Feasibility of Grid Compatible Microgrids
by ~~~MASSACHUET NTTTbyOF TECHNOLOGY
Patricia Janet Levi MAR 02 2017
B.A., Harvard University (2012) LIBRARIES
Submitted to the Institute of Data, Systems, and Society 4ARCHNES
in partial fulfillment of the requirements for the degree of
Masters of Science in Technology and Policy
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2016
@ Massachusetts Institute of Technology 2016. All rights reserved.
Signature redacted
Author ..............(j ) .....................
CJfY
Instuitute U Data, oy b tems, I U ouciety
May 6, 2016
Certifiedby.....Signature redacted..
Ignacio IPerez Arriaga
Visiting Professor, MIT Sloan School of Management
Professor, Comillas University (Spain)
Thesis Supervisor
Accepted by Signature redacted ..
Munther Daleh
William Coolidge Professor of Electrical Engineering
and Computer Science
Director, Institute for Data, Systems, and Society
Acting Director, Technology and Policy Program
2
Feasibility of Grid Compatible Microgrids
by
Patricia Janet Levi
Submitted to the Institute of Data, Systems, and Society
on May 6, 2016, in partial fulfillment of the
requirements for the degree of
Masters of Science in Technology and Policy
Abstract
There are 1.1 billion people in the world who lack access to electricity, mostly in rural
areas. The expansion of the central grid has been slow in many developing countries,
hampered by a lack of supply, poor finances, and politics. Distribution companies
in these countries are often cash strapped, in a tremendous amount of debt and are
unable to make adequate investments in infrastructure.
Off-grid technologies can be the most cost-effective choice in remote areas, and they
also can offer a solution for communities that will not receive reliable centralized
electricity for many years. These solutions include solar home systems and microgrids.
However, investment in microgrids has been discouraged by the risk of the central grid
expanding into the service area of a microgrid. An attractive solution is to create
technical standards for microgrids such that they are able to connect to the grid if or
when it arrives, and to provide regulations for the integration of these systems into
the operation of the main grid. This arrangement could reduce the risk to microgrid
investors significantly. While existing literature speculates on the value of such a
system, the costs and benefits have not been quantified.
This analysis uses the Reference Electrification Model, a tool developed in collabo-
ration by the Massachusetts Institute of Technology and IIT Comillas - Madrid, to
assess the costs and benefits that might arise when using grid compatible microgrids.
These results and an assessment of the regulatory context and forthcoming regula-
tions show that grid compatible microgrids can provide significant social value, but
only if supported by sufficient subsidies and a recognition of the costs imposed on
society by depriving so many people of electricity.
Thesis Supervisor: Ignacio Perez Arriaga
Title: Visiting Professor, MIT Sloan School of Management
Professor, Comillas University (Spain)
3
4
Acknowledgments
The help of many people was necessary for making this thesis come to life. Professor
Ignacio Perez-Arriaga was instrumental in supporting my work and allowing me to
wander off into various explorations of how to think about and use the results of REM.
He also gave me the opportunity to travel to many parts of the world, an experience
for which I am immensely grateful. Claudio Vergara and Reja Amayata provided
vital ground support. Reja was always a helpful sounding board for weird results and
tricky questions. Claudio was instrumental in wrangling REM into a useable system,
and helping troubleshoot difficulties. Also, his curiosity made sure our trips to India
always had some excitement.
Immense thanks are due to the team of students who I worked with along the way.
We shared some stressful times but also had plenty of excellent fun along the way.
Vivian's hilarious wit and perspective helped make the challenging moments of this
project easier to deal with. Yael provided an amazing amount of advice, support,
and friendship, without which I'm not sure I ever would have figured out how to
work effectively in our team. Doug exhibited remarkable patience in explaining so
many parts of REM to me. Our no-longer-so-newbie students, Cailinn, Turner, and
Stephen, integrated themselves seamlessly into this often crazy and messy project
and took on a series of amazingly helpful roles and projects. I'm sure that they'll do
even more amazing things in their remaining time at MIT.
Thanks are also due to our counterparts in IT Comillas, Madrid. They all made
our summer there a joy, and were an instrumental part of the team. Andres Gon-
zales Garcia made India a bit more warm and caring, and saved the day numerous
times with his endless supply of useful electronics accessories and Ricola. Fernando
de Cuadra, Pedro Cillier, and Carlos Mateo have been instrumental in developing
REM and fielding countless questions.
5
Just as important as my research colleagues are my wonderful support system of
family and friends, without whom I would have certainly gone insane before I fin-
ished TPP. Thanks to the whole TPP crew for making this wild MIT ride into a
ton of fun. You all kept me sane and reminded me to enjoy myself along the way.
My parents were always there to console me on the bad days and remind me what
actually matters. Finally, I could not -have survived this process as sanely as I did
without the support of my boyfriend, Pallav. Thank you for being an unwavering
source of support and strength. I've learned so much from your advice throughout
the last year, and I look forward to all that I have yet to learn from you.
6
Contents
Contents 7
1 Introduction 11
1.1 Overview - The Energy Access Challenge . . . . . . . . . . . . . . . . 11
1.1.1 Research questions . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1.2 The viability gap . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.1.3 Indian context of energy access . . . . . . . . . . . . . . . . . 18
1.1.3.1 Electrification regulation & programs in India . . . . 18
1.1.3.2 The trouble with India's distribution companies . . . 20
1.1.3.3 Barriers to investment in off-grid electrification . . . 27
1.2 Grid-compatible Microgrids . . . . . . . . . . . . . . . . . . . . . . . 29
1.2.1 When and why they are useful . . . . . . . . . . . . . . . . . . 29
1.2.2 What Makes a Microgrid Grid-Compatible? . . . . . . . . . . 32
1.2.2.1 Network requirements . . . . . . . . . . . . . . . . . 32
1.2.2.2 Generation interconnection . . . . . . . . . . . . . . 33
1.2.2.3 Islanded operation . . . . . . . . . . . . . . . . . . . 34
1.3 Preview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2 The Reference Electrification Model (REM) 37
2.1 O verview . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2 Precedents for REM . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Description of REM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.1 Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4 Scenario construction in REM . . . . . . . . . . . . . . . . . . . . . . 48
2.4.1 User's perspective . . . . . . . . . . . . . . . . . . . . . . . . . 48
7
2.4.2 Definition of Key Inputs . . . . . . . . . . . . . . . . . . . . . 49
2.4.2.1 Cost of Nonserved Energy (CNSE) . . . . . . . . . . 49
2.4.2.2 Discount Rate . . . . . . . . . . . . . . . . . . . . . 51
2.4.2.3 Annual Management Costs . . . . . . . . . . . . . . 55
2.4.2.4 Per Customer Costs . . . . . . . . . . . . . . . . . . 55
2.4.2.5 Network and component lifetimes . . . . . . . . . . . 56
3 Methods
3.1 Overview of Vaishali .................
3.2 Inputs for Case Study Area: Vaishali, India .
3.2.1 Cost of Nonserved Energy (CNSE) . . . . .
3.2.2 Discount Rate . . . . . . . . . . . . . . . . .
3.2.3 Taxes and Tariffs . . . . . . . . . . . . . . .
3.3 Scenarios run in REM . . . . . . . . . . . . . . . .
3.3.1 Business model and regulatory assumptions
3.3.2 Scenario 1: Grid-compatible microgrids . . .
3.3.3 Scenario 2: Non-grid-compatible microgrids
3.3.4 Scenario 3: All grid-extension . . . . . . . .
3.4 Financial analysis and assumptions . . . . . . . . .
4 Results
4.1 REM Sensitivities . . . . . . . . . . . . . . . . . . .
4.1.1 Results. . . . . . . . . . . . . . . . . . . . .
4.1.2 Other Model Result Observations . . . . . .
4.2 Scenario Results . . . . . . . . . . . . . . . . . . . .
4.3 Comparison . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Savings from a mixed on/off-grid approach
4.3.2 Value of grid-compatible microgrids . . . . .
4.3.3 Need for ongoing subsidy of grid-compatible microgrids
5 Regulatory Assessment
5.1 Regulatory requirements . . . . . . . . . . . . . .
5.2 Current status of regulation in India . . . . . . .
5.2.1 Status of off-grid regulation . . . . . . . .
5.2.2 Existing franchisee regulation . . . . . . .
101
. . . . . . . . . . . 101
. . . . . . . . . . . 103
. . . . . . . . . . . 103
. . . . . . . . . . . 105
8
57
. . . . . . . . . . 57
. . . . . . . . . . 59
. . . . . . . . . . 59
. . . . . . . . . . 61
. . . . . . . . . . 64
. . . . . . . . . . 65
. . . . . . . . . . 67
. . . . . . . . . . 68
. . . . . . . . . . 69
. . . . . . . . . . 70
. . . . . . . . . . 71
73
. . . . . . . . . . 73
. . . . . . . . . . 73
. . . . . . . . . . 86
. . . . . . . . . . 89
. . . . . . . . . . 94
. . . . . . . . . . 94
. ......... 95
98
5.2.3 Recent developments in off-grid policy
5.3 Lessons learned in other countries . . . . . . . .
5.3.1 Cambodia . . . . . . . . . . . . . . . . .
5.3.2 Tanzania . . . . . . . . . . . . . . . . .
5.3.3 M ali . . . . . . . . . . . . . . . . . . . .
5.3.4 Sri Lanka . . . . . . . . . . . . . . . . .
5.4 Assessment of India's Proposed Regulation . . .
5.4.1 Forum of Regulators Recommendations
5.4.2 Uttar Pradesh microgrid regulation . .
5.4.3 Conclusions . . . . . . . . . . . . . . . .
6 Discussion and Conclusions
6.1 Utility of REM . . . . . . . . . . . . . . .
6.2 Potential for Grid Compatible Microgrids .
6.3 Regulatory conclusions . . . . . . . . . . .
6.4 Future work . . . . . . . . . . . . . . . . .
6.5 Conclusions . . . . . . . . . . . . . . . . .
123
. . . . . . . . . . . . . . . 12 3
. . . . . . . . . . . . . . . 124
. . . . . . . . . . . . . . . 12 8
. . . . . . . . . . . . . . . 129
. . . . . . . . . . . . . . . 13 1
Bibliography
A Inputs for REM: Definition and Values Used
A.1 Discount rates . . . . . . . . . . . . . . . . . .
A.2 Cost of Nonserved Energy (CNSE) . . . . . .
A.3 Diesel cost . . . . . . . . . . . . . . . . . . . .
A.4 Network and component lifetimes . . . . . . .
A.5 Per customer costs . . . . . . . . . . . . . . .
A.6 Demand growth rate, years of demand growth,
A.7 Annual management costs . . . . . . . . . . .
A.8 Operations and maintenance (O&M) costs . .
A.9 Installation costs . . . . . . . . . . . . . . . .
A.10 Existing Network data . . . . . . . . . . . . .
A.11 Demand & Customer Type . . . . . . . . . . .
A.12 Network and Generation Catalog . . . . . . .
B Calculation of the Cost of Nonserved Energy
and k-factor . . . . . .
133
141
141
147
150
151
152
153
156
159
160
160
161
162
163
9
106
108
108
109
110
111
111
111
120
122
10
Chapter 1
Introduction
1.1 Overview - The Energy Access Challenge
There are 1.1 billion people in the world who lack access to electricity, mostly in
rural areas [551. Electricity access could dramatically improve their quality of life
and reduce poverty [46]. The expansion of the central grid has been slow in many
developing countries, hampered by a lack of supply, poor finances, and politics. Even
having a grid connection in a developing nation does not always mean access to
quality electricity. For example, villagers interviewed in Bihar, India, reported that
the average number of hours of service per day was 1.3 hours in bad months, and 6.3
hours in good ones [401.
The scale of investment needed to build the electricity infrastructure for these peo-
ple is phenomenal. At the same time, climate and other environmental objectives
necessitate that planners and policy-makers consider how to best provide electricity
service in light of the grave dangers posed by climate change. While greenhouse gas
emissions should not be used as an excuse to deny access to a basic level of electricity,
neither should electricity access be planned without regard to the climate implications
of growing access and consumption. The creation of a new electricity system in these
parts of the world presents a tremendous opportunity to design the best, cleanest sys-
11
tem possible, without the obstacles technological lock-in and stranded assets present
in developed-world electricity systems. Williams paints a rosy and optimistic view of
the opportunity that developing countries have in exploiting off-grid electrification,
both for accelerating electricity access and for developing a more modern, sustainable
grid;
"In the absence of legacy systems, many developing countries have the
opportunity to leverage decades of technological advancement and experi-
ence in developed countries as they build innovative modern infrastructure
systems. Microgrids, small electricity networks that have the ability to
operate autonomously, can play a key role in developing an electricity
infrastructure built around decentralized renewable energy technologies.
At the same time, microgrids can accelerate electricity access to areas the
central electricity grid cannot reach in the short to medium term. As these
microgrids develop, they can be easily interconnected, creating a decen-
tralized network that can aggregate loads and generation capacity, while
maintaining the ability to operate as isolated systems should the need
arise. Many developed nations are now devoting significant resources to
retrofit existing infrastructure and permit the integration of decentralized
technologies. Countries with underdeveloped electricity systems
are well positioned to leapfrog outdated centralized approaches."
[emphasis added] [60]
Putting this vision into action and balancing the two moral imperatives of electricity
access and GHG reduction in developing countries is no small feat. India and China,
for example, are building out their coal capacity just as fast as they can to meet
growing demand, and the developed countries have not been as forthcoming as hoped
in providing additional financing and other support for low-carbon power. However,
the attractiveness of renewable energy sources for many rural, unelectrified people
presents a glimmer of opportunity to meet both of these objectives at once. Such
technologies can be the most cost-effective choice in remote areas [301. Off-grid sys-
12
tems can offer a solution not only for extremely remote locations that will never see
the centralized grid, but also for communities that will not receive reliable centralized
electricity for many years.
Distribution companies in many developing countries are cash strapped. India pro-
vides a dramatic case study in the dysfunction of distribution companies; they are
in a tremendous amount of debt and are unable to make adequate investments in
infrastructure. These cost pressures mean that cost-efficiency is key for any rural
electrification program. Appropriate use of off-grid systems should be considered
as an important aspect of such a program, since these systems can be more cost-
effective than traditional centralized electrical infrastructure for many small, poor,
and remote communities. Even if they are not more cost-effective, they may still be
available sooner and be able to deliver electricity more consistently at more useful
times.
Microgrids and solar home systems could be installed in many communities years
before the central grid might be able to reach them. Their ability to leverage many
sources of capital could give them a huge advantage in nimbleness and scalability over
government-funded utilities which must negotiate bureaucracy, infrastructure, and
finances that are often arrayed against their progress. Additionally, in places where
the centralized grid is known to be highly unreliable, off-grid systems could more
reliably provide electricity for critical loads at peak hours. Finally, off-grid systems
built to be compatible with the centralized grid could retain their usefulness after
the grid reaches them; they could be integrated into the grid, ultimately providing
savings on distribution wiring costs for the utility, while providing electricity faster
to remote communities.
There are several obstacles limiting the potential of distributed, renewable energy
sources. First, regulators and state-controlled utilities may not know the magnitude of
the cost savings that could come from using a mix of on-grid and off-grid approaches,
so they may not even be aware of the opportunity. As a result, they may not have
put in place policies that could create a fertile environment for off-grid electricity
13
businesses to grow. Second, many of the countries with low levels of electrification
lack sufficient funds for a comprehensive electrification program; private investment is
almost certainly needed to provide a significant portion of the funds needed. Creating
a favorable investment environment is challenging in these countries, but there are
several routes by which regulators might do so.
Many nations have a two-pronged approach to electrification, both through promoting
the extension of the centralized grid and through supporting distributed electricity
sources like solar home systems and microgrids. However, for such a scheme to work,
regulators must craft a clear plan for what happens when the two approaches meet
1541. Without such a plan, investors in a local microgrid run the risk of losing their
entire business to the centralized grid, which is typically able to provide much lower
tariffs to customers thanks to significant government subsidies. Recent studies have
shown that this risk is a key obstacle hindering growth in the microgrid sector in India
[141. There are other downsides to the grid taking over in addition to putting local
microgrids out of business- Any renewable generation that might have been supplying
customers of that microgrid will probably be replaced by electricity generated by
coal or other fossil fuels that dominate the centralized grid. If the centralized grid is
unreliable, as it is in many parts of the rural developing world, customers additionally
may lose reliability (especially during peak hours) by switching over to the centralized
grid. This is clearly an undesirable scenario.
A potentially attractive solution is to create technical standards for decentralized
electricity systems such that they are able to connect to the grid if or when it arrives
(hereafter referred to as 'grid-compatible microgrids'), and to provide regulations for
the integration of these systems into the operation of the main grid. This arrangement
could assure investors in the original system of a continued return on their investment
in the event of the arrival of the grid, or at least an attractive exit strategy.
However, such an arrangement would impose higher initial costs on microgrid projects,
and brings up a host of regulatory and financial questions. How should tariffs be set
before and after the microgrid is connected to the grid to ensure equitable payment by
14
customers and cost-recovery for investors? Will the reduction in uncertainty create
a sufficiently attractive investment environment, despite higher upfront costs? What
level of subsidy will be needed to maintain this approach, and is such a subsidy
feasible? These are the types of questions addressed in this thesis. This analysis will
focus on the specific context of India, but my conclusions should be broadly applicable
in developing countries.
The official policies of India have focused on the expansion of the central grid as the
main means for providing electricity. Although they have allowed the development of
off-grid systems, and even promoted them for extremely remote areas, there is still
significant room for Indian policies - and, indeed, for the policies of many developing
countries - to exploit the benefits of off-grid electrification.
1.1.1 Research questions
Putting numbers to the benefits that grid-compatible microgrids could provide would
help direct policy discussions towards more effective electrification solutions. With
so much money and so many livelihoods on the line, it is important to base pol-
icy decisions on the best available data and analysis. The Reference Electrification
Model (REM), a rural electrification planning tool developed at MIT over the last
three years, gives us the tools to address the questions surrounding grid-compatible
microgrids quantitatively, and to frame policy discussions around a firm understand-
ing of the magnitude of the tradeoffs. In light of all this, this analysis examines the
following questions:
* What is the potential for grid-compatible microgrids to enable societal cost-
savings? Under what conditions would the higher up-front costs be outweighed
by the benefits?
* What type of regulatory factors are necessary to make these projects feasible
and financially appealing?
We focus on the Indian context but my results can be useful in a variety of contexts.
15
In order to answer this question we employ REM, a planning tool that is completing
its initial development phase, as well as financial analysis, and available literature and
stakeholder interviews. We will use the district of Vaishali, within the Indian state of
Bihar, as a case study.
1.1.2 The viability gap
Any discussion of rural electrification is incomplete without acknowledging the so-
called 'Viability Gap.' This concept refers to the gap between the cost of building
the infrastructure to supply electricity to rural areas, and the revenue collected from
rural customers. For rural systems, the upfront infrastructure to reach them is more
expensive, and the cost of operating and maintaining the infrastructure and collecting
tariffs will be higher due to the remoteness of the area. Electrical losses may also be
greater, leading to an additional financial penalty. The issue is not just that providing
electricity to rural consumers is more expensive than supplying urban consumers;
rural consumers also often have lower demand and a lower ability to pay for electricity
services, especially in developing countries. The combination of these factors means
that rural consumers often may not be able to pay for the cost of their electricity [8J.
The gap between what they can pay and the actual cost gives rise to the Viability
Gap.
The idea of the viability gap is not a new one. The cross-subsidization of rural
consumers by their urban counterparts is common throughout the world and through
the history of electric grids, both in developed and developing countries. In the 1930s,
electricity was uncommon in rural parts of the United States; it was only through
the Rural Electrification Administration (REA), established by President Franklin D.
Roosevelt as part of the New Deal, that electricity services were extended to rural
America. The REA provided grants, loans, and loan guarantees to rural electric
cooperatives, and it still exists today, providing loans and loan guarantees to cover
the viability gap1 .
'More info available at https://en.wikipedia.org/wiki/Rural_ElectrificationAct and
16
Today it is common for utilities to charge the same tariffs to their urban consumers
as their rural consumers, although the latter are more costly to provide for; this
arrangement allows the cross-subsidization of rural consumers by their urban coun-
terparts, and helps utilities remain solvent while extending their service to everyone.
Many developing countries face the issue that there are not enough urban consumers
to subsidize all the rural consumers. With a smaller urban to rural ratio, utilities
have few options for providing universal access while remaining solvent. Government
subsidy is vital. A key lesson from rural electrification programs in Brazil, China,
and southern Africa is that "in most successful programs, a substantial proportion
of the investment has been obtained at reduced or low interest rates or in the form
of subsidies/grants," [541. Even reduced-rate loans may not be sufficient to meet the
viability gap; grants are usually necessary.
These same conclusions hold true for off-grid investments; there is likely to be a gap
between the cost of service and what consumers can pay, and that gap will need to
be filled by the government, or a philanthropic organization. The inability of many
off-grid systems to turn a profit in the absence of a subsidy, then, should not be
seen as grounds for dismissing them as a solution. All we can strive to do is identify
ways to narrow that gap, and provide subsidies to bridge it while still encouraging
efficient business practices. Experience has shown that even poor consumers are able
to pay for a low level of service, such as the few hours of evening electricity for a
light bulb and mobile charger provided by the minimalist microgrids of Mera Gao [3].
However, above these basic levels, ability to pay may drop off sharply. Encouraging
private investment can open access to greater quantities of capital, encourage greater
experimentation with business models, and provide better incentives for efficiencies.
However, expecting it to eliminate the need for subsidies may ultimately be too high
a bar in many parts of the world.
http://blogs.usda.gov/2015/05/11/rural-electrification-celebrates-80-years-of-rural-productivity/
and http://www.rd.usda.gov/about-rd/agencies/rural-utilities-service. Accessed April 29, 2016.
17
1.1.3 Indian context of energy access
Although the Indian government has made repeated pledges and funded numerous
programs aimed to bring electricity to all people living in India, 29% to 37% of Indians
have no electricity access, or else have access to electricity that is extremely unreliable
[32]. The lack of adequate electricity supply in India has been widely reported, and
its effects felt in large blackouts as well as the persistence of regular outages. The lack
of sufficient generation is compounded by the problem of degraded and insufficient
transmission and distribution infrastructure to bring electricity to users. Even when
there is sufficient generation capacity, the distribution companies may be unwilling to
purchase electricity when wholesale prices are high, since tariffs are not high enough
to cover the costs of supply. India's government is aware of the issues and has made
significant efforts to improve electricity supply through reforms like the Electricity
Act of 2003 and many other policy programs for rural electrification [281.
1.1.3.1 Electrification regulation & programs in India
Electricity Act of 2003: In 2003, the Indian government passed the Electricity
Act of 2003, which remains the landmark piece of legislation governing the electricity
sector, and led to significant reforms. It led to the break-up of the State Electricity
Boards, which had been large vertically-integrated utilities that owned everything
from generation to local distribution. The State Electricity Boards were broken up
into separate generation, transmission, and distribution companies. The transmission
businesses and distribution businesses were allowed to exist as regulated utilities, and
regulatory bodies were created to oversee them. Generation was delicensed, allowing
private generation operators to sell to any licensed distributors or, in some cases,
directly to customers [9]. However, open access, though required by the Act, is being
implemented only haltingly by regulators [28]. The Act does not distinguish between
distribution and retail activities, although it does allow regulators to introduce some
measure of retail competition at their discretion, though this has not come to pass [9].
18
However, if regulators did so, it would likely be in name only, since retailers be unable
to compete with the heavily subsidized service provided by state-owned distribution
companies.
The Act containes several parts that were very relevant for rural electrification. It
encoded the 'universal service obligation', which mandates that the states, through
the distribution companies, are responsible for providing access to electricity to all
villages and hamlets. In many states this obligation is far from being met, and,
as discussed below, distribution companies are far from being able to meet it in the
near-term. Acknowledging that this mandate is outside the ability of local distribution
companies, regulators have often turned a blind eye to this requirement.
Perhaps the most important impact of the Act was that it allowed the generation
and distribution of electricity in rural areas (e.g. through a microgrid) without a
license. This law lowered the barrier for countless microgrid companies to enter the
market where the distribution companies were not able to. reach. There is a legitimate
debate surrounding the sagacity of such a lack of regulation for microgrid electricity
service providers, but it does seem to have jump-started the market. These issues are
discussed further in Section 5.
RGGVY and DDUGJY: The Rajiv Gandhi Grameen Vidyutikaran Yojana (RGGVY)
program, which has been since subsumed within Deendayal Upadhyaya Gram Jyoti
Yojana (DDUGJY) program, facilitates India's goal of providing universal access to
electricity for rural households. RGGVY was launched in 2005, with the ambitious
goal of providing universal access by 2009. In 2013, it was subsumed under DDUGJY.
The program aims to provide an electricity connection to all Indian villages with 100
people, with an emphasis on grid extensions, and provides for the strengthening of the
distribution system. Electricity connection is provided for free to households under
the poverty line under these programs. Where grid extension is deemed infeasible or
not cost-effective, there is some support for off-grid systems. However, the focus of
the program has been primarily on grid extension. The costs of rural electrification
19
are funded by a 90% subsidy from the Ministry of Power, or 100% subsidy for the con-
nection of households below the poverty line. Preferential loans are available through
the Rural Electrification Corporation (another Indian governmental organization) for
the remaining 10% of the cost. [39], [31 and [30] provide a detailed description of the
program and its effectiveness. In theory, states must meet requirements including
reliability of supply, and provision of supply for at least 6 to 8 hours per day, or else
the grants will be converted into interest-bearing loans. In practice, this punishment
has not been enforced [301.
DDUGJY added two additional goals to those of RGGVY, meant to address supply
in agricultural regions. It provides support for the separation of electricity feeder lines
in agricultural areas, so that agricultural uses will be fed by one line, and household
and commercial uses fed by another line. It also provides for the strengthening of
the sub-transmission and distribution network in rural areas, notably including the
metering of transformers, feeders, and customers [36]. The significance of this move
is discussed below.
RVE: The Remote Village Electrification Program (RVE) picks up where RGGVY
and DDUGJY leave off. It provides support for hamlets with less than 100 people,
and for areas where grid-extension has been deemed infeasible or not cost-effective.
It provides a 90% capital subsidy from the Ministry of New and Renewable Energy.
Under the program, electricity supply can be provided from the most economical
local resources, including conventional and non-conventional generation sources. In
practice, many households are provided with solar PV home lighting systems, which
are often seen as a second rate service, making this program politically unpopular
[391.
1.1.3.2 The trouble with India's distribution companies
The poor state of distribution companies (discoms) in India is a major cause of the
poor quality of electricity supply and the widespread lack of access. Discoms are
20
deep in debt across India. As of March 2015 they had cumulative losses of 3.8 trillion
Rs, or $USD 57 billion. Cumulative outstanding debt was 4.3 trillion Rs, or $USD
70.6 billion [45]. To put that in perspective, the amount owed to banks by India's
distribution companies is 2.5 times India's defense budget, and larger than its national
fiscal deficit [5]. The shortfall between tariffs and costs means that ongoing operating
costs are covered by even more debt. Discom losses in 2013 were 3% of India's GDP
[551. Although the discoms receive substantial subsidies from the government, these
subsidies are far from covering all losses, and even the subsidies that are promised
are not always delivered [30, 28, 51].
All this debt reverberates across the electricity sector. Banks are reluctant to lend
additional money to these over-leveraged institutions, so distribution companies often
have difficulty coming up with money to cover investments and even operational costs.
As a result, infrastructure is often poorly maintained, leading to outages, and due
to low tariffs discoms buy less power than their consumers demand when wholesale
prices are expensive, necessitating power cuts. Their lack of credit-worthiness makes
them a suspect counterparty for contracts with generation companies, which has a
chilling effect on investment in additional generation [301.
Reasons for poor finances of discoms These large and growing losses have
their root in the political capture of electricity regulators, tariff schedules that do
not allow discoms to cover their costs, and large subsidies that states do not always
refund. Solving these issues will be key to providing better electricity access across
India. Although this situation is evolving and some hopeful developments are on the
horizon, it remains an important and sobering backdrop to the discussion of off-grid
electricity in India, and understanding it will put the discussion of the regulation of
microgrids in a helpful context.
Low tariffs are the major reason for the poor financial situation of discoms [30]. High
levels of theft and uncollected tariffs (collectively referred to as 'non-technical losses')
as well as actual losses also contribute to the substantial gap between the cost of
21
supply and collected revenue for distribution companies. Social and political pressure
to keep tariffs low has meant that tariffs are rarely revised upwards, despite man-
dates from the Electricity Act of 2003 for annual revisions of the tariff. Tariffs are
now significantly below the cost of providing service. Between 2007 and 2011, the
average gap between cost of supply per kWh and the tariff ranged from 1-1.4 Rs, or
1.5-2 cents in USD [301. For the North Bihar Power Distribution Company, fiscal
year 2015-2016 saw collected revenues that were only 70% of the calculated revenue
requirement2 , a figure that is especially astonishing given that the revenue require-
ments are often calculated using overly optimistic estimates of the discom's ability to
improve their efficiency and reduce nontechnical losses. Distribution companies are
increasingly dependent on subsidies to cover their operational costs; in Bihar, such
subsidies comprised 33% of the state budget in 2011, an huge fraction in a state that
scarcely has the resources to afford such subsidies [301.
There are indeed many poor customers who should rightly have access to electricity
below the cost of providing it, as a humanitarian right. However, such a program
should be ideally limited in its scope to those who truly need it. Instead, the gov-
ernment provides limited subsidies, and the distribution companies often attempt to
recoup some of this extra cost by charging higher tariffs to commercial and indus-
trial customers. This cross-subsidization creates a vicious cycle wherein commercial
and industrial consumers are charged a rate above the cost of service, encouraging
them to set up their own private generation facilities which may be more reliable and
more cost effective. As more companies do this, the revenue received by the discom
drops, forcing them to raise commercial and industrial tariffs further to support the
cross-subsidization scheme [281.
Tariff reform is necessary for improving the electricity sector in India. Many in the
sector are aware of this fact, yet tariff reforms have not happened. The reason for
this lack of tariff reform can be attributed to the political capture of utilities and,
2Table 6.66 of the Tariff Order for the North Bihar Power Distribution Company. Found
at http://berc.co.in/media/Tariff-Order/Taxiff%200rder%20NBPDCL%20FY%202015-16.pdf Ac-
cessed April 6, 2016
22
importantly, the state regulators of electric utilities. The utilities owned by the state
government, and many appointments within them may be through political crony-
ism (which also contributes to an inefficient hiring). Electricity is a major political
bargaining chip in India, and the close relationship between state government, distri-
bution companies, and electricity regulators allows politicians to manipulate its price
and supply. A 2015 study showed convincingly that state governments frequently
manipulate the supply of electricity, especially in strongly contested areas [7]. In
November 2015, in the lead up to state elections in the state of Bihar, the state's
Chief Minister directed the distribution companies to provide electricity connection
to all customers for free (though they would still have to pay for the electricity once
connected) [24]. Reduced or free electricity is a common promise in Indian election
cycles.
The Electricity Act of 2003 created independent regulators to oversee the electricity
sector, including the discoms; part of their job is to ensure that the discoms do not
abuse their monopoly power, that they operate efficiently, and that they comply with
governmental mandates. The current status of the distribution companies indicates
that they have not been able to do so. The regulators are often 'captured' by the state
government, and tend to comply with the political will therein. Often, the regulator
is appointed not based on competence but rather based on political ties. As well,
they are dependent on grants from the state government to fund their activities, so
a government which disliked the regulator's activities could retaliate by defunding
them [28, 81. Even if a regulator did have the desire to enact reforms or to punish
a discom for missing agreed-upon milestones, they would have few tools available
to them. Financial incentives and penalties are the main tool of the regulator, but
such instruments do not have a great effect since the state can ultimately absorb any
fines, for example penalties levied for not meeting the universal service obligation or
the renewable portfolio obligation. The lack of independence and useful enforcement
tools makes it difficult for the state regulators to enforce sensible tariff adjustments,
especially when such adjustments are a politically hot issue.
23
6UAd4LU -, ,, , ,,, - ,ANW19h-- -- I I - I.., - -1 - . g4"" - I I Ad",
The political capture of the utility and regulator has consequences beyond tariff re-
form. Since efficiency can be less important than politics, utilities may maintain
an unnecessary number of employees, some of whom may be there through political
favors. The discom is incentivized to purchase power and other materials not from
the cheapest bidder, but rather from a politically favored company. They are also
unlikely to be open to key electricity sector reforms like open access, which would
allow private parties to buy and sell generation over the distribution grid in private
contracts [281.
Reasons for Poor Supply The factors outlined above and a few other issues as
well have combined to provide the discoms with an array of reasons and even incen-
tives to provide unreliable electricity service. The most important of these reasons
is that tariffs are below the cost of service, so that the more electricity the discoms
sell, the more money they lose. Discoms would rather purchase less electricity when
wholesale market prices are very expensive and cut supply to their own customers,
because tariffs are not high enough to cover electricity purchase costs [301. The lack
of adequate installed generation capacity in India has received significant attention
overseas, yet the issue of willingness to purchase may be considerably more impor-
tant. Arguably, if discoms were able to reliably purchase electricity, there would be
significantly more investment in generation.
The lack of supply is compounded by an aging infrastructure. Without sufficient
funds to maintain and upgrade their infrastructure properly, distribution companies
are often forced to cut supply even when electricity is available. These supply cuts
could be caused by the failure of a component like a transformer. Often, as well,
they are the result of upstream medium voltage distribution lines being too small to
meet the full demand of all the attached feeder lines at peak hours, so some of the
feeder lines must be cut to prevent over-capacity on the medium voltage line. Our
conversations with the operators of several grid substations during visits in 2015 and
2016 to Bihar indicated that this was a more common cause of blackouts than the
24
lack of generation.
When supply cuts must be made, rural agricultural areas are disproportionately likely
to have their service cut, further contributing to the miserable quality of service in
rural areas. Electricity supply to agricultural areas is often subsidized to extremely
cheap or even free levels. In 2011-2012 the agricultural sector consumed around 22%
of the total power sold in India, yet it accounted for only 8% of sales revenue for
power companies [511. Agricultural subsidizes were 64% of total subsidies provided
by states for power sales in 2011-2012; to make matters even worse, states often do
not fully cover those promised subsidies, leaving discoms even further in the red [51].
Since supply to agricultural areas is so unprofitable, the utilities are incentivized to
cut power to these areas first. Perversely, since the amount of power in these feeders is
often unmetered, discoms are incentivized to hide higher-than-expected non-technical
losses within the reported agricultural sales, in order to appear more efficient. Because
poor, rural consumers are typically on the same feeder lines as agricultural customers,
they end up with much less reliable power supply. One could ask if the free power for
agriculture is worth the abysmal reliability to these consumers.
These worries could make off-grid electrification seem even more attractive by com-
parison. Even if grid extension were the most cost-effective way to provide electricity
to a community in theory, in practice implementing this solution could be very dif-
ficult. If the infrastructure were built, it could still provide only very unreliable
service, rendering the few hours that a microgrid or solar home system might provide
attractive by comparison, since the supply could be more reliable.
Recent Reforms The issues with agricultural supply are why the DDUGJY pro-
gram has included as part of its objectives (1) the metering of supply on rural feeders
and (2) the separation of agricultural and household loads onto different feeder lines.
With these measures in place, rural households might suffer less collateral damage
from supply cuts, and discoms would be forced to be honest about their agricultural
sales. It remains to be seen how quickly this program has effect.
25
"I ~~ I - - '"AhUdahm- 1., .111, "_' -11 .. ' 1- 11-1-- 6
In the last year, India announced the UDAY (Ujwal DISCOM Assurance Yojana)
Scheme, an attempt to overcome the disastrous financial situation of the country's
distribution companies. It is an optional program, but incentivizes states to join it by
providing more funding for several electricity-related initiatives, including DDUGJY
and several other schemes. UDAY has four major parts [45]:
1. Improving operational efficiency. This is to be accomplished through smart
metering, upgraded transformers, energy efficient appliances and agricultural
pumps, among other things.
2. Reduction in the cost of power, primarily through making greater use of cheap
domestic coal, and other measures to make coal cheaper.
3. Removing a substantial amount of debt from distribution companies, and trans-
ferring it to states. By 2017, states are to take on 75% of existing debt from
distribution companies, which will allow the interest rates to be lowered from
14-15% to 8-9%. Debt remaining with the distribution companies will be con-
verted into lower-interest loans or bonds which must be at the bank's base rate
plus 0.1%.
4. Aligning the financial incentives of distribution companies and states. States
will be required to take on 50% of discom's losses at the end of every year.
Past efforts to bail out the distribution companies have failed, mostly due to political
reasons, but there is some reason to hope that this scheme is more comprehensive and
will be more effective than previous attempts [5]. The key political roadblock will be
getting the states on board with this plan, as they will be apprehensive to take on so
much of the distribution company's debt. However, if the incentives of extra funding
- and the threat of losing funding for federal electricity programs - is enough to drive
compliance by the distribution companies, this policy could have a major impact on
the state of India's power sector [5J.
Forcing the state to take on the debt of the distribution companies could be a very
effective way of aligning incentives between the distribution company, the state regu-
lator, and the state government. As the government is often unwilling to raise tariffs
26
for political reasons, and able to exert that will through the capture of the regula-
tor, tariffs stay low and losses stay high. The idea is that states will now have an
incentive to reduce those losses, potentially creating political will for significant tariff
reform. The reduction in interest rates for distribution company debt will also have
a noticeable impact; the requirement that debt be converted into loans or bonds with
an interest rate that is the base rate plus 0.1% represents a a significant improvement
over the current rates of 14-15%, given that base rates in India are currently around
9.3%. Stronger distribution companies with less debt will, in theory, be able to invest
in better infrastructure, purchase more power, and thus encourage positive develop-
ments in the Indian power sector as a whole. Time will tell how effective this scheme
will be.
1.1.3.3 Barriers to investment in off-grid electrification
1141 presents a very compelling analysis showing that the risk of central grid expan-
sion is the 'gateway barrier' preventing investment in microgrids. Through a series of
interviews with many people across the off-grid electricity sector, they show the per-
vasive worry that the grid will arrive and bankrupt a local microgrid business, since
they cannot compete with the low tariffs provided by the distribution company. One
of their interviewees notes that "The biggest risk for mini-grid[s] ... is the availability of
the grid.. .And the reason is of all of the 300 million people don't have access, about
90-95% of the villages covering the same population lives within 5 kilometers from
the existing transmission network. So if the utility decides to provide access, then
it can actually provide it the next day." Most distribution companies do not have
long-term plans for how the expansion of the grid will progress, providing potential
project developers and investors with no certainty regarding the magnitude of this
risk [14]. Indeed, since the discom is so closely tied to the state government, where to
extend the grid next is a decision often made on short-term political considerations.
This issue could be addressed by publicizing (and following) expansion plans, and
through the creation of a well-defined process for the off-grid business to receive some
27
I , -. 1. 1 1 11 . - -.- -, . . . - - .. - -- , 2.,.- -,- _Jj
value from the arrival of the central grid provided that they are grid-compatible, e.g.
by being bought out by the discom, or remaining as a franchisee in some capacity.
While the former option sounds easier in theory, the close political ties between the
discom and the state government, and the short tenor of many governmental admin-
istrators (many rotate after 2 years), mean that it could be much easier said than
done. Meanwhile, recent amendments to the Electricity Act have started to address
the issue of integrating grid-compatible microgrids. This development is discussed in
section 5.
Although this risk of grid expansion is perhaps the most preventative barrier, there
are still many hurdles to creating a viable microgrid business. Below are several of
the most prominent.
" Slim profit margins: Profit margins for microgrid businesses remain slim,
given the low ability to pay of many rural customers. Microgrid businesses
must achieve scale to realize greater efficiencies and more attractive revenue,
but progress in this direction has been sporadic, except for some ideal cases
such as those involving anchor loads, or physically isolated communities [14].
* Untrustworthy state subsidy: Although the government does provide a
subsidy for solar projects and for decentralized distributed generation in rural
areas, it does not always provide it reliably [30, 281. The uncertainty and
paperwork involved in obtaining the subsidy is burdensome for entrepreneurs;
several microgrid companies we spoke with in India indicated that they prefer
to operate without the subsidy as a result'.
" High transaction costs: Community engagement is key to the success of a
microgrid, and doing it properly requires a significant investment of time and
resources [3]. If it is not done properly, it can contribute to political retaliation
and theft of electricity and system components. This requirement makes it
difficult for an outside company with no local knowledge to come and establish
a business quickly.
3e.g. Meeting with Boond Power, July 2015
28
* Low availability of finance: Off-grid entrepreneurs have had difficulty ob-
taining finance from banks and investors, especially at reasonable interest rates
and tenor [3]. This situation is especially true for microgrid businesses, in part
because of the risks imposed by grid expansion; smaller-scale solutions such as
solar lantern businesses have slightly more access to capital [14].
1.2 Grid-compatible Microgrids
The concept of grid-compatible microgrids refers to the idea that a stand-alone micro-
grid could be built to certain specifications that allow it to interface with the central
grid if and when the occasion to do so arises. In theory, many types of microgrids
could do this. From a technical standpoint, the ability to synchronize local generation
with the main grid, if operating in AC, is the biggest hurdle. Even DC microgrids
could technically be connected to the central grid, if they had an inverter/rectifier
at the interface. However, in practice, due to political and regulatory constraints, a
grid-compatible microgrid must meet minimum standards for safety and quality in
the construction of its network. It is unlikely to be in DC due to the cost of an in-
verter/rectifier and the desirability for consumers of being able to purchase the same
appliances as everyone else. In the rest of this section I will discuss the situations in
which such microgrids might be useful, and the standards that they must meet.
1.2.1 When and why they are useful
There are three main situations for using microgrids4 to provide electricity access in
rural, developing areas; when customers are so remote and have such low-demand that
grid-extension will be economically unfavorable, where the existing grid is woefully
unreliable, and where the existing grid may not reach a set of villages for many years.
While the central grid may never reach customers in the first situation, negating the
4While some studies distinguish between minigrids and microgrids, I will refer to any collection
of households connected to the same isolated network with its own generation as a microgrid.
29
-
-- ---
utility of making it 'grid-compatible,' for the latter two situations, a grid-compatible
microgrid could provide several benefits. It could provide a more attractive invest-
ment framework for private developers, result in better quality service in the future,
and contribute to the sustainability of the electricity sector by encouraging distributed
renewable energy [60]. Unlike the solar home system kits typically sold in the devel-
oping world, a microgrid can be easily connected to the main grid when it arrives,
and can operate as an island in case of central grid blackouts [47]. (While isolated
solar panels can theoretically be attached to the main grid, it is less likely that this
arrangement would be feasible and attractive for poor rural customers, given the high
transaction costs and upfront equipment costs needed to sell electricity to the grid.
The panels could be retained for self-supply for a few appliances, however.)
For some remote, low-demand communities, a microgrid could be much more cost-
effective for supplying their electricity needs than the extension of the central grid
[40, 4, 25, 60, 291. Several studies have tried to estimate a radius outside of which
grid extension is no longer cost effective; although this approach is quite crude, it gets
across the point that under some combinations of factors, microgrids will make more
sense. The cost of building a grid line to reach a remote community may never be
recouped by the utility if demand is not high enough or tariffs are too low, because
the utility will never collect enough tariffs to come close to offsetting the investment
cost. Even for non-remote communities, if the grid is unreliable or there is no certain
date for grid expansion in sight, a microgrid could be cost-effective at addressing their
needs and ensuring that they do not have to wait years to receive reliable electricity
supply.
However, microgrids in any of these three scenarios have not yet achieved scale -
most remain pilot projects funded by philanthropic organizations or subsidized by
international development banks as investors endeavor to find a sustainable business
model and policymakers try to create an attractive environment [311. Attracting
investment to microgrid projects remains an issue, as it is difficult to cover project
costs with the limited payment capacity of poor rural villagers, and the risks of such
30
projects remain high [31]. Microgrid companies have had difficulty creating scalable
business models that turn a profit; even those that have figured out how to develop
sufficient demand and affordable supply often "struggle to find a replicable business
model that allows them to leverage the economies of scale that are critical for growth,"
[6].
Figuring out a favorable policy and business environment for such projects remains a
key objective. Doing so would unlock vast quantities of capital for rural electrification
[60]. The limited budgets of developing countries will not be enough to provide
electricity to the many who lack it in the near future, so this private capital would
be truly helpful. Utilizing private-sector investment in microgrids can also result
in better service than often-beleaguered state-owned utilities are able to provide;
these state-owned utilities are known to suffer from "inefficiency and poor technical
performance" [60, 221.
One of the key risks of investing in a microgrid is the risk of the central grid extending
to the location of a microgrid; [49] reports that they observed this in multiple case
studies [49]. Grid-compatible microgrids, and an accompanying regulatory frame-
work, can help address this uncertainty and reduce the risk to investors [49, 16].
There is a significant need to regulation for what happens when the grid is extended
to meet microgrids and what is required for the interconnection [4, 54]. Despite the
substantial literature on the need for microgrids and the advantages of grid-compatible
microgrids, there is no literature addressing the cost savings that could result from
smart grid-compatible microgrid regulation when electrifying a region. Such a study
would need to take into account not only the savings in upfront and ongoing costs,
but also the need to provide subsidies to private microgrid operators. These subsidies
are needed so that owners can cover their costs and customers can pay a uniform
tariff (the uniform tariff could be for central-grid-connected customers only, or for all
customers of microgrids and the central grid alike; this issue will be discussed'later).
Even accounting for this subsidy, a regional electrification scheme incorporating grid-
compatible microgrids could still be cheaper for the government and state utility than
31
grid extension. The newfound analysis capabilities that REM provides will allow us
to address the question of cost.
1.2.2 What Makes a Microgrid Grid-Compatible?
When the central grid reaches a microgrid, several requirements must be met if the
existing network and generation will continue to be used connected to the central grid.
This situation raises three sets of issues for compatibility. First, if the network is to be
maintained, it must meet certain standards for quality of components, voltage limits,
quality of service and safety and for the type of connection needed. Second, if the
generation is retained, it must be able to inject power to the grid at utility standards,
and to disconnect and reconnect safely from the central grid when it blacks out and
comes back online [211. Third, if the microgrid developer wishes to be able to operate
in 'islanded' mode while the central grid is down (a functionality that could be quite
useful in countries with unreliable grids), they must meet the above requirements
while also being able to shed load as needed while disconnected from the 'big' grid.
1.2.2.1 Network requirements
The microgrid network needs to be of sufficient quality to connect to the utility
network. Safety requirements must be met, which may require giving the lines a
certain berth, having a minimum pole height and line height, and having the necessary
electrical protections like circuit breakers in the event of short circuits.
In theory, there is no reason that a DC microgrid could not interconnect with an AC
utility grid; an inverter at the interface could connect them. However, in practice it
is highly unlikely that a microgrid not originally built as AC would be attached to
an AC centralized grid. DC microgrids are typically built in the developing world to
reduce cost; such microgrids may not be built to sufficient safety standards for the
central utility, and for communities where the cost constraint is so significant, the
cost of the inverter may be prohibitive.
32
Additionally, many of the appliances consumers might wish to use once they have
access to a larger amount of power require AC supply; consumers might be unhappy
that they are constrained to a limited (though admittedly growing) selection of appli-
ances designed especially for DC power supply. For these reasons, it is a reasonable
requirement that microgrids built to be grid compatible should be AC. Beyond this
requirement, it is possible to have compatible microgrids with a variety of configu-
rations: three phase, single phase, single phase earth return, or others. Single phase
earth return has been noted for its low cost, though it can result in large voltage
drops.
1.2.2.2 Generation interconnection
For the generator to remain useful after the microgrid has connected to the main grid,
it must be able to connect at the right frequency and phase, and produce power of
sufficient quality for the utility's standards. These specifications will impose different
requirements for synchronous generators, asynchronous (induction) generators, and
inverters (e.g. solar panels) [21].
If an isolated microgrid has a diesel generator, it is likely to be a synchronous genera-
tor, since it does not need an external power source to start, unlike an asynchronous
generator. These types of generators are the most difficult to connect to the grid, as
their phase and frequency must be matched to the main grid's before interconnection.
This matching can be performed by electrical equipment or by a trained technician
[21].
However, it is unlikely that a developer would find it economical to run their diesel
generator for power when significantly cheaper power can be purchased from the grid.
What is more likely is that existing renewable generation will be used, and any diesel
generation will be used only as a backup if the central grid is down. Asynchronous
(induction) generators are common in micro hydro installations, as well as in wind
turbines. Their interconnection requirements are not as complex as those for syn-
33
chronous generators, as they do not need to be synchronized before interconnection
[211.
Solar panels on an AC grid require an inverter to convert the DC output of the panel.
These inverters may contain all of the protective relays needed to connect to the grid.
However, it is important to note that many inverters are made either for islanded
operation or grid-tied operation, and fewer are able to do both. Many inverters used
for off-grid applications are not able to export power to the main grid [211. Investors
expecting their microgrid to be able to interconnect to the grid one day must ensure
that the project uses the correct type of inverter to do so.
1.2.2.3 Islanded operation
One of the potential advantages of a grid compatible microgrid is that it could continue
to provide some service to customers while the central grid is blacked-out, a frequent
occurrence in India. There are several technical requirements with this functionality;
resynchronization (discussed above), ensuring worker safety during islanding, and
managing undersupply during local operation.
During a blackout, utilities worry about distributed generation causing some lines to
remain 'live' without warning, which can present a hazard to workers who may be
working on the lines while trying to restore power [211. Actions must be undertaken
to ensure that worker safety is protected if the microgrid is live while the central grid
is down.
If a grid-compatible microgrid is to retain the ability to operate in an islanded mode
after being connected to the central grid, there are several additional requirements
it must meet. First of all, the microgrid must have the proper electronics in place
allowing the microgrid to disconnect from the central grid quickly and safely when
central power is lost or a major disturbance is detected. It also must be able to
determine, either through power electronics or human assessment, when it is safe to
reconnect to the central grid, and it must be able to reconnect at the correct frequency
34
and phase [211.
Consumers connected to the microgrid will be able to draw more power once the
microgrid is connected to the central grid. They may purchase additional appliances
and equipment that has a higher power draw because the grid can now support it.
As a result, after interconnection power draw from the microgrid consumers may
grow to a level that cannot be supplied by the local generation and storage of the
microgrid. In order to take advantage of the 'backup' capabilities of the microgrid, the
developer/operator would need to devise a method of ensuring that the power supply
of the microgrid is not exceeded by demand during periods of isolated operation. This
could be accomplished in several ways:
" Having two separate sets of electrical lines for the centralized network and the
microgrid. While this would create significant redundancy, it would give each
household a second set of plugs that could provide them with a limited amount
of power in case the central grid was out. Controls, such as load limiters,
could ensure that household draw on the backup plug was not too high. This
method would be simple for consumers to understand but would involve signif-
icant redundant expenditure, and would raise other questions, too. Would the
customers use the 'backup' outlets when both services were available? Would
the same tariff be charged for each?
" Variable load limiters that can be controlled remotely could be used to constrict
household power draw during a central grid blackout. If a blackout occurred
and a household was drawing too much power, the limiter would shut down
their power, until the household reduced their power load and reconnected their
breaker.
" The much less high-tech way of accomplishing this control is to have the micro-
grid operator communicate with the villagers regarding the status of the grid,
informing them that if they use more than a minimum number of appliances
they will cause the mini-grid to destabilize and black out. While inexpensive to
implement this solution might create some 'tragedy of the commons' incentives,
35
, , - -- -- _- 0101,161116506,- -_ kW._ 1.11 . _- --- . 1 11. - ll""Iiijimiligniiiiiiialkiiiiiiii- , -" ,- .- , -gkakdlduktM I "I - -u-, - - -
wherein it would be difficult to enforce cooperation.
These choices will be dictated by how the investor and community values the tradeoff
between efficiency and upfront cost; automated systems can raise the cost of the
system significantly.
1.3 Preview
In the rest of this document, I will lay out my approach for addressing the questions
described above. In section 2, I will describe the Reference Electrification Model
(REM), a model which has been developed by my team at MIT over the last three
years, and is useful for identifying the most cost-effective methods for rural electrifi-
cation. In section 3 I describe my methods for applying REM in order to arrive at
useful comparisons between different strategies of providing electricity. Section 4 lays
out the results of those analyses, as well as an assessment of the sensitivity of REM
to key parameters. In Section 5, I analyze the current regulatory structure in India
for off-grid systems, and suggest ways in which their proposed regulation regarding
off-grid electrification and grid-compatible microgrids could be improved. Section 6
provides the conclusion.
36
Chapter 2
The Reference Electrification Model
(REM)
2.1 Overview
The Reference Electrification Model (REM) is a planning tool meant to assist policy
makers, utilities, entrepreneurs, and investors in identifying the most effective ap-
proaches to providing electricity in a region. REM aims to design a near optimal
least-cost electrification solutions for a region at a highly granular level. It decides
between three electrification 'modes' - grid extension, microgrids and individual home
systems - for each household or other type of customer. The resulting design includes
the type, size, and locations of electricity generation, storage, and distribution assets,
and takes into account the specific amount and timing of consumer demand.
This model could be used by planners, including distribution companies, policy mak-
ers, and regulators, to estimate electrification costs and appropriate electrification
modes at a regional level. It is not meant to give a definitive implementation design
but rather to allow the decision-maker to understand the key factors influencing the
optimal design. REM can help answer 'what-if' questions like 'how important is cen-
tral grid reliability?' and 'if off-grid systems are restricted to low-carbon generation
37
sources, how does the cost change?'
Engineers and designers intending to provide electricity to specific areas could use this
tool to facilitate design decisions that appropriately balance level of service and costs;
since affordability is a key issue for many customers it may make sense to sacrifice
a certain amount of reliability in return for lower costs, but only up to a point.
Entrepreneurs could use the model to identify locations where their technology is
likely to be the most cost-effective solution. REM can also be used on a local scale
for the design of single microgrids. Entrepreneurs can use this capability to design a
local project and assess the financial viability of that project.
2.2 Precedents for REM
There are a sizable handful of papers assessing the suitability of on-grid vs. off-grid
electrification modes within a region. A number of them are focused on a regional
scale, incorporating high-level parameters that may be easier to find yet offering a
less detailed assessment. A few are more complex and able to handle more granular
local data and produce detailed designs, though none of them rival the granularity
and specificity included in the REM.
There are a number of papers that illustrate regional-level models using various com-
binations of regional parameters to show which areas are too remote and low-demand
to justify on-grid over off-grid; these papers, too numerous to discuss individually,
include [4, 40, 10, 38, 63, 52, 25]. There are also a handful of reviews, [60] and [26]
being the most notable.
For example, [29] estimates a radius around the existing grid, outside of which off-grid
systems would be considered more cost-effective. Their calculation uses a combination
of factors including the Levelized Cost of Energy (LCOE, $/kWh) for each generation
option, the project and component lifetimes, operation and maintenance costs, and
demand. Although the approach of applying a radius is rather crude, it can serve to
38
illustrate order of magnitude of the on-grid vs. off-grid tradeoffs, though it may not
accurately capture many costs.
A more granular approach than used above can provide a more accurate estimate of
the costs of electrification. Most significantly, the actual distribution of houses and
settlement patters can have a significant impact on the cost of the network required
to connect houses to a grid or a micro-grid [60]; using population density alone, as
done in many of these papers, could give a poor estimate of the cost-effectiveness of
microgrids, which depends significantly on how tightly the houses are clustered.
Several tools have been developed that operate at a higher level of detail, and begin
to approach REM's granularity. [341 goes down to a 2.5 km 2 resolution, and thus
lacks the household-level resolution of REM, but it does consider planned network
expansion, a greater variety of generation technologies (micro-hydro and wind, in
addition to solar and diesel), and uses the same basis for the calculation of diesel fuel
costs as REM. It decides between technologies on the basis of LCOE, which takes
into account the level of energy access desired, population density, grid characteristics
(though grid reliability is not considered), and the availability of renewable resources.
It does not consider the timing of demand, just the annual level; this omission could
result in a poor representation of the adequacy of resources like solar.
The United Nations' Dept. of Economic and Social Affairs offers a web-based tool
called the 'Universal Access to Electricity model,' which also helps the user determine
the most effective electrification mode on a regional scale, using a 10 km 2 grid 1.
Modi Labs at Columbia University has produced several papers addressing this topic 2 ,
culminating in the release of the Network Planning tool [27, 41, 43, 48, 64]. [64]
describes the use of a minimum-spanning tree for developing the least-cost distribution
network, based on household locations taken from satellite imagery, and defines the
'homogeneity index' for describing the amount'of clustering of houses in a region. [43]
has been widely cited in this literature; it introduces a model that selects the most
'Available at https://unite.ui.org/sites/unite.un.org/files/app-desa-electrification/index.html
2 Available at http://sel.colunbia.edu/publications/
39
- , - - , - - -- _'k,&h"L, , I - - MONIMM" , -- __ _ _ . _,,- .- I I , - - 1 1.
cost-effective electrification mode, and aggregates households into 'demand notes'
which are typically aligned with the smallest administrative unit in a country, to
best take advantage of demographic data and reduce computational complexity. It
uses a clustering algorithm based on a minimum spanning tree, and allows the user
to select a 'penetration rate,' representing the fraction of the households that the
planner wishes to have connected to an electricity source (presumably to represent
the gradual expansion of access). [431 finds that this penetration rate has a strong
influence on household connection costs; they also find that their model is not very
sensitive to the existing location of the grid, a result that seems counterintuitive.
The Network Planner, available from Modi Labs, is available online and allows the
user to upload their own datasets. This tool, described in [271 and applied to a case
study in [411, provides least-cost electrification mode decisions and a detailed cost
projection for off grid and on-grid solutions, and can be used on a variety of scales,
but does not go to the individual household level. It clusters individual households
at the community level, and considers the options of grid extension, diesel-powered
microgrids, and small solar systems with a diesel backup. While this tool is quite
detailed and sufficiently developed to be useful to planners, it does not have the
same household-level spatial resolution that REM does. It also does not consider
the relative timing of generation and demand; for hybrid PV-diesel systems, the tool
assumes that household, educational, health, and public lighting demand is met by
PV generation, and commercial and productive demand is met by diesel generation;
this assumption is questionable as PV generation would be unable to meet evening
and nighttime demand created by public lighting and household uses, unless a battery
is used (which would significantly affect the price).
In sum, although there is a considerable volume of work attempting to provide tools
for evaluating the best way to provide electricity, no available paper or tool that I
have found rivals REM's temporal and spatial granularity. While lower-resolution
data can make such tools easier to use, as data can be scarce in unelectrified ar-
eas, representing higher-resolution data could significantly improve the realism of the
40
model's outputs. In particular, the clustering of households within a community, and
the temporal relation between demand and renewable generation potential can have
significant impacts on the most cost-effective plan; REM is the only model examined
that captures both of these. Additionally, it is more flexible in determining the source
of generation for off-grid systems than most examined tools, and appears to be the
only option that places a value on the reliability of the electricity provided.
2.3 Description of REM
Here I will provide a mid-level overview of the algorithms used by REM to determine
the most cost-effective electrification mode. I will go into the detail necessary to
understand how the following analyses are conducted and understood. For a more
detailed examination of the algorithms within REM, see Ellman (2015), the original
thesis describing this model[17]. This model is also described in [12]. Updates to the
algorithms since the Ellman and Borofsky 2015 theses are described below.
The broad objective of REM is to determine the most cost-effective way to serve pre-
dicted demand, choosing between the options of extending the main grid, or building
a microgrid or isolated system. Included in the 'cost' of any possible option is the cost
of not meeting predicted demand. This parameter represents the cost of intermitted
supply. Requiring that any option meet 100% of predicted demand would likely result
in a system that is overbuilt and too expensive. This cost tells REM how to trade off
reliability against cost savings, and the determination of this parameter is discussed
extensively in section B. With this representation of the social cost of unmet demand,
REM determines the most cost-effective way forward for society to address the lack
of supply.
41
2.3.1 Inputs
REM requires a broad variety of inputs, detailing the location and type of customers,
the existing electricity infrastructure, and the prices of various options available for
meeting demand. The level of detail of the inputs can be adjusted based on the
available information.
The model requires the location and load profile of all the individual customers to be
connected. The geographic location of each connection can be gathered using individ-
ual household census data, satellite image processing or crowd-sourced pinpointing
of users. Our satellite image processing algorithm is under active development, in
this analysis I use the latest results from our building extraction efforts based on
images taken from Google Maps. This algorithm makes use of convolutional neural
networks to identify pixels that most likely belong to houses, based on an initial set
of training data. Census data from 2011 tells us that there are 600,000 households in
the district of Vaishali, so we sample (without replacement) that many pixels from
those identified to arrive at a set of possible household locations. While this method
is imprecise, it provides a sufficiently accurate representation for large-scale plan-
ning. Larger buildings may be assigned more than one connection point, but as such
buildings are likely to have a higher demand, this outcome may actually improve
representation of demand.
For each identified household connection point, we must also determine an expected
demand profile. The accuracy of the load profiles will vary greatly depending on the
available information. It might be estimated according to the average income level
of these customers or, where a increased precision is needed, deducted from specific
surveys that assess the energy services required by these users, and accounting for
factors like weather and sunset times. In the future, we may be able to moderate
this profile based on building size. For this case study of Vaishali, we use the same
demand profile parameters for all connection points, based on demand for a light,
phone charging, and fan, with the timing of these loads based on temperature and
42
sunrise and sunset timing, with an additional stochastic component.
REM also requires data on the existing power infrastructure, especially the location of
existing medium voltage lines and MV/LV transformers. This information is used to
determine the distance of a settlement from the centralized grid, and future iterations
of REM could use it to determine upstream alterations that might need to be made
to the existing infrastructure to accommodate increased demand. The data used in
this analysis was obtained from the North Bihar Power Distribution Company as
described in [171.
Available components, including wires, power electronics, batteries, and generation
options, with their respective costs, are another important input to REM. These
should reflect any relevant standards for grid infrastructure in the relevant locality.
Finally, the tradeoffs that REM makes are influenced by inputs like reliability con-
straints (how much of existing demand must be served), penalties for not serving
demand, and discount rates. These inputs are discussed further in Section 2.4.2 and
Appendix A.
2.3.2 Algorithm
Below, I outline the overall process for determining and designing the optimal electrifi-
cation options for a region'. Many of these steps could be performed in an automated
fashion, in the case of regional planning, or performed with more user-interaction, in
the case of design for a specific location. These steps are briefly described below:
Division of area into separate analysis regions: If the area of interest is large
(e.g. a country) it must first be divided into regions, which can be analyzed and
optimized separately, in order to manage computational burden. This could be done,
for example, based on demand density or based on distances between demand points.
REM currently employs a k-means clustering methodology for this process.
3The version of REM used in this study and described here refers to the version that was current
as of April 10th, 2016. This corresponds roughly to the 0.3.1 version.
43
Division of analysis regions into electrically isolated clusters: It is non-
trivial to determine which demand points will be electrically connected together (into
microgrids or grid extension clusters) or remain isolated with home systems. REM's
approach to clustering houses into groups starts with creating a minimum spanning
tree (MST) that connects all the houses. The MST identifies the nearest connection
to another house for each house, and this set of connections defines the possible
connections that can be made in grouping houses into clusters. REM first groups
houses into clusters for off-grid electrification, and then determines the clusters for
on-grid electrification.
The process of off-grid clustering trades off the additional costs of building more
electrical lines with the economies of scale of larger generation units. In order to avoid
the computationally-intensive task of determining the proper generation technology
and capacity for every candidate microgrid, REM first builds a 'look up table' which
contains estimates of the generation needed for microgrids of a sample of different
numbers of households, following the process described below. If different types of
demand are used, the lookup table will contain samples with different fractions of
each type as well. In the clustering process, REM interpolates between the samples
in the lookup table to arrive at an estimate for the relevant size of microgrid, where
size is defined by peak demand. This table is designed for each analysis region, to
account for possible differences in diesel prices and insolation which might affect the
optimal generation set-up in different places.
For the off-grid clustering, REM tries connecting houses together, starting with the
shortest branches of the MST and working up to the longest. To decide if the MST
branch should be a physical electrical connection, it compares the cost of electrifying
the houses on both ends of the branch to the cost if they remain unconnected. Note
that either or both ends of the MST branch may be single households or clusters
already connected houses. To do this, REM compares the cost of an electrical line
between the geographical centers of each cluster with the difference in cost of gener-
ation, operations, and management for the unconnected and the connected clusters.
44
If the cost of providing electricity to the houses or clusters on either side of an MST
branch is cheaper, then REM puts the two into the same cluster. REM loops over all
possible branches in the MST several times, until no additional lines are connected
or until a maximum number of loops is reached; this looping helps prevent the search
algorithm from getting stuck in a local minimum.
In this way, REM determines the least cost way of clustering houses for off-grid
solutions (microgrids and isolated home systems). It then determines optimal grid
extension clusters, which consist of combinations of off-grid solution clusters. The
logic for this process is similar to that of creating the off-grid solution clusters. It
begins with the remaining unused MST lines as the candidate connections. For each
possible connection of two clusters it considers five possible scenarios (which are
described in more detail in [171. The main change since the publication of [17] is the
inclusion of operations and management costs in the comparison).
1. The two clusters are connected together, and cluster one is connected to the
nearest grid connection point.
2. The two clusters are connected together, and cluster two is connected to the
nearest grid connection point.
3. Each cluster is connected to the grid separately.
4. Cluster 1 is connected to the grid, while cluster 2 is off-grid.
5. Cluster 2 is connected to the grid, while cluster 1 is off-grid.
Just as is done for the off-grid clustering, REM loops through all the candidate
connections for the on-grid clusters until no new connections are made in a loop or
the maximum number of loops is reached.
Design of electricity supply for each isolated cluster: For each cluster, local
generation and storage and/or a grid connection must be selected to optimally serve
the cluster's demand. This will entail selection of supply -components from a catalog of
components, based on the cluster's demand, energy resources, and costs. Currently
REM can choose between solar, battery storage, and diesel generation, employing
45
whichever combination of the three is most effective. REM utilizes a heuristic-based
pattern search to find the optimal design; this algorithm is based on the method
described in Hooke & Jeeves, (1961) and its application in REM is described in [171;
however, since the publication of [171, the approach has been modified in a way that
avoids local optima. REM can be used such that the generation for each cluster is
the value interpolated from the lookup table, from the previous step, or it can re-
optimize the generation for the particular parameters of the cluster in question; the
former saves a significant amount of model run-time and so has been used here.
The search space for the generation optimization algorithm includes the size of the
generator, the size of the battery, and the size of the PV panel. From a starting
point, the algorithm will try changing one of the variables at a time and see if the
cost is reduced. However, this arrangement can lead to getting stuck in local optima if
getting out of the local optima requires changing more than one variable at the same
time. A typical example of this would be a candidate design with a diesel generator
but no PV or battery, when the optimal design has no diesel generator but a PV
panel and a battery; adding either the PV panel or the battery alone will increase
the cost.
To avoid getting stuck in these local optima, the Hook & Jeeves algorithm has been
changed to be implemented in a master-slave setup, in which each 'master' fixes the
size of the diesel generator at a certain level. The master's 'slaves' then search the
space of PV panel and battery sizes, holding the diesel generator size fixed, and
return the optimal design given the diesel generator size to the master. Comparing
each master's result gives the optimal design, avoiding the problem of local optima.
REM identifies the cost of a generation design by simulating the operation of the
microgrid over a sample of time periods. This simulation reveals how much fuel is
used, how much demand is is not met, and how often the generator and batteries
must be replaced. These costs are used in the design search algorithm, in addition
to the upfront capital costs of the components. The simulation is done for a repre-
sentative sampling of time periods throughout the year; the number of samples and
46
the length of each sample can be determined by the user. The sample periods are
evenly spaced throughout the year, ensuring that different seasonal effects are taken
into account. See 117] for greater detail on how the simulation is implemented. The
regional-modeling version of REM does this detailed testing only for generating the
look-up table (using the interpolated values as the final design values), while the ver-
sion of REM used for the design of local infrastructure runs the detailed simulation
for determining the final design as well.
The current implementation of the simulation assumes that consumer's, demand re-
quests can be curtailed based on available generation, and that they can be differ-
entially curtailed based on whether they are high or low priority. This assumption
would clearly require some electronics to regulate demand in the actual implemen-
tation of a microgrid, and we have not included the price of such electronics in the
current analysis. Please see the forthcoming MIT Masters thesis by Vivian Li (2016)
for a greater discussion of the implementation of demand-response in the context of
REM.
Design of distribution network for each isolated cluster: Given power profiles
of demand and generation points, REM must select the appropriate wires, transform-
ers, and other distribution network components for minimum cost, considering tech-
nical constraints and losses. This process uses the Reference Network Model (RNM),
developed at IIT Comillas, with a few modifications to make it more appropriate for
a rural setting.
The RNM was developed at the Institute for Research in Technology (IIT) at Comillas
University in Madrid, and has been used by the Spanish Energy Regulatory Com-
mission to remunerate the distribution companies in Spain, with about 25 million
customers, with the agreement and collaboration of these companies [33, 441. RNM
has been used for similar purposes in several other countries and in several Euro-
pean research studies. RNM determines the most cost effective arrangement of wires,
transformers, and other components to supply electricity to a given set of connection
points with a given level of demand. Within REM, RNM is used to design both
47
the wiring of grid extensions and microgrids, using a catalog of network components
based on data we have received from the North Bihar Power Distribution Company.
RNM also assesses the cost of operation and maintenance for the network it designs,
based on expected preventative maintenance and maintenance in response to failures.
Determination of final power system: Multiple design options have now been
created for each household; on-grid, off-grid and isolated system. REM selects the
ones that are least-cost, where cost includes some penalty for not meeting consumer
demand reliably. After this selection process, REM is finished and reports the optimal
electrification design to the user.
2.4 Scenario construction in REM
2.4.1 User's perspective
By default, REM is geared towards a 'universal social planner' perspective, because it
seeks to minimize all costs to society, regardless of who incurs them. However, some
parameters of REM implicitly depend on the perspective and assumptions the scenario
is meant to have. The user of REM must still be clear about what perspective they
wish the analysis to take. The possible uses of REM are quite broad. The user can
ask 'what if..' questions, such as 'what will be the best solution if grid reliability can
be improved?', or 'what would be the impact of reducing the cost of capital available
to off-grid businesses?'. These sorts of questions tend to be more 'predictive', in using
the results of REM to attempt to forecast the most cost effective (and, thus assumed
most likely) way forward. On the other hand, REM can be used for more 'prescriptive'
questions. The user, for example an international development organization, could
take the outputs of REM to determine how to intervene to improve electricity access
most effectively. In any of these situations, the user must clarify what business models
and regulatory environment they are assuming for all modes of electrification, as these
will affect the inputs to REM. In particular, the proper selection of discount rate and
48
cost of nonserved energy will depend on the source of funding and the objectives, as
discussed below. As well, the user may wish to consider the timescale over which
they are planning, and what changes to the physical and business infrastructure may
occur over that time period. These changes may include the growth of the central
grid, improvements in reliability, and increased access to financial services.
2.4.2 Definition of Key Inputs
Appendix A contains a definition of the vast majority of the inputs for REM. Here
I will confine us to a discussion of the most interesting and relevant variables for
constructing a scenario.
2.4.2.1 Cost of Nonserved Energy (CNSE)
This value refers to the penalty added to the cost of supply in REM when projected
demand is not met, on a per kWh basis. The cost of nonserved energy, the demand
curve, and the minimum reliability threshold all work in concert to determine the
appropriate level of investment and type of service. Although the minimum reliability
threshold parameter does ensure that the demand of all households is met to some
extent, it does not ensure that the access happens at times of critical demand; as
discussed in Appendix A, this parameter refers to a crude average over a significant
length of time. Because it is currently a crude control, CNSE must account for some of
the costs of not having electricity in the first place, as well as the costs of intermittency
in supply. The selection of CNSE in REM is important to inform the proper level of
investment. Through the value of CNSE chosen, the user can specify how important
it is to build high-capacity, high-reliability electricity infrastructure, and how to make
the tradeoff between cheaper infrastructure and more reliable infrastructure.
In reality, the concept of nonserved energy incorporates a variety of costs which vary
along many dimensions, which are discussed at length in [151. These dimensions
include:
49
o Differences between customers: residential, business, urban, and rural con-
sumers all may have different types of demand and possible substitutes for
electricity, leading to a different valuation.
* Timing of the outage: the impact of an outage can vary depending on the time
of day, of the week, and the season.
* Duration of the outage: longer outages are often seen as more costly per unit
of electricity foregone and per hour as the length of the outage increases.
o Frequency of outages: more frequent outages may lead to larger costs, for ex-
ample a business may decide it is necessary to purchase a back-up generator.
e Advance notice given for outages: if an outage is anticipated, consumers can
mitigate its impact.
9 Capacity restrictions: limited supply and the inability to have larger loads can
restrict the usefulness of electricity supply, and reduce business opportunities.
Unfortunately, REM is not able to capture all of these dimensions. The only differen-
tiation available is in the distinction between critical and non-critical load, which are
assigned different values for CNSE. Critical values are higher than normal values, and
are used for loads that are higher-value to the consumer. For example, a few lights
at night would be considered critical, while a fan might not be. The assignment of
critical or noncritical is made within the definition of the consumer demand. This
distinction captures, somewhat crudely, an certain amount of preference for timing of
service (since critical loads will be higher at particular times) and preference between
appliances (since the value of CNSE is assigned to the particular use the electricity
is put to in the demand curve).
The basis of CNSE should be in the value placed (either by the purchaser or by
society) on the supply of electricity and the services provided by it. The perspective
used will depend on the motivations of the planner and the anticipated business
model and subsidy structure. For example, if the model is used with the assumption
that the government will subsidize access to and quality of electricity, it may be
more appropriate to use a 'social' CNSE. This more global cost could capture values
50
like lost health from kerosene pollution, reduced hours for businesses and students
to work, and reduced economic growth. Although REM would typically be used in
cases where the planner has already decided that it is worth it to provide electricity
services to all or most people, capturing these values is still important because they
will be reflected in the level of service and reliability that is most appropriate for the
situation.
Even though REM.is meant to address situations where the user has already decided
that it is worthwhile to provide electricity to everyone, the consideration of the cost of
not having any access at all in determining the REM inputs can be useful in ensuring
that REM makes the tradeoff between cost and reliability appropriately. Again, this
is because the minimum reliability constraint in REM is currently a very crude tool.
If the minimum reliability threshold is not a binding constraint, incorporating this
cost is necessary. However, if it is high and/or binding, the CNSE may only need to
include the costs associated with outages and capacity constraints. As well, it can be
easier to put a value on the cost of not having access to electricity than the cost of
outages, since it is difficult to know the preferences of consumers who have not yet
begun to consume, and many methods of determining the CNSE for outages rely on
consumer preferences.
2.4.2.2 Discount Rate
These figures (along with the lifetime of the investment) are used to calculate the
annuity (i.e. the cost per year) from the upfront costs. They are critical inputs
because the annuity is the basis on which REM compares different options. There
is one for grid extension, for microgrid, and for isolated systems. They represent
the time value of money for the relevant actor; typically this would come from the
financing terms (i.e. debt or equity) that they receive, but more generally it reflects
the opportunity cost of putting this money towards the system in question, rather
than investing it elsewhere (for example to purchase other equipment that could be
used to start a business). Thus, the actor who is providing the money to build the
51
electricity infrastructure influences what the discount rate is, because they will have
different opportunity costs for their money depending on who they are. For example,
for a bank, that opportunity cost may be not investing in stocks or other projects, and
for an individual consumer 4 who may decide to invest in an isolated home system it
may be not purchasing medicine, food, or other necessities; clearly for these two actors
the opportunity costs, and hence the discount rates, would be different. The discount
rate would also be expected to vary based on government regulation, business model
supposed for each mode of electrification, and financing arrangements.
The annuity value is what is used to compare different options of network and gen-
eration. The annuity is the sum of annual costs (like O&M costs, and non-served
energy costs) and the annualized upfront capital cost. The discount rate is used to
calculate the annuity with the following equation (where r = discount rate and L
lifetime of upfront capital cost):
Annuity =Upfront capital cost *r
(1-(1+r)-L)
The discount rate depends significantly on the regulatory and business scenario we are
considering. What discount rate we use depends on who will be funding the various
modes of electrification and what the overall goals of the planner are. The discount
rate encapsulates two main ideas; first is the opportunity cost of capital, and second
is the risk associated with waiting for a return on investment in an uncertain world.
The opportunity cost reflects the returns that one could get from putting that capital
towards something else; what that cost is depends closely on what opportunities are
available to the agent holding the capital. For a bank, this opportunity cost would
be the potential returns they could get though investing in a different project or in
the stock market. In the case of a local entrepreneur who might set up a microgrid
4The typical use of REM is to examine the options available to large-scale planners and investors,
not to individuals. However, the consideration of the possibility of investment by individual house-
holds is useful in two scenarios: (1) the planner wishes that the scheme for isolated home systems
is one in which the systems are purchased by individuals, possibly with the aid of subsidy and/or
financing from the government, and (2) the planner wishes to use REM in a more 'predictive' (rather
than 'perscriptive') manner, to estimate how electrification modes might evolve in the absence of
intervention, assuming that the more cost-effective mode is likely to prevail in real life as it does in
REM (acknowledging that this may be a heroic assumption).
52
business, these other opportunities could include other business possibilities, sending
their children to school, or buying a new piece of farming machinery.
The risk portion of the discount rate refers to the idea that the future is unpredictable;
there is uncertainty about market conditions, about the opportunities that might
be available in several years, about the likelihood of your investment to yield the
expected returns. Different businesses have very different levels of risk; a utility may
be relative low-risk, since it has a regulated rate of return, while a microgrid could
be significantly more risky, contending with issues like local acceptance, maintenance
quality, demand growth, and the arrival of the central grid, among others.
The Capital Asset Pricing Model (CAPM), which is commonly used to determine the
rate of return that investors should require from a project, exemplifies the dual nature
of the discount rate. (Note that discount rate and interest rates/rates of return can
be two sides of the same coin.) CAPM calculates the rate of return as the sum of the
'risk free' rate of investing, plus an added rate to account for the risk inherent in the
specific project and the returns that are expected above 'risk free' returns.
Whose discount rate?
The resources used to purchase electricity and build infrastructure have different
opportunity costs depending on whose wallet they are taken from, as well as the type
of opportunity costs that are considered. For this reason, we have allowed REM to
use a different discount rate for each mode of electrification; different sorts of actors
may invest in them and own them. The function of the discount rate is to incorporate
the time value of money and risk into the cost comparisons made within REM. Who
bears these costs and what types of costs we allow our analysis to consider both affect
the discount rate the user should use in REM.
With regards to who bears the cost, the actor or actors who actually invest their
money in electricity infrastructure have different opportunity costs and hence dis-
count rates. Even if REM is run from a 'global social planner' point of view, where
the user is interested in finding a design that maximizes overall societal welfare, it
53
might need different discount rates for different types of systems. The ownership
structure and risks of different business models are important because by choosing to
invest resources in electricity infrastructure, the actor in question would be forgoing
investment in other opportunities that could also create value for society. For exam-
ple, if we consider that a solar home system may be owned by a single household, the
discount rate used will be different than if we consider that the system is owned and
leased by a larger company; in the latter scenario, we would use lower discount rates
because they have less uncertainty and different alternative opportunities.
The perspective of the analysis we wish to conduct with REM also affects what types
of opportunity costs we wish to incorporate. There may be opportunity costs than an
individual entrepreneur may not consider but that a social planner (for example in
the central government) would want to incorporate. Examples of these opportunity
costs include the GDP, development, and/or political impact of other investments the
government could make with the money needed. to fund or subsidize an electrification
project. A user of REM who is more socially-minded would want to incorporate
these broader benefits. Conversely, a user of REM who is more market-driven (like
the owner of a business who wants to see where it is most likely that their off-grid
technology would be preferable to customers) would not want to include these social
opportunity costs in their analysis.
It is likely that a user of REM would be interested in funding or subsidizing to
an affordable level all modes of electrification; such users include government and
international development banks. In this case, since the source of the capital would
be the same across all modes of electrification, the opportunity costs would be the
same. This perspective, then, would dictate that the opportunity cost portion of
the discount rate be the same for all modes. However, different modes would still
have different risks associated with them, so this user may still wish to differentiate
between the modes slightly based on risk level.
Sensitivity to discount rate
54
Discount rates typically vary within a rather tight window, and small changes in the
rate can have significant impacts on the annuity calculated for different modes and,
hence, the ultimate recommendation on electrification mode from REM. The sensi-
tivity to discount rate also hinges significantly on the lifetime used for computing the
annuity (which varies between components to reflect their different useful lifetimes).
The longer the lifetime, the more sensitive the annuity is to the discount rate used.
For high values of the discount rate, e.g. above 15%, extending the lifetime does not
have a significant impact.
2.4.2.3 Annual Management Costs
This category includes the definition of the number of customers in a 'small' and
'medium' microgrid (the number for a large microgrid is assumed to approach infin-
ity), as well as the annual management costs for a small, medium, and large microgrid.
These figures are used to define the costs in administration and revenue collection as-
sociated with each additional household (or building) added to a microgrid. The
distinction between small/mededium/large is used to simulate economies of scale,
and these figures are used to fit a smooth exponential curve that is used to define
costs for all sizes of microgrids.
2.4.2.4 Per Customer Costs
For each mode of electrification, we define the per customer costs as the cost of the me-
ter or load limiters, if they are employed, any provided internal wiring, switchboards
and lighting. It does not include labor costs for installing these components.
This value is not the same as the connection charge that is charged by the utility or
microgrid provider. Sometimes the utility may inflate the connection charge in an
attempt to balance their books, and the charge may also be used as a commitment
mechanism to make customers invested in their new electricity infrastructure. Mera
Gao, for example, reports that they charge a $2 connection fee as a commitment
55
mechanism [131.
2.4.2.5 Network and component lifetimes
Each capital investment modeled by REM, from the network to the charge controller,
has an associated lifetime, which allows us to calculate an amortized value (the yearly
annuity) which we can use as a standard measure between options. The lifetime
represents how long until the component in question needs to be replaced, or how
long until the developer wishes to recoup their investment.
56
Chapter 3
Methods
3.1 Overview of Vaishali
All of the scenarios in this analysis use the data of the district of Vaishali, within
the state of Bihar, India. Bihar has the lowest rate of electricity access in India. In
2011, the Indian Census observed that only 16.4% of people in Bihar had access to
electricity. As of 2011, the district of Vaishali had a population of about 3.5 million
people over its 2000 sq. kilometers, a population density of 1700 people per square
kilometer. Vaishali is a flat, fertile, and mostly agricultural district. The high density
and relative flatness of Vaishali mean that we would expect grid extension to be a
cost-effective method of providing electricity to a large portion of its citizens if the
reliability of the grid were good.
Over the last 2.5 years, our team has worked with the local government and the
North Bihar Power Distribution Company to gather as much data as possible about
the existing infrastructure. We have observed reliabilities in the rural feeders that
are quite poor, as low as 15% in the peak hours of 6-8pm, and typically lower than
50% except in the middle of the night. The causes of this unreliability include lack
of supply, but poor distribution infrastructure is a much larger cause; grid operators
reported to us that the capacity of the line coming into a distribution substations
57
Figure 3-1: Map of Vaishali produced by UNICEF
was often insufficent to meet demand on all of the feeders, forcing them to shut down
some of the feeders at peak hours. Since agricultural loads are given free electricity,
the utility then has an incentive to cut the feeders leading to agricultural areas more
frequently. Reliability can be significantly better in urban areas, though it is still far
from ideal.
Vaishali is administratively divided into subdistricts, as shown in Figure 3-1. It in-
cludes three urban areas, shown in yellow. We have used the boundaries of these
subdistricts to divide the region into separate analysis regions, which are processed
individually within REM. Conversations with local officials indicated to us that mi-
crogrids would be less likely to cross the boundaries of these subdistricts, so it seemed
a natural way to divide up the region in the analysis.
58
VAISHALI DISTRICT
unicef*
3.2 Inputs for Case Study Area: Vaishali, India
To assess the value of grid compatible microgrids we will make a comparison between
three different scenarios run by REM. While each scenario requires its own set of
assumptions and inputs, many of the inputs are common across all three scenarios.
There are a great deal of assumptions that must be made to put together the inputs
for REM. The most critical and complex of these assumptions are discussed in this
section, and a comprehensive listing can be found, in Appendix A. For each scenario,
this section will discuss how the inputs are changed from the base case defined here
and in Appendix A.
3.2.1 Cost of Nonserved Energy (CNSE)
Values Used:
Critical: $2/kWh, based on [591. This is higher than estimates from substitution
analysis and from the Power Corp. of India's Value of Lost Load assessment (discussed
below), but these numbers are known to be left-skewed and to not capture important
effects like the GDP impacts of lack of electricity access.
Noncritical: $0.51/kWh is a lower bound, taken from the Power Corp. of India's
Value of Lost Load assessment.
Reasoning:
There are several possible ways of calculating the Cost of Nonserved Energy, but
none of them are likely to ever find the 'true' value. Indeed, there is arguably no
one 'correct' value waiting to be found. By examining several methods and taking
into account what they miss, we can assess a reasonable estimate for the purposes of
running REM.
Via substitution: See Appendix B for more detail on how to calculate CNSE by
looking at substitutes for electric lighting. In summary, we use kerosene lighting
59
costs as an estimate for CNSE based on the units of 'cost to provide the service
of illuminating a house'. This is a reasonable basis for a CNSE because people's
expenditure on kerosene is an indicator of the lower bound of the value they place on
energy services - particularly on the service of lighting. Lighting is one of the highest
value uses for electricity (with agricultural water pumps and cell phone charging the
likely competitors), so a CNSE based on this use is likely to be 'a lower bound on the
upper bound' of an individual's valuation of electricity services.
This method does not capture added benefits of using electric lighting over kerosene
such as health, increased quality of light, and the social value to development and
GDP of energy access. While an individual consumer may not take all these factors
into account, a social planner who has the interests of the whole country in mind
would wish to incorporate these values into a higher CNSE.
Via reported VOLL: Power companies regularly determine a 'Value of Lost Load'
(VOLL), representing the cost of an outage, to guide their investment decisions.
While the cost of an individual outage is likely to be lower than the cost of chronically
unreliable or low-capacity supply, this is still a useful yardstick. A report by Wartsila
India (2009) reports that the Power Grid Corporation of India estimates the VOLL
in India as Rs.34/kWh to Rs.112/kWh, or $0.51 - $1.67/kWh [581. The range is used
because the cost of an outage is dependent on a variety of categories. [59] notes that
customer type (industry, service sector, households), perceived reliability level, time
of occurrence, duration, and advance warning can all affect the determination of the
VOLL [581.
[59]notes that for developing countries the literature strongly suggests a range of
VOLL that is 1-10 $/kWh, and likely 2-5$/kWh, higher than the Power Corp of
India's suggestions [58] ([59] notes their values are likely to be left-skewed). The
VOLL for rural consumers in India would likely be on the low end for variation within
consumer type, low end for perceived reliability level, and the high end for variation
based on time of occurrence, duration, and advance warning. Without information
on the relative magnitude of these effects, it is difficult to say how these customers
60
are likely to compare to the average.
Summary: Substitution analysis suggests a minimum of $0.37 to $1.84/kWh (CFL vs
Incandescent) for critical CNSE. VOLL studies from India suggest a range of $0.51 -
$1.67/kWh and a literature review of VOLL studies in developing countries suggests
a range of $2 - $5/kWh [151. All of these values come with the caveat that they are
likely to be left-skewed (i.e. underestimates).
3.2.2 Discount Rate
Values Used:
" Grid extension: 13.3% for NBPCDL
" Microgrid: 11% at minimum
" Isolated system: 14% based on SELCO financing rates
Reasoning:
Grid: For the extension of the grid, we could use 10% in the absence of other data.
This number is based on a paper from Climate Policy Initiative "Solving India's
Renewable Energy Financing Challenge" which bases their discount rate on the gov-
ernment cost of borrowing (7.83% at the time of writing, 7.77% on Jan '16, based on
Indian 10 year bonds), plus a risk premium of 2% [50]. This risk premium is meant
for large-scale renewable energy, so it may not be representative for a utility. Indeed,
this number may be too low as evidenced by regulatory documents in India.
The regulatory documents for the North Bihar Power Distribution Company Ltd
(NBPDCL) for Fiscal Year 2015-2016 from the Bihar Electricity Regulatory Com-
mission (BERC) grant NBPDCL a rate of return on their equity of 14%, and an
interest rate on their loans of 13%1. The rate of return on equity is established by
Regulation 73 (2) (c) of the Bihar Electricity Regulatory Commission regulations of
'Taken from Table 6.57 and Table 4.62, found at http://berc.co.in/media/Tariff-
Order/Tariff%200rder%20NBPDCL%20FY%202015-16.pdf Accessed April 6, 2016
61
2007. These regulatory documents assume that the debt:equity ratio for capital in-
vestments will be 70:30, which gives us a weighted average cost of capital of 13.3%
2
Microgrids: This number will vary based on the regulatory context that we consider.
The risk premium will be lower if we consider a regulatory structure that accounts
for grid-compatible microgrids in a reliable and trustworthy way, but it will be higher
if the regulation does not provide investors with sufficient reassurance that their
investment will be properly valued and compensated when they connect to the main
grid. The risk premium will again be higher when we consider non-grid-compatible
microgrids, as the companies planning these will lose their whole investment when
the grid arrives; additionally, previous conversations with them have revealed that
they typically look for a relatively short payback period (on the order of 5 years).
TaraUrja, a microgrid company with several installations, uses a discount rate of
10% in the calculations that they sent to us, and the World Bank cites 10.3% as
the interest rate "that usually meets the short and medium-term financing needs of
the private sector." In light of the justification for 10% as an appropriate interest
rate for a utility or large project (cited in [50]), this rate does seem optimistic. [501
cites 12.3% as the commercial rate of interest applied to renewable projects in India,
so this is perhaps a decent lower bound, though it is meant for larger utility-scale
generation installations. [3] and [231 use an interest rate of 12% for their calculations
regarding solar microgrids (though neither specifies a source for this figure). It seems
that 10% is a lower bound for microgrids. One would expect microgrids to have a
higher discount rate than utilities as they are often riskier investments, so in that
light a higher rate may be justified.
Individual home systems: Individual home systems could be provided through a
variety of business models which could be relevant to REM. They could be provided
2 However, these regulatory documents note that NBPDCL "has projected funding of capitaliza-
tion through Grants at 94% and through Loans at 6%," meaning that the capital investments of
NBPDCL could be more likely to be dictated by the preferences of the central government grant-
maker than by NBPDCL and their discount rate.
62
through the utility or as part of a larger private off-grid service provider. They could
be supplied on a fee-for-service model by a private company, similar to the business
model used by Off-Grid Electric in east Africa3 . Or, single households can purchase
them themselves from the government or a private entity, with or without financing.
In all of these models, the government could choose to provide a subsidy or subsidized
loans; for example, the government may sponsor a program to subsidize the provision
of such systems through the utility, or they may sponsor a program of discounted
financing for households to purchase their own system. The business model supposed
here has a large effect, perhaps even more so for this mode than the other two because
they range of possible business models is so great. Very high discount rates could
make individual home systems very unattractive relative to other solutions within
REM.
Individual customers are known to have very high individual discount rates (on the
order of 20+%), likely due to high uncertainty surrounding the future. However, a
lower discount rate would be appropriate when financing is available to households,
especially if government-subsidized financing is available. [231 observes: "Ekholm et
al. (2010) use discount rates of 62-74% for rural households and 53-70% for urban.
Reddy and Reddy (1994) estimate an internal rate of return of 28% for a switch from
kerosene lamps to electricity- which could be a lower bound on the discount rate".
One highly-successful example of a business model for solar home systems is Grameen
Shakti, a social enterprise which supplies solar home systems with financing through
the microcredit scheme supported by the Grameen Bank in Bangladesh. Households
wishing to purchase a system can have a 3-year loan with a 12% interest rate, 2 years
for 10% interest, and 1 year for 9% interest [611. In India, SELCO works to provide
purchasers of solar home lighting with financing; they state interest rates ranging
5%-14% over 3 to 5 years with a 10-25% upfront deposit (25% is required by the
Reserve Bank of India) '.
3More information at http://offgrid-electric.com/#home
4http://www.selco-india.com/finance.html Accessed April 6, 2016
63
3.2.3 Taxes and Tariffs
Values Used:
" Tax: 14.5%
" On-grid tariff: Rs 170/month for the 65% of APL households, 60/mo for the
35% BPL.
" Off-grid tariff: Rs 120/month for a grid-compatible microgrid, Rs 100/month
for a non-grid-compatible microgrid.
Reasoning:
These two values are not used by REM, but are critical for post-processing. TaraUrja
reported 5 that the tax charged to electricity providers is 14.5% of tariffs collected.
Tariffs charged for microgrid service to households are typically a flat monthly fee.
During a variety of conversations in India we have heard of tariffs ranging from 60
Rs/month to 150 Rs/month for very similar services, typically a light or two and
phone charging for a few hours in the evening. Mera Gao reports that they charge 100
Rs/month for evening service from their solar-based microgrids', while Husk Power
charges 60 Rs/month for evening service from their biomass-based microgrids'. The
Subdistrict Magistrate of Mahnar District, Vaishali, India reported to us that many
villages have a single diesel generator supplying some service to the village, and the
charges are typically Rs 90-100/month for one to two lights'. TaraUrja reports to us
that they charge 120 Rs/month for one light and a phone charger, and an additional
40 Rs/month for an additional light.
Some microgrids to have volumetric tariffs. TaraUrja charges volumetric tariffs to
its larger customers, which range from 16-22 Rs/kWh, with lower rates for those
demanding more capacity. In the Indian village of Darewadi, a microgrid operated
5Meeting held January 2016
6
"Microgrids Lend a Shine to Solar" Times of India June 7, 2014.
http://timesofindia.indiatimes.com/business/india-business/Microgrids-lend-a-shine-to-
solar/articleshow/36168208.cms Accessed April 6, 2016
7Conversation with Sushanta Chatterjee, July 2015
8 Conversation July 2015, Vaishali, India.
64
Category Type of Charge Rate in Rs (USD)
Kutir Jyoti (Below Unmetered Rs. 60/month
Poverty Line) Metered (Rural) 1.7 Rs/kWh for first 30 kW(same as DS-1 after)
Metered (Urban) 2.05 Rs /kWh for first 30
kW (same as DS-1 after)
Unmetered Rs. 170/month
DS-1: Rural single Metered: first 50 kWh 2.1 Rs/kWh
phase <2kW Metered: 51-100th kWh 2.4 Rs/kWh
Metered: Above 100 kWh 2.8 Rs/kWh
Table 3.1: Selections from the Bihar electricity tariff schedule for FY 2015-
2016. http://www.biharchamber.org/content/new-tariff-rates-electricity-nbpdcl-
and-sbpdcl-fy-2015-16, Accessed April 6, 2016
by Gram Power charges 20 Rs/kWh [3].
Meanwhile, grid tariffs are heavily subsidized. Although most customers are not
charged volumetrically, the charges come out to approximately 1.5-3 Rs/kWh [3, 28],
significantly lower than microgrid costs. The approved tariff rates for the North and
South Bihar Power Distribution Companies for fiscal year 2015-2016 are shown in
Table 3.1. Data from the Reserve Bank of India indicates that 35% of rural people in
Bihar are below the poverty line, and 65% are above it', so I use the appropriately
weighted average tariff in my calculations.
3.3 Scenarios run in REM
REM's detailed cost projections allow us to compare the potential costs of a variety of
scenarios, and to assess who will be responsible for those costs. In order to assess the
attractiveness of policy supporting grid-compatible microgrids, we need to understand
how much such a policy will cost, who will pay that cost, and who will receive the
benefits. To answer those questions, we use the following three scenarios in REM.
They are each described in greater detail in the following sections.
9 Reserve Bank of India https://www.rbi.org.in/scripts/PublicationsView.aspx?id=15283, Ac-
cessed May 10, 2016.
65
1. Mixed solution with grid-compatible microgrids: This is the default
operation of REM; it selects the optimal electrification mode for each cluster of
houses and builds all systems to meet the grid code. This scenario is run for
both a high and low grid reliability scenario.
2. Mixed solution with non-grid-compatible microgrids: This scenario is
similar to the above; the main difference is that REM is given a different catalog
of network components for use with the microgrids than for the central grid.
As well, the discount rates and operational costs would be adjusted to account
for a different business model.
3. All grid-extension: Force REM to connect all customers to the central grid
by removing isolated generation as an option.
Each scenario will be run with a range of key variables to assess their sensitivity.
Their output can be used in three main ways.
First, we can examine what the cost savings of using a mixed on-grid and off-grid
approach would be for the region of Vaishali, by comparing the output of scenarios
3 and 1. It is important to get a good estimate of this number for discussions with
policymakers, because it can help address the tendency we have seen in state-level
administrators to prefer grid extension as a near-universal solution. These cost savings
refer both to the change in overall, societal costs, as well as the difference in cost
actually paid by the government (including costs to the state-owned distribution
company as well as any subsidy paid to off-grid operators).
Second, we can assess how much additional investment is required to bring a microgrid
up to compliance with grid-compatible standards - both for an average microgrid and
for the region as a whole - by comparing scenarios 2 and 3. These costs include higher-
quality network infrastructure, and power electronics capable of interfacing with the
grid (for example, higher-quality inverters built for both modes of operation).
Finally, using the output of scenario 2 and an estimate of the rate of grid expansion
we can assess the magnitude of the investment likely to be 'lost' as the grid expands
66
and puts non-grid-compatible microgrids out of business. Unfortunately, difficulties
with scenario 2 rendered the results of the second and third analyses less useful, but
there are still worthwhile insights to be gained from these comparisons.
3.3.1 Business model and regulatory assumptions
Crafting these scenarios requires not only finding the proper technical data inputs,
but also imagining the regulatory and business context, since these factors will in-
fluence costs of capital (e.g. the discount rate used), tariffs, and economies of scale
in operational costs. We assume that the expansion of the central grid would be
undertaken by the utility, while the microgrid projects could be undertaken either by
the utility (as has been done in Tanzania [54]) or by private entrepreneurs, which has
been the norm thus far in India.
The choice of business model will affect the discount rate and operational costs for
off-grid projects. In the utility-runs-all scenario, there will be savings in capital cost
and in operational costs. In the private entrepreneur model, there is a potential range
of economies of scale. For example, we could envision a situation in which all isolated
systems are owned and managed by one company (for example, in a business model
like that of Off Grid Electric's), and each microgrid company is responsible for a
number of microgrids. Alternatively, we could envision private investment happening
in a more isolated way, in which each isolated system and each microgrid has its own
owner/developer. Of course, there is a continuum between these two visions.
For the two scenarios involving off-grid projects, we examine two cases, one on each
end of this business model continuum. One will assume that all projects are centrally
owned, funded, and operated (e.g. by the distribution utility or by a large devel-
opment investor). The other will assume that all projects are funded in a one-off
fashion, resulting in higher operational costs and costs of capital; while there is very
little data publicly available regarding how these cost might change exactly, I have
made some assumptions based on the reasonable distribution I have observed of these
67
costs in the field and the literature.
We can also envision two possible scenarios for how tariffs might turn out under either
scenario. While this input is not relevant for REM, it is relevant for the subsequent
financial analysis. Regulators may require tariffs to be harmonized between the mi-
crogrids and the central grid, or they may allow the microgrids to charge whatever
their customers find reasonable. My assumptions for this regulation will be described
in the sections for each scenario.
3.3.2 Scenario 1: Grid-compatible microgrids
In this scenario, REM is run under its standard settings; it is allowed to choose any
electrification mode for each cluster of households, and it uses the same standards for
building the central grid and microgrids. Since both are built to the same standard,
the microgrid will meet grid-compatibility standards for its network. After running
REM, additional costs for interfacing between the central grid and the microgrid will
be added to the financial assessment.
This scenario assumes that there is sufficient regulation to enable investment in grid-
compatible microgrids, meaning that the investors feel that the risk posed by grid
expansion is sufficiently low to enable them to lend at rates that the project can
pay for. This change in the cost of capital is a key effect, as it is the clearest ben-
efit of regulation which aims to remove the downsides of the risk of grid expansion.
Other benefits include changes in the greater structure of the grid to encourage more
distributed, renewable generation, which could reduce losses as well as the climate
impact. However, these effects are beyond REM's modeling scope.
This scenario assumes that grid-compatible microgrids are allowed to charge a differ-
ent tariff from the grid prior to connection. The pros and cons of such an arrangement
are discussed in Chapter 5. I assume the tariffs charged by the central grid are those
charged by the North Bihar Power Distribution Company to households, shown in
Table 3.1.
68
Scenario-specific inputs:
* discount rates for microgrids
- Integrated business model: 13.3% (same as grid)
- One-by-one business model: 11%
" per customer costs for microgrids: $25, includes a meter.
" Network lifetime for microgrid: 20 yrs
" Tariff charged for microgrids: Rs 120/month
3.3.3 Scenario 2: Non-grid-compatible microgrids
This scenario is similar to the above scenario except the microgrids are not built
to the same standard, so they are cheaper but not grid compatible. This scenario
attempts to replicate the requirement for a very basic microgrid, of the type that we
have seen Mera Gao implement; these microgrids typically provide very basic service,
1-2 lights and a phone charger, for a few hours per evening. The wiring for these
microgrids is typically in DC, not in AC; while RNM is not capable of handling DC
grids, we attempt to replicate its characteristics as closely as possible by ensuring
maximum power delivered and voltage drops are represented accurately. To do so,
the resistance and current rating of the wires are adjusted - an important side effect
of this adjustment is that losses are significantly under-estimated in the design of
the wiring, which may lead to a slight underestimate in the cost of the system. The
line characteristics are based on a set of aluminum wires with diameters from 1.1
to 2.5mm; this range is based on characteristics of a sample line provided by the
microgrid company SELCO, who uses aluminum wires that are 1.78mm in diameter.
Ultimately, however, this method produced results that were somewhat nonsensical.
As a result, we employed a cruder method which encouraged cheaper, low-quality
microgrids by lowering CNSE to $0.4/kWh and $0.8/kWh for normal and critical
load, respectively, and increasing the discount rate significantly.
We assume the tariffs charged by the central grid are those charged by the North
69
Bihar Power Distribution Company to households, shown in Table 3.1.
Scenario-specific inputs:
e discount rates for microgrids
- Integrated business model: 13.3% (same as grid)
- One-by-one business model: 14%
- Additional runs assuming faster rate of grid expansion, with discount rate
200 basis points higher.
9 Network lifetime for microgrid: 10
* per customer costs for microgrids: $15 (no meter)
* tariff charged for microgrids: Rs 100/month
o Network catalog attributes and costs
- Impedance is 0
- Inverter cost is 0
- MV and HV lines eliminated
- Assume aluminum wiring
3.3.4 Scenario 3: All grid-extension
In this scenario, generation option are so expensive that REM should never choose an
off-grid system; instead, every household will be forced to be connected to the grid.
The minimum reliability requirement will ensure that every household is connected
to the grid. Tariffs charged by the central grid are assumed to be those charged by
the North Bihar Power Distribution Company to households, shown in Table 3.1.
Scenario-specific inputs:
All inputs are the same as in scenario one except generation options have been made
very expensive, so that REM is forced to choose all on-grid options.
70
3.4 Financial analysis and assumptions
After running these three REM scenarios, their output is used to calculate the cost
of each scenario to the government. We assume that the government will make up
any shortfall in revenue for both on-grid and off-grid projects through subsidies. This
is common for the distribution companies, though currently the subsidy system in
India is not very effective at subsidizing off-grid projects. Stakeholders in that sector
report that obtaining the subsidies requires navigating a significant amount of red
tape, and the timing is highly uncertain.
Since the question we are concerned with answering is 'how much government support
would each of these scenarios require,' it is appropriate to assume that a negative
NPV for a project would be made up in the form of government subsidy. For off-grid
projects, clearly a more efficient method of providing the subsidy would need to be
devised for this hypothetical scenario to play out.
REM outputs the costs of all components and their lifetime, allowing us to assess
upfront and ongoing capital expenditures. Operations and maintenance costs are
provided as well. REM also outputs information about the number of households
within each electrification project, and the total demand served; this information
allows for the calculation of revenue. With the costs and revenues in hand for each
electrification project, we can determine the overall NPV of the project, assuming
that the grid never expands.
If scenario 2 had worked for realistic simulation of non-grid-compatible microgrids,
we could have determined the value invested in off-grid projects that is abandoned as
the grid expands. It would be a worthwhile analysis to conduct if this functionality
can be implemented successfully. This value is important both because it represents
a deadweight loss to society, and because it will serve as a warning to other poten-
tial entrepreneurs and investors, potentially scaring them off or at least significantly
driving up the rates of return they expect from the project, so they can make a profit
before the grid arrives.
71
Assumptions regarding the rate of grid expansion could significantly affect the cost of
different scenarios. Although REM determines an 'optimal' arrangement of on- and
off-grid projects, it is reasonable to assume that the grid will continue to expand in
at least some areas. This expansion will be driven by policy directives as well as by
increasing demand for electricity, which will expand the range of locations where the
grid is cost-efficient.
Grid expansion affects the government subsidy provided to off-grid projects, which
would need more subsidy once they are connected to the central grid and must charge
lower tariffs. As well, it affects government spending on the grid; although REM is not
particularly well-suited to assessing this cost, but it can be estimated by comparing
the financial costs of the grid in scenarios with low and high grid expansion. Assump-
tions about the rate grid expansion will also affect the risk to non-grid-compatible
microgrids, and change the value that is abandoned when the grid arrives. A range
of assumptions should be tested regarding how long it will take before the grid en-
croaches on the microgrid's territory.
72
Chapter 4
Results
4.1 REM Sensitivities
Before utilizing REM for a policy analysis, it is vital to understand how sensitive
the model's results are to the values of various inputs, in order to assess whether the
model behaves as expected as well as over what range we should expect the results
to hold true. REM's inputs are comprised of hundreds of separate values, and not
all of them have been tested in this analysis. Instead, the analysis focuses on those
variables we suspect might have the greatest impact.
4.1.1 Results
Stochasticity of runs There is some inter-run variability, even between runs with
identical inputs. This difference arises from stochasticity in the demand profile genera-
tion algorithm described in [17]; this is the only portion of the model with randomness
involved. However because this aspect is very important for building the look-up ta-
ble and determining the costs of supply, it can have wide-reaching effects within the
model results. It is important to quantify this variability first, to determine at what
point the effect of a change in inputs can be considered significant. Six identical runs
73
.1 - -_ - I. I -_-,W"",; Ibl - - --. 1- - . . ." I -- I I~ , - -UL&W .Wa -- I
Inter-Run Variability in Customer Assignment
100% - -- - - - -
95% --- 47409 46337 46154 4653 - 32
66 87
90% --- m- !M7---5--7
12 85%
80%
75% Grid Ext.
70% 412507 413578 413761 426887 41334 427769 i Isolated System
Z 65% Microgrid
60%
55%
50%
1 2 3 4 5 6
Run number
Figure 4-1: Inter-run variability for assignment of customers to different electrification
modes. Please note that the Y-axis starts at 50%, to make the inter-run variability
more visible.
using the input parameters described in Appendix A yielded the following inter-run
variation, with surprising results.
" There appears to be a bimodal distribution of the model outputs, where there
are solutions around two somewhat different optima. Figure 4-1 illustrates this
bimodality. The two optima, or 'modes', are reasonably different from each
other. However, within each mode the variations are tiny. Between modes, the
average number of households in a microgrid shrinks by 20%, while the average
battery size drops dramatically, from 46.2 kWh to 4.64 kWh. This difference
is not due to an outlier effect; rather, the whole distribution of battery size
changes significantly, as shown in Figure 4-2.
" Assignment of customers to electrification mode: The largest difference across
the 6 runs was a change in assignment for 3.2% of consumers, who mostly switch
between microgrid and grid extension. The variability in customers assigned to
isolated systems was extremely low (.0025% of total customers). A graph of
this variability is shown in Figure 4-1; it illustrates the breakdown in customers
between microgrid (blue), isolated systems (red), and grid extension (green).
The maximum inter-run variability for the number of customers in microgrids,
74
Size of batteries in microgrids Size of batteries in microgrids
127 microgrids have no battery 122 microgrids have no battery
80 80
F6 960
E
C C
240 240
20 20
0 00 200 400 600 800 1000 0 200 400 600 800 1000
Size of batteries (KWH) Size of batteries (KWH)
(a) Histograms of battery size distribution within the two runs that show the less common mode.
Size of batteries in microgrids Size of batteries In microgrids
0 microgrids have no battery 0 microgrids have no battery
80 80
F60 ?60
E E
240 2 40
20 20
0 - C
0 200 400 600 800 1000 0 200 400 600 800 1000
Size of batteries (KWH) Size of batteries (KWH)
(b) Two representative histograms of battery size distribution from the four runs that show the
more common mode.
Figure 4-2: Difference in distribution of batteries between identical REM runs. There
appear to be two 'modes' that exhibit little variation within themselves, but great
variation between the two modes. Note that the Y axis is the percent of microgrids
in the simulation with the relevant size of battery.
isolated systems, and grid extension within the more common mode is 0.27%,
0.0004%, and 0.27%, respectively.
e Cost of electricity per kWh: There is little variability here. Microgrid cost per
kWh varied by 0.75%, grid extension cost by 4.04%, and isolated system cost
by 5.6%. Within the more common mode, the inter-run variability is 0.21%,
0.12%, and 1.1%.
e The variation in total financial cost (i.e. cost excluding the cost of non-served
energy) was 0.14%. Excluding the less-common mode does not affect the spread
in financial costs.
75
Changes in the lookup table The lookup table is generated by the algorithm
described in section 2.3.2. For a sample of microgrid sizes it produces sample designs
and an associated cost. In theory, the relationship between the number of households
in a microgrid and the cost per customer should be a smooth curve. If it is not so, then
the clustering step may become stuck in a local optima that is not a global optima.
It important to check this intermediate output before interpreting any runs, because
on rare occasions the model can produce lookup table curves that have local optima.
In the 50 or so runs done for this thesis, 2 had undesirable lookup table curves, which
would arise unpredictably. Figure 4-3 shows an example of the lookup table curves
for a set of four runs, which only differ in the diesel price used; they have values of
1.75, 1.5, 1.25, and 1 $/liter. Run 5, in the upper right, and Run 6, in the lower left,
have local minima which could result from switches between generation. However,
the actual curve used to determine prices in the clustering step is a smoothed version
of the curve displayed. Nevertheless, knowing this, we might be more skeptical of the
results from these runs.
Impact of diesel price The price of diesel is a strong determiner of the model's
outcome. Especially because Vaishali is a relatively flat, dense region, grid extension
is a relatively competitive option for a large number of households. When the cost
of diesel drops such that microgrids are competitive with grid extension, it changes
how customers are assigned to electrification mode quite dramatically. Figure 4-4
illustrates this sharp transition as microgrids reach cost parity with grid extension.
Even at high diesel prices there are a number of communities for whom microgrids
are still the most cost-effective option, and that choice remains constant up to very
high diesel costs. Diesel appears to make very large microgrids economical, and at
this scale they can compete with the grid. In the absence of large diesel generators,
the average size of a microgrid drops to below 100 households; diesel generation is
not completely absent in all these smaller microgrids, but they do tend to be based
on PV with a battery, and maybe a small supplementary diesel generator.
76
200
180
160
0
8140
120
100
80
0 5000 10000
Number of customers
(a) Run 4
0 5000 10000
Number of customers
(c) Run 6
15000
200
180
160
8 140
CL
120
100
80
15000
0 5000 10000
Number of customers
(b) Run 5
0 5000 10000
Number of customers
(d) Run 7
Figure 4-3: Changes in lookup table over a small range of diesel prices. Run 4 has a
price of 1.75 $/liter. Run 5: $1.50/liter. Run 6: $1.25/Liter. Run 7: $1/Liter.
77
200
180
160E
3140
r)
120
100
80
200
180
160
8 140
120
0
100
80
15000
15000
~~Cost of diesel e $criiatoL md Cost of diesel $/Lor sz
to
:g 00 0
CC
E
0 0
0 0
0~0
a) > N
.0 0D
0 0 0 0 0 0 0 0 1ODo000
Z :,0.5 1.0 1.5 2.0 2.5 0.5 1.0 1.5 2.0 2.5
Cost of diesel, $/L Cost of diesel, $/L
Figure 4-4: The price of diesel causes sharp transitions in the decision outcome of
REM
Interestingly, the most current price of diesel we have, from January 2016, is $0.82/liter,
right at the transition point in Figure 4-4. Two years earlier, when oil prices were
much higher, we would certainly be below the transition shown. Of course, this tran-
sition might happen at different values if the costs of the grid were to change, for
example by changing the cost of supply or changing the reliability.
Grid reliability The reliability of the grid also has a very strong impact on REM's
output, creating sharp transitions similar to those observed for diesel price. See
Figure 4-5 for an example of this tipping point; under the current model settings, the
transition happens between 75% and 85% grid reliability. This tipping point indicates
that for many households in the study area, the cost of the grid extension is less than
the cost of a microgrid or solar home system, if the costs of non-served energy are
ignored. Of course, these results have the caveat that upstream reinforcement in the
transmission and distribution system is ignored because the model cannot currently
capture it, though it would certainly be required and would likely be expensive. We
do not know how the magnitude of that cost would change the difference between
grid extension and microgrids.
78
Cost of diesel and electrification mode Cost of diesel and microgrid size
Importance of Grid Reliability Effect of Grid Reliability
For Customer Assignment On Total Cost
E I 4.0E+07
0. J3.5E+07
6 -I I 3.0E+07
0:4 6 2.5E+07-
10.2 2.OE+07
0- 1.5E+07
50% 60% 70% 80% 90% 100% 50% 60% 70% 80% 90% 100%
armd Roab"ty Grid RelabNty
(a) Effect of changing grid reliability on assign-(b) Effect of changing grid reliability on the an-
ment of customers to grid extension nualized cost of building and maintaining elec-
tricity infrastructure for Vaishali
Figure 4-5: These figures show the strong effect of grid reliability on the assignment
of customers to electrification mode, and the effect on total financial (i.e. excluding
the value of unmet demand) cost.
Discount Rate The discount rate, both in absolute value and the relative value
between the various modes of electrification, has a modest effect on REM's output.
The absolute value of the discount rate has a small effect; with discount rates for all
modes set equally, a shift from a 5% to a 10% discount rate changes the assignment
of the customers from 12% connected to a grid extension to 21%. A higher discount
rate appears to favor grid extension in the scenarios examined. Smaller changes in
discount rate, in the 10-14% range, appear to have a dramatically smaller impact on
model output. Shifting the discount rate for microgrids from 12% to 14% decreased
the customer assignment to microgrids for only 1.2% of customers, and increased the
financial cost by 1.9%.
The relative value of the discount rates can have a much stronger effect. A 5%
discount rate for grid extension and a 10% discount rate for microgrids leads to 87%
of the customers assigned to a grid extension, although no parameters other than the
discount rate were changed relative to the all-equal runs. It is very likely that the
discount rate exhibits a similar characteristic to grid reliability and diesel price, in
that its effect has a sharp transition between two modes, and the values tested tend
to be near this transition, creating large differences between otherwise similar runs.
79
Number of Management Management Management Management
Customers Cost Cost Cost Cost
per year per year per year per year
Case1 Case2 Case3 Case4
1 $15 $15 $11 $10
100 $11 $12 $10 $10
Infinite $7 $10 $9 J $10
Table 4.1: Sample inputs for defining economies of scale in annual management costs
Management Costs REM contains some parameters for defining the economy of
scale in the management and collections costs for microgrids. Since the strength of
economies of scale is important for the clustering algorithm, we might expect this
value to have a strong impact on the outcome. In order to define this curve, REM
fits a curve to data such as that provided in Table 4.1; everything but the definition
of 'infinite' for a large microgrid is changeable by the user. The table shows the input
values used for the range of cases tested, which were chosen to examine the impact
of economy of scale.
From case 1 to case 4, the average microgrid size shrank by 25.6%, indicating that
these management costs do have some impact on clustering decisions. The results are
shown in Figure 4-6. However, despite these changes, the assignment of customers
to different modes of electrification was not strongly affected by differences in man-
agement economies of scale. There was no clear trend within the four cases, which
varied between each other by less than 1% of the total customers. The total nonserved
demand and the total financial costs likewise showed no clear trends.
Cost of Nonserved Energy (CNSE) The cost of nonserved energy has a very
strong effect on the model outcome. It determines the preference for reliability vs.
cost. Table 4.2 shows the change in total financial cost, cost of nonserved energy
incurred, and customer assignment to the three modes of electrification under different
CNSE assumptions. The effect exhibits a sharp 'tipping point' effect, much like diesel
cost and grid reliability. This tipping point occurs because most consumers and most
80
0 1 2 3 4
Case Number
Figure 4-6: Decrease in microgrid size with increase in management economies of
scale.
clusters of consumers are relatively similar, so if one option is more cost effective for
one cluster, it is likely that many others (anywhere from 80 to 99% of customers) will
have the same result.
A low CNSE for critical demand biases the results towards grid extension in this
case, as the grid has a 75% reliability in these runs and generally a lower financial
cost per person. A high CNSE biases the results for low demand biases away from grid
extension. Because the microgrids designed in these scenarios are better at meeting
critical demand than a grid with 75% reliability (which in actuality is much better
than what we have observed for reliability during peak hours), when emphasis is
placed on meeting this demand microgrids are preferred.
Operations and Maintenance Costs for Generation To test the impact of
operations and maintenance (O&M) costs for generation, we run three scenarios in
addition to the base case. In the base case, the cost of labor is $0.93/hr, a figure
roughly derived from the monthly salaries we have heard from the microgrid company
Tara Urja. O&M costs for generation are divided into two components; a fixed
component and a component that varies by the size of the generation. The former is
given as a fraction of the initial cost, and the latter is given as a number of service
81
Management Cost Economy of Scale
Affects Microgrid Size
700
600
500
400
300
200
100
0
I Base case High CNSE Low CNSE
Critical CNSE 2 2 1
Normal CNSE 0.51 1.5 0.51
% customers with microgrids 89.23% 99.94% 1.45%
% customers with grid extension 10.75% 0.030% 98.53%
% customers with isolated system 0.018% 0.028% 0.016%
Annualized financial cost $34.88 $38.70 $18.51
($ million)
Annual cost of nonserved energy $3.59 $0.49 $16.75
($ million)
Average microgrid size 647 472 16
(customers)
Average $/kWh, $0.590 $0.587 $0.638
microgrid
Average $/kWh, $0.324 $0.378 $0.285
grid extension
Average $/kWh, $2.197 $1.179 $2.081
isolated system
Table 4.2: Assumptions about the cost of nonserved energy have significant effects on
several key outcomes of REM. A few illustrative cases are shown in this table; a base
case, a case where the 'normal' CNSE has been raised, and a case where the 'critical'
CNSE has been lowered.
82
hours per year'. For the fixed maintenance requirements, the battery is assumed to
require 5 hours of maintenance, solar panels 5 hours, diesel generators 25 hours, and
the AC/DC converter and charge controller 2 hours per year. The three variants on
this base case are as follows:
1. Reduce cost per hour of labor to $0.50/hr
2. Reduce required hours of labor (fixed cost) by half for all components.
3. Raise cost per hour of labor to $2/hr.
The results of this comparison are more or less inconclusive. They do not suggest
that O&M costs have a large impact on the electrification mode chosen, at least not
within the range tested. Within the range of labor costs tested (0.50, 0.93, and 2
$/hr) there is no strong trend. Quadrupling the labor cost from $0.5 to $2/hr results
in a 0.5% increase in costs per customer for microgrid systems, and a 1.3% increase in
the total cost of the electrification plan for the whole region. Only 0.08% of customers
are assigned to a different electrification mode between the high and the low scenario.
The comparison of the reduction in fixed labor costs with the base case is difficult to
do, however, because the base case seems to be a dramatic outlier in this set of four
cases. It is possible that this difference is due to the stochastic variation discussed
above. Indeed, the differences are consistent with the variations seen in the test of
inter-run variability - most parameters change a modest amount, but the average size
of the battery in a microgrid changes from 45 kWh in the base case to 5.0 kWh in
the reduced labor hours scenario. In scenarios 1 and 3, average battery sizes are also
significantly smaller, at 4.6 and 4.9 kWh per microgrid. Microgrid size is not a good
explanation for this change, as the average size between the base case and scenario 2
differs only by 6.05%.
While these results are somewhat puzzling and support further investigation, they
do suggest convincingly that generation O&M costs do not have a dramatic effect on
'Future versions of REM will incorporate some economy of scale into these fixed costs, recognizing
that the time to service a generator and a PV panel at the same microgrid takes less time than
servicing an equivalent generator and PV panel at different sites.
83
Table 4.3: Description of cases run for demand level sensitivity
REM's outcome.
Demand Level To test the importance of demand level on the output of REM,
we ran two cases with lower demand than the base case, and one case with demand
higher than the base case. For a full description of the demand inputs, see Appendix
A. The three cases are described in Table 4.3.
At the lowest level of demand, no customers are connected to the grid. This conclusion
is striking, as the demand levels used (two lights, with a possibility of a fan and a
small amount of TV usage) would be very typical for below-poverty-line households in
India, though even then the TV is likely out of reach. At the lower and base levels of
demand, approximately 90% of the customers are assigned to microgrids; 79% for the
higher demand case. The level of demand does affect the levelized cost of energy and
the generation chosen for microgrids, but not in a monotonic way. Figure 4-7 shows
these impacts; isolated systems unequivocally become cheaper with higher demand,
but for microgrids and grid extension it seems costs rise with increasing demand,
before hitting some sort of economy of scale and falling again.
84
Case Description Peak demand / Average daily
household (kW) demand /
household
(kWh)
Lowest Only two lights are possible, 0.153 0.255
and the option to use a TV
at night is eliminated.
Lower Four lights are possible, and 0.183 0.405
nighttime TV is eliminated.
Base 2 lights are considered 0.266 0.623
critical, with an additional 4
noncritical. Fan, TV also
included. (See Appendix A)
Higher Same as base case, but with 0.266 0.795
more customers (75%)
assigned a fan.
$/kWh and Demand Level
$3.249 *Qub
$2.080 $2.197 k hIOMAdsien
$1.574
$0.5 $05 $0.5 $0.52
$$0 2 
$0 7
Lowest demand Low demand Base demand Higher demand
(a)
Average Battery Capacity per Customer
0.08
0.07
0.06
0.05 -
0.04
0.03
3 0.02
10.01
0
Lowest Low demand Base demand Higher
demand demand
Average Generator Capacity per Customer
0.08
0.07
-0.06I0.05
0.04
0.03
z0.02
i 0.01
Lowest Low demand Base demand Higher
demand demand
(b)
Average PV Capacity per Customer
0.022
0.0215
0.021
0.0205
0.02
0.0195
(c)
Lowest Low demand Base demand Higher
demand demand
(d)
Figure 4-7: Impacts of demand on the cost per kWh and the generation options chosen
for microgrids.
85
Network Opex to Capex ratios
500-
400-
0300- r
o m grid
200 - microgrids
E
100-
0-
0.0 0.1 0.2 0.3 0.4
Ratio of opex to capex
Figure 4-8: Histogram of average network operations and maintenance costs as a
fraction of upfront installation, for each cluster
4.1.2 Other Model Result Observations
Bimodal Network Opex Costs As shown in Figure 4-8, the cost of operations and
maintenance (opex) for the network has a bimodal distribution. For many systems,
the ratio of the average annual network opex cost to the upfront cost for installing the
network (capex) is well below 30%, but for many systems, all microgrids, the cost is
almost 40%. These high opex costs are found generally in smaller microgrids; ratios
less than 20% are found only for microgrids that are more than 1000 households.
The reason for this bi-modality is unclear, and has to do with the way in which
the Reference Network Model (described in Section 2.3.2) calculates operations and
maintenance.
Cost per kWh The literature has several estimates of what the cost per kWh for
a solar or diesel-powered microgrid should be. [14] reports a total LCOE of $0.446
to $0.569 per kWh, depending on the generation technology used, with higher values
for diesel-based grids, assuming a microgrid size of 270 (the average village size in
Gujarat). [13] estimates LCOEs for solar-and-battery microgrids ranging from under
86
$1/kWh to as high as $23/kWh for very small systems, though for scenarios where
peak demand was at least 45W per customer, the range was under $1 to roughly
$3.5/kWh (See Figure 27 in [13]). In the runs examined, every customer is assumed
to have at least 2 lightbulbs of 15W each which are considered critical load, and 4
additional optional lights. So these runs would be assumed to be in the lower range
of [13]'s estimates.
The range of costs predicted by REM appears to be in line with what we would
expect from [14] and [13]. Across the range of runs done for this sensitivity analysis,
the average LCOE for microgrids was $0.595/kWh, only slightly higher than what
[14] predicts and well within the range expected by [13]. Isolated systems tend to
have an LCOE around $2/kWh. Extension of the grid, meanwhile, clocks in with an
average LCOE of $0.29/kWh. The range of these values is illustrated in Figure 4-9a.
Fraction of Demand Served The fraction of demand served in microgrids appears
to be very closely related to the number of customers in a microgrid and the size of
the diesel generator. Without any diesel, microgrids in the base case tend to achieve
a reliability of around 40%. After the microgrids grow larger than about 25 people,
diesel generators begin to be economical and reliability quickly improves. While the
exact numbers vary between cases, the behavior (with a low base reliability, increasing
to high levels once some minimum economy of scale is achieved) is relatively similar.
See Figure 4-10 regarding the relationship between the fraction of demand served
and the number of customers in a microgrid, and the relationship between number
of customers and generator size. In Figure 4-10a there are microgrids with similar
number of customers that have significantly different levels of fraction of demand
served; in particular, the fraction of demand served seems to increase rapidly, save
for a few outliers. It is not clear why REM found these outlier solutions optimal, but
these systems seem to have significantly smaller diesel generators than comparably
87
Distribution of LCOE
15-
10~
5-
0-
0.5 1.0 1.5 2.0
Average LCOE ($/kWh)
grid
isolated
microgrids
2.5
(a) Distribution of LCOE across all the runs of REM conducted for the sen-
sitivity analysis
Grid Extension Size and LCOE ($/kWh) Microgrid Size and LCOE ($/kWh)
0.4 0.7
0.35 0.65
0.2S 0.6
0.2 IM 0.55
0.15 0.5 0
0.1 040.05 0.45
0 0.4
0 500 1000 1500 2000 2500 3000 0 200 400 600 800 1000 1200 1400
Households in Grid Extension Households In Microgrid
(b) The average size of a grid extension or microgrid does not have a strong effect on the LCOE,
although there is a small correlation for microgrids.
Figure 4-9
88
Fraction of Demand Served In Small Microgrids Fraction of Demand Served In All Microgrids
J00 % 0000 200 0 0 O 5000 1 00 0
Nmr0 0 C Or
w6 0 oE - 0 E 0
90
C5
LL 0I LJL~
0 100 200 300 400 0 5000 10000 15000
Number of Customers Number of Customers
(a) The fraction of demand served changes rapidly as microgrid size increases, until it hits a saturation point
E Generator capacity per customer
4.2 cenaio Rsu.t
0
0; I I I I I
S 0 2000 4000 6000 8000 10000
Number of Customers
(b) The installed capacity of diesel generation per cus-
tomer increases rapidly at first, and then saturates and
declines slightly.
Figure 4-10: Microgrid reliability is closely related to the number of customers and
the size of the diesel generator.
sized systems.
4.2 Scenario Results
Recall from Section 3 that the three main scenarios in question are:
1. Mixed approach to electrification, utilizing grid compatible microgrids
(a) Low grid reliability (75%)
(b) High grid reliability (95%)
2. Mixed approach to electrification, utilizing non-grid-compatible microgrids
3. All grid extension.
89
Figure 4-11: Map of the output from the Scenario la. Grid extension network lines
are in red, microgrids in blue. Green circles represent microgrid generation sites;
those with a red border are connected to a medium voltage network. Thick network
lines represent medium voltage, thin lines represent low voltage. Small orange dots
represent isolated systems. The black lines represent existing medium voltage grid
lines
90
REM's default approach is in line with scenario 1. For scenario 1, we examine a
situation from each side of the grid reliability threshold: one with 75% grid reliability
and more microgrids, one with 95% grid reliability and fewer microgrids. Note that
75% grid reliability is still higher than what we have observed in agricultural feeders
in Vaishali, based on historical log books. The map of the low-reliability scenario is
shown in Figure 4-11. It is interesting to note that in the dense urban areas in the
lower left, microgrids are preferred, while grid extension is preferred in the less-dense
areas on the right side.
To simulate scenario 2, we wished to replicate the cheap, DC microgrids used by
companies like Mera Gao, which supply a few hours of evening light to their customers.
However, attempts to adjust REM's parameters to simulate a DC system within the
AC architecture of the Reference Network Model used to design the network were
unsuccessful.
Instead, cheap microgrids were simulated by lowering the cost of nonserved energy
for low-priority and high-priority demand to $0.4 and $0.8, respectively, such that
low-priority demand would rarely be met by a microgrid, but high-priority demand
(2 light bulbs at night) would often be. As well, the discount rate was raised and the
lifetime was lowered for these systems, to replicate the high risk environment of these
projects. These settings resulted in many microgrids that met only a small fraction
of load. However, the efficiency of these microgrids is likely still vastly overestimated
in this scenario, since some of the microgrids are much larger than would be expected
from a DC grid.
For scenario 3, the generation costs available were raised to very high levels. Solar
panels were priced at $10000 (instead of $225 and $50 for the different sizes). Diesel
cost was raised to $2/liter. Even under these high cost scenarios, 2 microgrids and 3
isolated systems powered by solar PV and a battery were found to be cost-effective,
but the rest of the consumers were placed on grid extensions.
For scenarios la, 1b, and 2, each run was done twice; once assuming that the mi-
91
crogrids are managed by private entrepreneurs with costs of capital and management
costs the same as the base case, and again assuming that the microgrids are managed
by the utility with the same cost of capital and no economy of scale in management
costs. The effect of the change was negligible, with the relevant output variables
changing by less than 2%. As a result, this section presents those figures derived
from the scenarios run with the assumption that microgrids are owned by private
companies, for clarity.
The high-level results of these runs are laid out in Table 4.4.
For the all-grid scenario, the government could reduce the subsidy required by 45%
if it were willing to raise the tariff only for customers above the poverty line to 340
rupees per month, which is twice the current rate. Below-the-poverty-line customers
make up 35% of the customers in rural Bihar, and arguably their rates should be
kept low. However, adjusting the tariff for those above the poverty line to be more
reflective of the cost of service would significantly reduce the subsidy needed.
A doubling in microgrid tariffs, from 100 Rs/mo to 200 Rs/mo, does not have an
enormous impact on the government subsidy required, even for the scenario 1(a) run
in which 88.5% of customers are on microgrids. This indicates how little of the cost
of the microgrid customers might pay. Indeed, few if any of the microgrids in this
scenario are able to cover the ongoing costs O&M and equipment replacement costs
of the microgrid, ignoring the upfront investment costs.
It is worth noting that for scenario la the number of households per microgrid has
a bit of a bimodal distribution. There are many microgrids with fewer than 350
households, and many with 1300-1600 households, but none in between. Future work
should investigate the reason for this bimodality.
92
Metric Scenario la: Scenario 1b: Scenario 2: Scenario 3:
Compatible Compatible Cheap All-Grid
microgrids, microgrids, Microgrids Low grid
Low grid High grid Low grid reliability
reliability reliability reliability
Annualized financial $35.64 $20.64 $19.00 $16.98
cost
($ million / yr)
Total cost including $39.46 $25.41 $32.42 $39.91
nonserved energy ($
million/ yr)
Annualized financial $77.47 $44.86 $41.3 $36.9
cost per customer per
year
Customer assignment, 88.51% 0.62% 1.57% 0.00217%
microgrid (10 houses)
Customer assignment, 11.47% 99.37% 98.40% 99.99%
grid extension
Customer assignment, 0.025% 0.013% 0.022% 0.0007%
isolated systems (3 houses)
Average microgrid size 517 9.3 1247 5 customers
(customers)
Average $/kWh, $0.630 / kWh $0.669 / kWh $0.737 / kWh $9.132 / kWh
microgrid
Average $/kWh, $0.314 / kWh $0.278 / kWh $0.294 / kWh $0.383 / kWh
grid extension
Average $/kWh, $2.077 / kWh $2.079 / kWh $2.080 / kWh $18.62 / kWh
isolated system
Percent of demand met 97.44% 37.19% 38.48% 38.48%
by off-grid systems
Percent of demand met 94.83% 94.63% 74.41% 75%
Subsidy cost to $453.0 - $504.4 $130.7 - $131.0 $256.1 - $257.0 $56.20
government, NPV. ($
million)
Subsidy cost, amortized $22.88 - $25.47 $6.6 - $6.62 $12.93 - $12.98 $2.84
over 20 yrs ($ million)
Table 4.4: Results from the two runs examined in Scenarios 1-3. The annualized financial cost
represents the sum of the annual costs incurred by the microgrid, and the annuity of the capital
costs. The annuity represents the value that could be paid yearly so that the net present value
of those payments is the same as the actual costs incurred; this value is a way of distributing the
infrequent capital costs over the lifetime of the project. The range in subsidy projection is based on
a range of microgrid 'tariffs from 100 Rs/mo to 200 Rs/mo, and a project lifetime of 20 years. The
subsidy calculation assumes a lifetime of 20 years and a discount rate of 9.5%.
93
4.3 Comparison
4.3.1 Savings from a mixed on/off-grid approach
The results from the scenarios above show a modest savings available through the use
of a mixed approach over an all-grid plan. These savings would likely be considerably
higher if the cost of upstream reinforcements in the transmission and distribution
infrastructure were considered, as this would drive up the costs of grid extension
significantly. The magnitude of the additional load attached to the grid would neces-
sitate considerable grids in in Bihar's already-overburdened infrastructure.
The use of solar/battery powered microgrids even when panel prices are $10000/panel
in scenario 4 clearly indicates that for the furthest consumers, off-grid systems can
be eminently cost-competitive with the extension of the grid. The difference between
the all-grid scenario (3) and the low-cost microgrid scenario (1(a)) shifts just 1.57%
of customers off of the grid and onto relatively low-quality microgrids. However, this
scenario could enable societal cost savings of at least $7 million per year; this number
is certainly an underestimate, as our cost of nonserved energy estimates are likely to
be underestimated, and the reliability of the grid in real-life can be much poorer than
75% in many rural areas - and likely would be so in other rural areas that have yet
to be electrified. This difference is illustrated in Figure 4-12a.
When the costs of not supplying electricity to consumers and businesses are factored
in through the cost of non-served energy, it is clear that solutions involving more
microgrids are vastly more favorable. The poor reliability of the grid in India results
in huge losses for consumers who are connected to the grid but receive only very poor
service. Whether these costs are sufficient to motivate additional investment in and
subsidy for microgrids depends on the resources and perspective of the planner. The
total costs are almost the same between scenario 3 (all grid) and scenario la (grid-
compatible microgrids); however in the former there are very high costs of nonserved
energy, while in the latter there are very high microgrid costs. Figure 4-12b shows the
94
high level of government subsidy that would be required to sustain those microgrids
and make them viable. These figures would be even higher if the microgrid is require
to charge a tariff consistent with on-grid rates; Uttar Pradesh, for example, has
suggested a tariff of Rs 60/mo, half the level that is used in these calculations.
Looking ahead, if the government is able to raise grid tariffs for above-poverty-line
(APL) households, it would in fact have a considerable impact on the subsidy required.
A doubling of the APL tariff, considering that 65% of rural households in Bihar are
APL, could reduce the total subsidy required for the grid by 45%.
These numbers make a strong case for microgrids as a way to reduce the societal
burden of non-served electricity demand, though not as effective way to reduce the
cost borne by the government. It is very likely that the subsidy costs borne by
the government would in reality be somewhat lower, as these simulations do not
account for any non-residential load; commercial load would likely use more power
and likely could pay a higher price, making the microgrids more viable. (Though it is
important to note that abuse of these cross-subsidies from commercial to residential
customers could drive commercial customers to seek their own generation source if
they are abused.) Even a small number of microgrids, as in scenario lb and scenario
2 increases the subsidy cost borne by the government considerably over scenario 3.
4.3.2 Value of grid-compatible microgrids
As discussed above, scenario 2 is a very rough approximation of a solution involv-
ing cheap DC microgrids, which makes a comparison between non-grid-compatible
microgrids and grid-compatible microgrids difficult. A more useful simulation of the
non-compatible approach would contain less grid extension and more and smaller mi-
crogrids, and could certainly be produced in future research. Although the cost per
customer for a microgrid should be much lower in scenario 2 than the other scenarios,
it is not significantly lower than the costs per customer seen in, any other scenarios
where a small fraction of customers are connected to microgrids. However, based
95
Source of costs in different scenarios
$45,000,000
$40,000,000
$35,000,000
$30,000,000
$25,000,000
$20,000,000
$15,000,000
$10,000,000
$5,000,000
Scenario 1, High Scenario 1, Low Scenario 2 Scenario 3
reliability reliability
E GE Financial Cost N MG Financial Cost a IS Financial Cost U Nonserved Energy Cost
(a) Breakdown of scenario costs between actual financial costs for the different
system types (MG = microgrid, GE = grid extension, IS = isolated system),
and the costs of nonserved demand. These costs are the yearly annuity that
covers upfront and intermittent capital expenditures in addition to ongoing
costs. The scenarios referenced are described in section 4.2 and 3.
$40.00
$35.00
02$30.00
! $25.00
$20.00
$10.00
Scenario 1, high Scenario 1, low grid Scenario 2 Scenario 3
grid reliability reliability
0 Annual financial cost W Annual subsidy required
(b) Estimation of the annual value of subsidy costs that would need to be paid
by the government, over a lifetime of 20 years, as compared to the annuity
cost of the microgrid. The scenarios referenced are described in section 4.2
and 3.
Figure 4-12
96
on the limited data available, we can draw some first-order conclusions about what
grid compatible microgrids and their accompanying regulation need to accomplish in
order to make them a good investment.
An attractive theory for grid compatible microgrids is the idea that these businesses
will bear the costs of the network and generation, covering them with tariffs and
perhaps some subsidies, and pave the way for the future expansion of the grid; once
the grid arrives, the theory goes, its costs will be greatly reduced because it will not
need to build a significant amount of network infrastructure. These results question
that theory, given the tremendous cost of building the microgrids in the first place;
it is almost double the cost of extending the grid initially. Of course, some of those
costs are generation, which would reduce the operating costs of the utility. However,
it is unlikely that diesel generators would be economical to run in the presence of
grid power, and diesel makes up a sizable fraction of the generation capacity in the
scenarios examined. As the majority of the microgrids built in the scenarios showing
a high penetration of microgrids have a significant amount of their generation from
diesel, the contribution of that generation to reducing the operating costs of the utility
might be limited. This is certainly a question for further research.
A comparison of scenario 3 with scenario la shows that the financial costs of provid-
ing electricity service to most consumers through high-quality microgrids are more
than double that of providing electricity service through the extension of the grid
alone. Accounting for the costs of nonserved energy, however, reveals that the two
scenarios are almost the same cost, with the grid extension scenario being slightly
more expensive. Which one is preferable in this situation is dependent on the values
of the planner and how much weight they place on the cost of nonserved energy.
Considering that the costs of nonserved energy are likely an underestimate and the re-
liability of the grid is likely an overestimate, these financial costs should be considered
strongly as in reality they are probably much higher than what is simulated here. An
important unaddressed aspect is the cost of reinforcing the existing network as it is
extended to more customers. Doing so would be incredibly necessary as line capacity
97
limits are already constraining in some rural areas. Accounting for this cost would
raise the cost of scenario 3 significantly. An additional consideration is the impact
of distributed generation on the transmission and distribution costs incurred when
the microgrids are connected to the grid. It is possible that local generation could
reduce the transmission capacity needed, resulting in lower costs overall. However,
the cumulative impact of local generation on transmission and distribution expenses
is highly dependent on the actual situation present, the timing of generation and
demand, and the intermittency of supply. Future studies should certainly take this
into account.
4.3.3 Need for ongoing subsidy of grid-compatible microgrids
The results of these scenarios point towards the need for ongoing subsidy to support
grid compatible microgrids. The average annual cash-flow shortfall for microgrids in
scenario la was $4,503 per customer per year, indicating that a subsidy of a similar
magnitude would be needed to make the project break even. This figure assumes
a tariff of 120 rupees per month per customer, and excludes upfront costs. Figure
4-13 shows the close relationship between the number of customers in a microgrid
and the average subsidy need per customer. The economies of scale start to saturate
around 100 to 200 households in these scenarios, indicating that governments should
encourage microgrids that will be at least this big for the most efficiency.
98
Subsidy Costs Per Customer
C) 0 0O00 0
E
.200
4o000 I~
.1000
C:
9-12000
<.140000
0 50 100 150 200 250 300 350 400
Number of customers per microgrid
Figure 4-13: Average annual ongoing subsidy needs (as expressed in cash-flow short-
fall) per customer in scenario la, for microgrids below 400 customers each. Note that
there are a significant number of microgrids with more than 1300 households, but the
same trend continues for these larger microgrids; subsidy costs decrease with number
of households. There are no microgrids with 400-1300 houesholds
99
100
Chapter 5
Regulatory Assessment
5.1 Regulatory requirements
Tenenbaum et al (2014) makes the astute observation that many countries have a
"two-pronged approach" to electrification: grid extension for most people and
microgrids for remote areas, but for this approach to work countries must have a set
of regulations for when the two meet [541. Bad regulation that leaves microgrid
investors feeling that they may be treated unfairly when the central grid arrives is
just as bad as no regulation at all, if not worse. India has started to develop
regulations governing off-grid, grid-compatible microgrids, and how these businesses
should be dealt with and regulated once the grid arrives. This chapter reviews those
emerging regulations and assesses how they might be improved to ensure their
success.
Cost is an important dimension of grid-compatible microgrids, and regulatory
structure is a key driver of these costs. Regulation must ensure the reliable and
equitable supply of electricity while allowing investors to recover their costs and
minimizing risk [60]. Regulation for off-grid systems must also be light-weight and
quick. If it involves a lot of paperwork, red tape, and delay it can act as a
significant barrier to entry for off-grid operators. Since these businesses often
101
operate with slim profit margins, regulation that is time-consuming and costly to
comply with could make an off-grid electricity business unprofitable [54j. As well,
light-weight regulation can reduce the burden on state electricity regulators, who
are often already underfunded for the work they have to do [30]. The key is to have
just enough regulation to ensure the fair and equitable provision of electricity, and
no more: not an easy balance to strike. In order to strike this balance, regulators
should seek to minimize the number of times the business is required to interact
with the regulator and the amount of information that must be provided, and they
should standardize and publicize documents and processes.
Before determining how to implement regulation for off-grid systems, regulators
must determine the broad frameworks and requirements. There are several
proposals in the literature for what these frameworks might be. One example comes
from Tenenbaum et al (2014), which notes four main options for the relationship
between grid-connected microgrid and utility, and suggests that all should be
available options for a microgrid developer [541.
1. "Small power distributor," (SPD) in which the microgrid effectively becomes a
small distribution company, buys electricity wholesale (or from the distribution
company) and sells at retail to customers. The generator is no longer used.
This model has been used in Nepal, Bangladesh, Vietnam and Cambodia.
2. "Small power producer," (SPP) in which the microgrid sells electricity to the
national grid but does not sell to customers, and may receive a feed-in-tariff.
The utility can take over network operations. This option is most similar to
what has been proposed as a model for Indian grid-compatible microgrids by
ABPS Infrastructure and the Shakti Foundation, and endorsed by the Indian
Forum of Regulators [1].
3. Combined option: in which the microgrid buys wholesale electricity and also
maintains generation assets that it may use to sell power back to the main
grid.
4. Buyout: the central utility buys out the microgrid owner at whatever is
102
determined to be a fair value. Book value (the depreciated value of the
physical assets) is a common way to determine this value.
The first and third of these options are most similar to the existing franchisee
structures in India, described below. [211 also provides a useful taxonomy of the
possible ways in which an existing microgrid and an encroaching central grid could
interact, shown in Figure 5-1, which encapsulates and extends the options offered by
Tenenbaum with intuitive graphics.
5.2 Current status of regulation in India
The Electricity Act of 2003, and its subsequent amendments, remains the most
prominent piece of legislation concerning the electricity industry in India, as
discussed in Section 4.2. The Act gives distribution companies (discoms) a
'universal service obligation,' compelling them to provide service to everyone living
in their service area. However, discoms have been unable to meet this obligation due
to their generally poor financial situation, and due to its in-feasibility, as discussed
in Section 1.1.3.2 regulators do not enforce it. The Indian central government
announced the UDAY (Ujwal DISCOM Assurance Yojana) scheme in early 2016,
which is aimed at improving the failing finances of the state distribution companies.
The effect of this policy may enable distribution companies to meet their obligations
more effectively but it is too early to tell. (see Section 1.1.3 for more details.)
5.2.1 Status of off-grid regulation
Meanwhile, policies have enabled other electricity businesses to grow in India in
several ways. The Electricity Act of 2003 allows the distribution of electricity in
rural parts of India without a license. This provision is how many small microgrids
103
Customers
Mini-Grid
Small Power Producer
Option 2 (SPD) : local distribution mini-grid system continues
to operate, obtains bulk electricity supply from national utility
Customers
National
Grid
tamge Plants
t
Customers
Mini-Grid
Sm wer ucer
Option 4 (SPD with backup): local distrIbution system
continues to operate, but with backup electricity provided by
old generator
Customers Customers
Mni-Grid
National
Grid
Small Power ProducerLarg PlatsjAresftybadWPfar Woad ak*4
Option 1(SPP): renewable energy generator interconnects
with main grid, becoming another power plant on the grid.
Utility takes over electricity distribution to retail customers
Customers
Extension
of a Mini-Grid
National
Large Plants Small Power Producer
Option 3 (SPP + SPD): renewable energy generator
interconnects with the main grid, becoming another power
plant on the grid, Continues to sell electricity to retail
customers, some of which may be purchased from utility.
Customers Customers
MiniwGrid
National
Grd
Large Plants s Small Power Producer
Option 5 (No SPP or SPD): renewable energy generator and
mini-grid are abandoned; customers in the area switch over
to the main grid
Customers Customers
National
Grid
Gr d
Larg Plants Sm~a r 'rucer
Key: = power from SPP = power from utility = meter
Figure S1. Options for interconnection of mini-grid with main grid.
Figure 5-1: This figure taken from [21] illustrates the complete range of options for
the interaction of a central grid and a minigrid
104
Option 0: Before the mini-grid connects to the grid
Customers
National
Grid
Large Plants
operate in India. In justifying the lack of license or tariff regulation, the
Government of India cited the idea that a low cost of entry could provide sufficient
threat of competition to keep tariffs reasonable for consumer [54]. However, it is
highly unlikely that a competitor would consider building another microgrid in a
location where one already exists; the barrier to entry is already quite high due the
the magnitude of the initial investment requirements. Thus, this justification for the
lack of tariff regulation is not a very good one.
Currently, microgrid operators have no recourse when the central grid arrives to
their area of service; they will most likely go out of business as they lose customers
to the heavily-subsidized central utility service. Recent amendments to the
Electricity Act of 2003, discussed below, require state regulators to provide a more
formal path for isolated microgrids to interact and connect with the central grid.
5.2.2 Existing franchisee regulation
India does have regulations allowing for licensed franchisees that are somewhat
similar to the SPD model described above, though they are meant for on-grid
businesses. [37] estimates that in 2012 there were 37,000 franchisees covering
216,000 villages in India, however recent conversations with stakeholders in the
sector have suggested that this model has not proved to be as attractive or popular
as hoped, and it has faced opposition from states and state-owned distribution
companies who do not want their business infringed upon by franchisees '. This
animosity is important to keep in mind when assessing proposed new regulations.
The Ministry of Power, through the RGGVY program (now subsumed under
DDUGJY), has specified six models for a more formal franchisee setup [1, 31:
* Collection-Based Revenue franchisee: The franchisee is responsible for billing,
collections, and facilitating connection of new consumers. They are given a
'Meeting with Akanksha Chaurey, July 2015; Meeting with Sanjay Prakash Jan 2015; Meeting
with Arnol Gupta (World Bank) July 2015
105
share of the fees collected, and are given a revenue collection target each
month.
" Input-Based Revenue franchisee: The franchisee is responsible for revenue
collections as in the collection based franchisee, but the target for collection is
based on the amount of electricity that the utility measures going into the
franchisee's area. This setup creates incentives for loss reduction.
" Input Based franchisee: This franchisee buys electricity from the utility at a
predetermined rate, and must cover those costs with the revenue it collects.
This creates incentives for loss reduction and efficiency.
" Operation and Maintenance Franchisee: This franchisee has all the rights and
responsibilities of the Input Based franchisee, with the added obligation of
conducting operations and maintenance work on the medium and low-voltage
lines. In return they are able to purchase electricity at a cheaper rate.
" Rural Electric Cooperative Society: In this arrangement, the community owns,
operates, and manages the infrastructure and collects payments, buying power
from the utility.
" Electric Cooperative Society: This arrangement is similar to the above, but
the community may decide to contract out the operation of the grid to a
private company.
With regards to the cooperative arrangements (the last two bullets), [19] states "It
is unclear if these franchise concepts were ever enacted. It does not appear that any
[Ministry of Power] supported ownership co-ops were initiated in recent years."
5.2.3 Recent developments in off-grid policy
India itself recently enacted new regulations in an amendment to the Electricity Act
that directed state electricity regulators to create regulations for microgrids to
connect to the grid and receive payment from the central utility [201.
The recent tariff policy amendment, passed Jan 28, 2016, acknowledges the risk to
106
investors of the grid reaching the microgrid, and specifies that the State Electricity
Regulatory Commissions (SERC) create regulations addressing the following:
"Micro-grids supplying renewable energy are being set up in such
areas where the grid has not reached or where adequate power is not
available in the grid. Investment invblved in setting up of such
microgrids is substantial. One of the risks of investment is grid reaching
the area before the completion of the project life and thereby making
power from micro grids costly and unviable. In order to mitigate such
risk and incentivize investment in microgrids, there is a need to put in
place an appropriate regulatory framework to mandate compulsory
purchase of power into the grid from such micro grids at a tariff to be
determined under section 62 of the Act considering depreciated cost of
investments and keeping in view industry benchmark and with a cap if
necessary, as approved by the Appropriate Commission. The
Appropriate Commission shall notify necessary regulations in this regard
within six months." 2
The details of this regulation have yet to be worked out by State Regulatory
Commissions; Uttar Pradesh was the first to release draft regulations, in early 2016.
Suggestions for this policy are found in ABPS's report "Final Report On Policy and
Regulatory Interventions to Support Community Level Off-Grid Projects," written
in conjunction with the Shakti foundation, and in the draft regulations promulgated
by the Indian Forum of Regulators, which build off of these suggestions [1, 21. These
suggestions are discussed in more depth below. The advantages and disadvantages
will be discussed after reviewing best practices from around the world.
2http://powermin.nic.in/upload/pdf/TariffPolicy-ResolutionDated_28012016.pdf Accessed
March 28, 2016
107
5.3 Lessons learned in other countries
There are several existing examples of similar regulation of independent small
generators and microgrids to draw on in considering a new regulatory structure.
Rwanda has proposed a system in which "private operators can own and operate
decentralized grids while selling the power to the state-owned utility that retails the
electricity using existing prepaid electricity systems and distribution channels" [601.
Several countries, including Cambodia, and Tanzania have implemented similar
"small power producer" (SPP) regulations [60, 541. Some of their experiences are
briefly detailed below, offering useful lessons in what to do and what to avoid.
One item that stands out in reviews of these programs is the idea that light-handed
or no tariff regulation can promote a lively rural micro-grid sector. Tenenbaum et
al. [541 notes, "most successful decentralized rural electrification schemes for isolated
mini-grids have involved little or no price regulation." This observation supports
India's decision to allow rural microgrids to operate without a license and charge
whatever tariffs their consumers will agree to. The arrangement certainly lowers the
barriers to entry by removing red tape. However, it is important to note that while
this lack of regulation may be beneficial for the market in the short term, in the
long term it may not encourage the most equitable or effective electricity provision.
Some tariff regulation and minimum quality standards is recommendable to ensure
that consumers are adequately protected, as a local microgrid does not experience
sufficient competition to prevent it from exerting monopolistic power.
5.3.1 Cambodia
In Cambodia, Small Power Producers (SPPs) sprang up without regulation,
determining tariffs through an agreement between the entrepreneur and the
community. Later, a national regulator came into existence, provided licenses to the
so-called 'rural electrification enterprises,' and regulated their tariffs. Cambodia has
108
no uniform national tariff; tariffs are generally higher in rural areas than urban
areas, for example. The regulators found they had to allow the SPPs to adjust the
tariff based on fluctuating diesel costs to allow them to remain profitable [541.
Many of Cambodia's isolated mini-grids converted to small power distributors when
the main grid arrived, selling their diesel generators and buying electricity wholesale
from the utility. They were required to get a new license from the regulator to
covert to a small power distributor; 82 had done so by 2013. Utilities benefit from
this arrangement because they do not have to deal with a proliferation of
low-demand customers, and the micro-grid owner benefits by being able to stay in
the business. The regulators ensure their financial viability by setting a sufficient
margin between the wholesale power tariff and the retail power tariff. The certainty
provided by this regulation encourages microgrid investors to invest in their
systems, and to ensure their facilities meet sufficient technical standards [541.
5.3.2 Tanzania
In Tanzania, Tanzania Electric Supply Company (TANESCO) owns and operates
both the main grids and some existing microgrids, and they allow independent
power producers (IPPs) to feed into both. The IPPs receive regulated feed in tariffs,
but these are based on avoided cost and are thus around three times higher for
microgrids than for the main grid [541. When the main grid connects to the
microgrid, the tariff will drop down to the main grid tariff. For small power
producers receiving a feed-in tariff the date of arrival of the main grid could add
significant uncertainty to the expected lifetime profitability. The low grid feed-in
tariffs, moreover, mean that only rather inexpensive generation, like hydro or
biomass, will be competitive for supplying the main grid [541.
Independently operated microgrids also exist in Tanzania. Small Power Producers
(<1 MW) do not need a license in Tanzania, just a registration with the regulator,
109
which does not require their approval. For producers < 100 kW, the regulator does
not even need to review the proposed retail tariffs, though they reserve the right to
review if there are significant consumer complaints [54].
5.3.3 Mali
Mali has successfully promoted a large number of isolated microgrids. They offer a
substantial capital subsidy for every consumer connected, but no operating
subsidies. In Mali, microgrids are allowed to charge tariffs substantially higher than
the main grid's, sometimes two to three times higher, though there is regulatory
oversight of tariff and revenue levels. The capital subsidy grant comes with
minimum service requirements as well as maximum tariff specifications. The
flexibility of this regime has its risks; Tenenbaum et al. (2014) notes that some
villages have reported that they were not adequately consulted before the
government granted the license to the microgrid operator; requiring an agreement
between the developer and the villagers may prevent this sense of
disenfranchisement [541.
In areas where the microgrids are close to the designated concession areas of the
national utility, this creates some tariff envy. In response to this envy, the regulator
ordered seven isolated microgrids located close to the main grid to be connected to
the grid in 2011 [54]. The authors of [541 note that at the time of writing Mali was
actively considering what rules might govern the handoff of the microgrid assets to
the central grid. 3
3
We have not been able to locate any evidence of these regulations.
110
5.3.4 Sri Lanka
In Sri Lanka, the government promoted the creation of 'Village Electricity
Consumer Societies,' which created small mini-hydro microgrids owned by the
community. When the central grid has expanded into the areas served by these,
microgrids, consumers have typically switched over, and the microgrid goes out of
business. Sri Lanka also has many small power producers, typically micro-hydro,
which sell wholesale to the national grid 1541.
5.4 Assessment of India's Proposed Regulation
The amendments to India's Electricity Act of 2003 require state regulators to
propose regulations for buying power generated by microgrids when they are
connected to the main grid. This description does not capture the whole story,
however, as the costs of the network and other non-generation components of the
microgrid are also significant investments requiring some compensation. These
aspects are addressed in the Forum of Regulators' model regulation in this area.
5.4.1 Forum of Regulators Recommendations
ABPS Report
The ABPS report which served as a basis for the Forum of Regulators' model
regulations recommends an "off-grid distributed generation based distribution
franchise model" (ODGBDF) over other methods including distribution franchisees,
relying on renewable energy certificates (RECs) for revenue, and generation-based
subsidies. This scheme is most similar to the Input-based Revenue Franchisee, as
described above, and like that model it would be approved under the RGGVY
guidelines.
111
Under this scheme, the microgrid developer will be a franchisee of the discom,
collect a regulated tariff from its customers, and receive a feed-in-tariff from the
distribution company to cover the viability gap. The feed in tariff would be
determined by state regulators, and the Forum of Regulators is to define guidelines
for how it should be calculated. The franchisee will receive the feed in tariff from
the discom, minus the revenue collected from customers. The report also makes a
passing mention that the state regulators could choose to include a franchisee fee
paid to the franchisee by the discom. As well the Government of India could
determine that it wishes to provide subsidies through the existing rural
electrification support programs (DDUGJY).
Once the microgrid is connected to the grid, the franchisee would purchase power
from the utility, but beyond this the recommendations are quite vague. It seems that
the feed-in-tariff might continue after grid connection, and the Central Electricity
Authority would need to provide guidelines for grid interconnection. The nature of
the relationship between the microgrid and the discom upon interconnection, however,
is left to state regulators to determine. The SERCs would specify the tariff at which
the franchise would purchase power from the utility, as well as most of the specifics
of the regulation governing the incorporation of the microgrid into the grid. The
SERC's decision regarding what price the franchisee would have to pay to purchase
power from the central grid (once it is connected to the grid, if it remains a franchisee)
is a significant choice. The margin between the power purchase price and the tariff
collected by the franchise is a key determinant of the viability of the microgrid once
it is connected to the grid.
Forum of Regulators Recommendations
The Indian Forum of Electricity Regulators has set forth a set of draft model
regulations which build off of the ABPS report (and were developed in conjunction
with ABPS) to guide state regulators in setting up proper regulations for the
112
connection of microgrids to the central grid [2]. The arrangement described in this
regulation follows the suggestions for the ODGBDF. Once the central grid reaches
the franchisee, the discom is to buy out the franchisee's network assets at book
value. The franchisee will continue to own the generation and will continue to sell
generated power to the grid at the pre-determined feed-in-tariff.
The model regulations set forth a framework in which a microgrid would operate as
the franchisee of a utility, with a license and an exclusive right to provide service
within their geographical area. The microgrid would be required to charge tariffs
that are no higher than the grid's tariffs, which would certainly be below the costs
of supply. To reimburse this viability gap, the costs of generation would be
subsidized through a Feed in Tariff agreed upon in a Power Purchase Agreement
with the utility and the rest through franchisee fees. Before receiving their license,
the microgrid would be required to obtain the consent of the local governing body,
after presenting them with a comprehensive plan.
The State Electricity Regulatory Commission would be required to determine the
amount of the Feed in Tariff for each renewable generation technology recognized by
the Ministry of New and Renewable Energy. The determination of the Feed in Tariff
would be through a cost-plus methodology, in which the tariff is set to cover the
expected costs of operation plus a rate of return on capital investments, and would
thus require only infrequent re-examination. Microgrid operators could request a
revision in this tariff if they felt it was unfair. The other costs of distribution and
management would be subsidized through a franchisee fee paid by the discom to the
microgrid operator, determined by mutual agreement of the discom and the
microgrid. The discom would also be expected to assist the microgrid in applying
for state and central subsidies as relevant.
Upon arrival of the central grid, the microgrid business would retain ownership of
the generation assets and would continue selling power to the discom at the feed in
tariff agreed on in the power purchase agreement. The rest of the assets of the
113
microgrid (the network etc.) would be purchased by the discom at their depreciated
book value.
This arrangement has many positive aspects. The requirement for local consent
avoids the disenfranchisement felt by some customers in Mali, as noted above, who
were upset when the government granted distribution monopolies without their
consent. This process also promotes village 'buy-in' to a microgrid, which is often
critical to a project's success by promoting adherence to tariff payments and
discouraging theft. Also noteworthy is the maintenance of the same feed in tariff for
reimbursing generation both on- and off-grid avoids the revenue risk faced by
microgrids in Tanzania, and provides long-term certainty regarding the value of
generation assets. However, there are areas requiring some careful examination and
improvement.
Recommendation 1: Require a grid expansion plan
The franchisee model has been somewhat unsuccessful in India partly due to the
animosity of the incumbent utilities, who are loath to see some of their service area
passed off to a competing business, as discussed above. Regulation should be
designed considering that the discoms may have an incentive to hinder the
development of off-grid franchisees. In this respect, there are a few worrying parts
to the Forum's proposed regulations.
The regulations provide that "Within one month of receipt of application from the
Rural System Operator, the Distribution Licensee shall process application and
confirm acceptance / rejection of the application. Provided that the Distribution
Licensee can reject application only on the grounds of violation of norms of design
of the system, or plan to extend electricity grid within two years," [2]. While this
aspect of the regulation does limit the discom's ability to deny microgrid's
franchisee applications out of malice, the provision for rejection in case of the
extension of the grid is a loophole that could be easily abused.
114
In almost all of the conversations that our team had with the distribution company
in Northern Bihar over the last two and a half years, a common refrain has been
that all of Bihar would be connected to the grid within two years. This claim
appears to be based on political promises, and those who work outside of the discom
will freely opine over the implausibility of this promise. The history of electrification
in India is littered with ambitious goals and unmet promises. If these common and
untrustworthy promises are sufficient grounds for rejecting a microgrid's application,
then the discom could have sufficient ground to deny a great deal of applications.
This escape clause for the distribution company only makes sense in the presence of
a realistic grid expansion plan, produced by the discom and approved by the
regulators. Such a plan does not exist in Bihar or in many other states.
The regulation should allow the discom to deny a microgrid's application only in the
presence of a grid expansion plan vetted by the regulators which indicates that the
area will be electrified in the next two years. This requirement would prevent abuse
of power by the discom. It would have the added benefit of providing a significant
amount of useful information to microgrid entrepreneurs who are trying to figure
out where the best areas to set up a business would be.
Recommendation 2: Provision of subsidy to microgrid franchisees by a
third party
The Forum's draft regulations specify two ways in which the viability gap will be
covered for microgrid franchisees: the feed in tariff and franchisee fees. Both of
these revenue streams will be paid by the distribution company. This arrangement
raises the issue both of the discom's ability to provide those payments and of their
willingness to do so. The distribution companies do not have enough revenue to
cover their own costs, let alone pay the expensive subsidies that will be required for
grid-compatible microgrids. As the discoms have an incentive to raise barriers to
entry to these potential competitors, they may find reasons for payments to be
115
delayed or underestimated. Investors may well feel uncomfortable if the viability of
their business depends on the benevolence and solvency of the distribution company,
which the current form of the regulations would suggest.
These subsidies should ultimately be covered by state or central government funds,
since the distribution companies are unlikely to be have the extra revenue to cover
them. In light of the worries about payment by the discom, it would be wise to have
these subsidy streams delivered in total or in substantial part by an entity other
than the distribution company. Possible candidates for this third entity are the
state government, central government, or a contracted private party that manages
government money. In the future, India may have developed more sophisticated
financial infrastructure which facilitates the delivery of subsidies directly to the
consumer. It could be undesirable to place the risk of subsidy provision on the
poorest consumers who can least bear that risk, especially if the subsidy remains
somewhat uncertain. However, there is reason to believe that the political clout of
all those customers would make the payment of such subsidies a priority.
Recommendation 3: Provide option to continue as grid-connected fran-
chisee
Under the proposed regulations, the off-grid franchisee has one choice when the grid
arrives - sell the network assets at book value, and retain the generation assets,
which will receive the feed in. tariff. This arrangement ignores the benefits that
could come from the discom leveraging the existing relationship with the community
that the off-grid franchisee has developed. They may be better positioned to
effectively collect tariffs and prevent theft than the discom, which has no personal
relationship with the customers.
Cambodia, for example, has effectively employed an arrangement in which existing
microgrids can become a franchisee of the distribution company, purchasing
116
electricity wholesale from the distribution company and reselling it to the local
customers [541. The off-grid franchisee should have the option of converting to what
might be called an Operations and Maintenance Franchisee as described in the
taxonomy in Section 5.2.2, while retaining ownership of the generation and receiving
a feed-in-tariff.
Recommendation 4: Allow islanded operation of microgrid with feed in
tariff
Although the generation assets of the microgrid are guaranteed the same feed in
tariff once they are connected to grid, they may still experience a significant drop in
revenue if the grid is unreliable. The generator cannot sell power on to the grid
while the grid is down, and for all of these hours they will lose revenue. One flawed
option for addressing this gap would be to force the discom to pay a penalty to the
generation owner when outages force the owner to lose revenue.
A more useful option would be to provide regulations allowing the islanded
operation of the microgrid even when the main grid is blacked out. This
arrangement would also offer more flexibility to the franchisee to provide service
even when the main grid is not providing power. Such an arrangement could result
in a much higher quality of supply in areas where the grid is unreliable. The discom
would still have to pay the generation owner the feed-in-tariff for their generation in
this time, lessening the loss imposed by the grid outage.
Of course, more infrastructure is required to operate in islanded mode than to
simply sell onto the grid. The proper controls must be in place to disconnect and
reconnect from the grid safely, and to govern power consumption during the
blackout. (These requirements are described in greater length in Section 1.2.)
Recommendation 5: Allow limited flexibility in tariffs For equity reasons,
the tariffs charged by microgrids should not be very different from those charged by the
117
discom. Allowing different tariffs on different systems would be sure to cause political
discontent, and could make the microgrid businesses' relationship with the community
somewhat tense. Situations where customers who are located close to each other, but
are connected to different electricity sources and are paying significantly different
rates, could create political discontent, as evidenced in Mali (discussed above). Any
attempt to pick a dividing line between lower and higher tariffs will result in some
perceived unfairness.
However, given that the grid's tariffs are known to be dramatically unsustainable
and should be revised, and given the known unreliability and difficult of obtaining
government subsidy in India, some flexibility may be beneficial. As long as subsidies
are viewed as not sufficient or sufficiently reliable to make up the viability gap between
the grid tariffs and the cost of service, regulators should provide for a limited and
monitored flexibility in tariff rates. Without this flexibility, microgrid companies will
have difficulty making sufficient profits, ultimately resulting in fewer viable businesses
and thus more consumers without electricity access. For example, the cheap or even
free rates provided to agricultural customers require significant subsidies to maintain,
yet even the state-owned discoms do not always receive the state reimbursement for
these subsidies in a timely manner [301. A microgrid operator, with less connection
and political clout, cannot be expected to fare better in receiving the prompt payment
of subsidy. Even if subsidies are ultimately provided, if the timing is unreliable the
microgrid business cannot trust it to support their cash flow.
Of course, the first-best solution would be to provide subsidies in a more consistent
and reliable way, as described in Recommendation 2. However, the issue of subsidy
unreliability is rooted in serious systemic, bureaucratic, and often political issues.
Regulators must be realistic about the rate of reform, if they are serious about pro-
moting electricity access as soon as possible.
In light of this situation, regulators may wish to offer an exemption from adherence
to the grid tariffs, especially for microgrids that are not very large. This exemption
should still set maximum tariffs that can be charged. The maximum could be based
118
on the cost of service that would be incurred by the distribution company in the local
area (i.e. the tariff that they would charge if their tariffs were rational), or it could
be based on estimations of the efficient cost of service for a microgrid. The proper
level allowed will depend on the local economic conditions and the costs of generation
(which could decline with improving technology). Of course, in setting this maximum
regulators should still maintain carve-outs for below-poverty-line populations and
other vulnerable customer groups; ensuring timely supply of subsidies to enable cheap
supply to these people should be the utmost priority.
The requirement for the consent of the local governing body to the terms of service
would serve to validate the reasonableness of a proposed tariff schedule in a village.
As well, there should be a formal process for villagers to appeal to the regulator if
they feel that the microgrid is charging tariffs that are unreasonable.
Recommendation 6: Ensure Feed in Tariff accounts for time-of-service
The proposed subsidy structure accounts for the costs of generation through the
feed-in-tariff, and the costs of the network and operations through the franchisee fee
prior to the arrival of the microgrid. However, this structure does not account for
the value of time-of-delivery. The Feed-in-Tariff incentivizes total production, not
production at useful hours. Given that a large amount of off-grid generation in
India is solar PV, which does not produce during the peak morning and evening
hours, this time-of-use aspect is crucial.
In particular, most renewables-powered off-grid systems (and the regulations have a
heavy bias towards renewable generation) will address this need through the
purchase of a large battery, which is expensive and must be replaced every 3-5
years. While it could be possible to include the cost of the battery in the franchisee
fee, this does not address the issue of how to ensure the value of the battery is not
lost if and when the central grid arrives; the ongoing feed-in-tariff would not cover
it, and the discom's purchase of the network assets is unlikely to include it.
119
1-11". -1 - __j-'Jk ... " , - ---
Time-of-use pricing is not typically proposed for rural microgrids due to the cost of
smart meters for each consumer. However, the cost of a single smart meter to
observe the production from the generator is a significantly more manageable cost,
especially if it allows the microgrid to be properly reimbursed for their battery
costs. Regulators can incentivize electricity service at useful hours by paying a
higher feed-in-tariff to solar-powered microgrids for peak hours that do not coincide
with solar production. The difference between the daytime feed in tariff and the
nighttime feed in tariff should be set to account for the cost of the battery, including
financing, operations, and depreciation costs.
This arrangement would be useful for properly reimbursing the microgrids for
providing a useful peak-hours service during off-grid operation, and it could serve to
address the central grid's peak-hour supply deficit once connected to the main grid.
(The regulator may wish to adjust the hours of 'peak tariff' to account for possibly
different hours in which the generation is most valuable to the central grid.)
Recommendation 7: Allow less-rigorous grid standards for microgrids
In order to be connected to the main grid, the microgrids are required to meet the
same electrical and safety standards required for the central grid. However, these
requirements can be overkill in rural areas. Brazil, for example, has significantly
reduced the costs of its rural electrification by establishing standards for rural
electrification which maintain system safety but are less onerous and less costly than
those in urban areas [111. While it is important that safety and the integrity of the
electrical system be maintained, regulators should specify the least onerous way
that this can be accomplished in rural settings.
5.4.2 Uttar Pradesh microgrid regulation
Uttar Pradesh released, in early 2016, regulations for microgrids that provide for
their connection to the central grid [561. It offers two options for microgrid owners
120
to 'exit' when the grid arrives: a total buyout of the microgrid including the
generation assets, or a buyout of the network only, allowing the owner to remain as
a power producer selling to the central grid. As noted above, this addition of
allowing the microgrid business to remain as a franchisee could have benefits.
The policy offers a 30% upfront subsidy to microgrids, but only under several
onerous requirements. The first of these is that the microgrids eligible for this policy
must meet the standards of the central grid, a requirement which could be improved
by modifying, as recommended above. Second, these microgrids must charge Rs
60/month for customers whose peak power draw is <50 watts, and must supply for
8h per day (3h in the morning, 5h in the evening) for all domestic customers. These
rates are quite low, lower than what we have observed in many off-grid arrangements
for a lower level of service. Under these conditions, microgrids may not be able to
break even, especially given the added costs of network compliance. Indeed, the
microgrids modeled using REM rarely break even with tariffs that are twice as high.
Finally, the regulations require all the solar-powered microgrids to not use any
fossil-based generation in addition, with an exception for limited use of a genset to
charge the battery. The restriction to renewable sources is not inherently
problematic, but requiring the back-up generator to operate only through charging
the battery is misguided. It would be much more efficient for the generator to
deliver that power directly to the grid, because charging and discharging the battery
results in a non-negligible amount of electricity losses.
The regulation does allow other microgrids that do not receive the subsidy (and
thus do not need to meet the tariff limits) but meet the grid standards to have
access to the two grid connection 'exit' strategies: the total buyout, or buyout of
network only. So, despite the shortcomings outlined above, this policy still leaves
room for the development of other grid-compatible business models, which is a great
leap in and of itself. Although there is room for improvement, both this policy and
121
the Forum of Regulators suggestions are significant improvements over the previous
status quo.
5.4.3 Conclusions
Indian policy makers, especially in the central government, are certainly aware of the
challenges facing distribution companies and microgrid companies in rural India. The
recent history of policy and regulation show that significant strides are being made
towards addressing the major risks facing off-grid investors and the challenges facing
the distribution companies. The UDAY scheme (discussed in Section 1.1.3.2) is the
first to require that states off-take the distribution company's debt on an ongoing,
creating an alignment of incentives which may spur much-needed tariff reforms. The
recent Electricity Act amendments acknowledging the significant risk to microgrids
from the expansion of the grid and requiring regulation are merely the capstone of a
significant body of work and thought regarding these issues in India. While there are
important issues to be addressed in the proposed regulation, they are an admirable
start. The next few years in India will tell how successful they are.
122
Chapter 6
Discussion and Conclusions
6.1 Utility of REM
REM remains under active development, and the version used here is by no means
the final version. However, even in its intermediate stage, REM shows its usefulness
in the analysis presented in this thesis. The flaws identified in the sensitivity analysis
are worth investigating, yet they do not give us significant reason to doubt the relative
conclusions reached in the results. The model largely behaves as anticipated under a
variety of conditions.
The most significant added value of REM could be its ability to showcase the relative
importance of different factors to electrification plans. While planners, policymakers
and academics may all know that a set of factors are important, they may not be
able to determine which are more important than the others. REM allows us to see
which factors are more influential.
For example, both diesel price and the efficiency of managing a microgrid are impor-
tant factors in the ongoing costs of a microgrid. There are more than a few papers
devoted to discussing possible business models for microgrids and how they can be
more efficient ([31 is one example). Although both appear significant, REM's sensi-
123
............
tivity analysis reveals that the diesel price can be much more influential on the final
suggestions than the efficiency of management, or the economy of scale in manage-
ment costs. Knowing the relative importance of these factors will help planners focus
their efforts on constraining and improving the most important issues.
Still, REM is only useful within a comprehensive understanding of the underlying
social, political, and regulatory context. [121 offers an in-depth exploration of many
of these factors. The importance of non-technical factors is evident in the motivation
for grid compatible microgrids themselves; the uncertainty of grid expansion cannot
be captured adequately within REM, yet it is a key factor in thinking about the
relative value of grid compatible microgrids.
6.2 Potential for Grid Compatible Microgrids
A key motivation for this analysis was to determine what the potential is for grid-
compatible microgrids to enable societal cost savings, and in what circumstances
the higher upfront costs of a grid compatible microgrid could be outweighed by the
benefits. The results show that there is significant room for grid compatible microgrids
to create value for society, but they are unlikely to be profitable on their own, and
that value is significantly dependent on the value assigned to non-served energy.
Although REM's analysis shows that ongoing subsidies will likely be needed to make
these businesses profitable, the necessity of subsidies does not mean that they cannot
create societal value. In the case of subsidies for the poor, the government has clearly
made the choice that it is more valuable to society to provide these people with a basic
level of access. As well, the value to society of greater electricity access (which would
be reflected in the government's willingness to subsidize it) includes values that may
not factor into individual decision making and individual willingness to pay; these
values could include the returns of future commercial growth, and improvements in
health, safety, and education. These are exactly the type of values that are captured
124
in the cost of non-served energy within REM1 .
The all-grid electricity provision plan is more expensive when taking into account
the costs of non-served energy. Under this scenario more of the costs are borne by
individuals, who incur costs and forgo significant opportunities due to their lack of
electricity access. As well, none of the costs are borne by private parties. In the
mixed approach, in which both grid expansion and microgrids are used, the costs are
borne less by individuals and more by the government and private investors. The
financial costs borne by the government in this scenario are likely to be higher due to
the high levels of subsidy that might be required. However, the total societal costs
will decrease due to the reduction in the cost of non-served energy, and in the long
run the government and society at large will benefit from the economic improvement
which typically follows energy access.
Not only is an all-grid electricity provision plan more expensive when taking into
account the costs of non-served energy, but it also presents the possibility for more
significant delays in electricity access. The expansion of the central grid in India has
been hampered by a lack of funds, political manipulation of the electricity sector, and
extensive bureaucracy. As a result, the costs of nonserved energy shown in Figure 4-
12a on page 96 for the all grid scenario are likely an underestimate, as there would
be significant unmet demand as the grid expanded slowly. Under the right set of
regulations, the expansion of microgrids could be much more rapid.
There are five other key assumptions present in the analysis conducted which would
have the impact of underestimating the value of microgrids relative to grid extension.
The first is the assumption of only household demand for all customers, which neglects
the significant number of businesses, schools, health centers, and other users who
would have a significantly different usage profile and ability to pay than the household
customers we have simulated. Including these customers in our analysis would likely
'It is worth noting that in future versions of REM, we expect the cost of non-served energy to
represent only the costs associated with intermittency of electricity service - not the costs of not
having it in the first place. In lieu of including those costs in the cost of non-served energy, we
expect to have a more robust implementation of a minimum service threshold within REM.
125
improve the financial viability of microgrids, as these customers tend to buy more
power at hours when it can be cheaply provided by solar and are willing to pay more
for it. TaraUrja, a microgrid developer in India, has told us in conversation that
they try to only build microgrids in villages with a significant number of commercial
entities, as these locations are the ones where they will be able to create a sustainable
business. Granted, these customers would have a similar effect on the viability of
grid extension. However, commercial load which typically operates in the daytime
(as opposed to the morning and evening peaks for households) can make an even
bigger difference on a microgrid powered by solar than on the grid, since solar-powered
microgrid power is abundant during the day but more scarce and costly to deliver in
the evening.
Next is the influence of grid reliability, and the upstream reinforcements of the cen-
tral grid which would be required to maintain and improve that level of reliability
when more customers are added. The line capacities are already insufficient for peak
demand in many rural areas of Bihar, according to the system operators. When more
consumers are added, the distribution company will either need to invest in signif-
icant upstream upgrades, or else provide power to an even smaller fraction of their
customers than before. Both of these options have the impact of significantly raising
the cost of any grid-based approach. Third, the grid reliability used for the 'low grid
reliability' scenario is 75%, which is consistently higher than what we observed in
rural feeder data in Bihar, so these cost of nonserved energy estimates are signifi-
cantly underestimated, especially given that reliability in a given hour seems to be
anti-correlated with demand in that hour [17].
Fourth, the wiring costs modeled for microgrids could be significantly reduced, since
these values currently are based on the same standards as used for urban grid lines.
These high standards are not necessary in rural areas to maintain safety and effec-
tiveness; reducing them could have a significant impact on costs for microgrids.
Finally, the costs of nonserved energy used in this assessment are likely to be under-
estimates, as discussed at length in Appendix B. A higher cost of nonserved energy
126
would bias the result against a grid that is unreliable, or even a reliable grid that is
slow to expand. The sum of all of these assumptions is a reduction in the cost of an
all-grid approach and an increase in the costs of a microgrid-based approach.
The determination of the value of a grid-compatible microgrid over lower-quality
microgrids was hampered by the failure of the attempt to simulate DC microgrids
within the REM framework. Although scenario 2 featured a set of microgrids under
a set of lower penalties for non-served energy costs, with higher discount rates and
lower lifetimes to simulate increased risk, these systems are not a good proxy for
existing DC microgrids, in part because hey were very large, with an average size of
1257 households per microgrid. As a result, a direct cost comparison is not useful.
As well, it was not a useful proxy for determining the value of the investment that
would be lost if and when the grid expanded into the domain of a DC microgrid.
We were expecting to see a larger effect from higher discount rates on the attractive-
ness of grid-compatible microgrids, representing the reduction in risk occurring from
better regulation. However, the effect of the discount rate in the range tested (9%-
14%) appeared to be small. Perhaps a wider range would have given more informative
results, given that investors are known to avoid this sector entirely due to the risks
of grid expansion [14]. This effect might have been captured with a higher discount
rate or shorter lifetimes. Whatever the mechanism, the risk of grid extension has a
much larger effect in real life than captured in the REM simulations. It could be
useful to think of the benefits of making microgrids grid-compatible not just in terms
of decreased financing costs, but also in terms of the value of non-served energy that
arises from the deterrence of investment in this area.
There is some room for non-grid-compatible microgrids to be a cost-effective interim
solution, especially if their costs are very low relative to the costs of a grid-compatible
microgrid and if the grid is slow to expand. If the loss of the depreciated book value
of the remaining microgrid assets at the time of grid expansion is considered as a
cost to society like cost of nonserved energy, and if that depreciated book value lost
is less than the cost of nonserved energy that would have been incurred, we can say
127
that it would be preferable to have a non-grid-compatible microgrid servicing critical
electricity demand in the interim.
6.3 Regulatory conclusions
The takeaways from the analysis with REM, and the assessment of the literature
and best practices highlight the need for regulation that reduces the economic risk
imposed by grid expansion. They suggest that containing costs through efficient
regulation and well-designed subsidies is key to ensuring the financial survival of
microgrid businesses.
As shown by [14], the threat of grid extension is a key risk that is preventing significant
investment in microgrids, and all the business innovation that comes with it. There
are two main ways that regulators can combat that risk. The first is by forcing a
discom to publish and abide by a reasonable grid expansion plan, which would show
potential entrepreneurs where their projects would remain unmolested by the central
grid. The second is to reduce the financial losses suffered by the microgrid when the
central grid does arrive, through specifications for making microgrids grid-compatible,
and regulations governing the ways that the two entities can interact and the process
for determining payments from the discom to the microgrid.
These two approaches for mitigating grid expansion risk are highly complementary.
A grid compatible microgrid will be able to plan its investments more prudently if it
knows the date at which it will connect to the main grid. Of course, both are easier
said than done. The animosity of discoms towards microgrids who may be viewed as
competitors requires smart regulations and empowered regulators to ensure that such
regulations are useful in practice, and not simply another source of uncertainty.
A highlight from the REM analysis is the significant need for ongoing subsidies to
support grid compatible microgrids. These subsidies may not be as high in real
life as they are in this idealized analysis, once accounting for commercial customers
128
and potentially lower network costs. However, they are likely to still be necessary.
Regulators should ensure that their grid compatible microgrid scheme accounts for
this need, without encouraging inefficiency on the part of the microgrid business.
India is well on its way to promoting this type of regulation. The proposal from the
Forum of Regulators discussed in Chapter 5 provides for on-going subsidies and pro-
vides for a more certain end-game for the microgrid when the grid arrives. However,
these suggestions might not be taken up. The Uttar Pradesh microgrid regulations
provide for no ongoing subsidy, and they require the microgrid's tariff to be quite
cheap, which appears to be a recipe for insolvent microgrids. There are additional
federally-funded subsidy schemes, through DDUGJY for rural electricity access and
the Jawaharlal Nehru National Solar Mission, which could potentially help make up
this deficit, but they both only provide upfront grants, still leaving a significant gap
in ongoing funding.
Regulators must also directly confront the issue of the discoms being a suspect partner
in implementing pro-microgrid regulations. Discoms are not, at the moment, very
trustworthy sources of subsidy due to their poor financial situation. The perceived
competition from microgrids may make them even less helpful, as they may seek
to avoid handing over parts of their established territory to a competitor. Future
regulations might consider addressing this latter issue through actually enforcing the
universal service obligation with real financial penalties, and making microgrids a way
to avoid these penalties.
6.4 Future work
There is a significant body of work that could build off of the initial analyses presented
in this thesis. Investigating and rectifying the shortcomings and inconsistencies noted
in the sensitivity analysis and results presented in Chapter 4 is the first order of busi-
ness. Indeed, improvements to REM have been ongoing during the writing of this
thesis and will continue afterwards. It could even be worthwhile to conduct the exact
129
same scenario analysis with the improved version of REM in order to understand how
better clustering methodologies, the inclusion of topography, and upstream network
reinforcement costs affect the outcomes. As well, Vaishali is just one case study, rep-
resenting a fairly homogenous and flat area. We do not know how different the results
might be for a region with different geography and significantly different settlement
patterns.
The capability to model the costs of upstream network improvements, which is being
actively developed for REM, suggests the possibility of assessing the impact that the
presence of distributed generation will have on the need for network reinforcement.
The impact of DG on these costs may affect regulator's assessment of the value of
storage, to meet peak demand hours. Such an analysis could help establish the levels
of incentive (for example, through a time-variable feed in tariff) that regulators would
wish to provide. As well, the capability of predicting network improvement needs is
one that was explicitly requested by the managing director of the North Bihar Power
Distrbution Company to us in meetings.
Demand is notoriously difficult to predict, especially for customers who do not yet use
electricity. As well, the omission of non-residential loads from this analysis probably
had a significant impact on the calculated viability of different electrification options.
Future work should consider in more detail the impact of having multiple demand
types.
Finally, there is room for a more thorough assessment of the adequacy of proposed
and future grid-compatible microgrid regulations to bring certainty to investors. It
is difficult to know from afar how reassuring these regulations are, and to what de-
gree they are expected to be abused and manipulated to the detriment of microgrid
entrepreneurs. Semi-structured interviews with investors and other stakeholders in
India and other countries with similar regulations would elucidate thse questions. The
answers are important, because they can help guide regulators' efforts in improving
future regulation.
130
6.5 Conclusions
The regulatory and economic context for grid compatible microgrids is changing
rapidly in India and elsewhere. Over the coming years, the recent amendments to In-
dia's Electricity Act of 2003 will be implemented in the form of state-level regulations
governing the connection of microgrids to the main grid. This activity will generate
significant interest and development in the relevant regulatory ideas, and create the
opportunity for new ideas and approaches to enter an often-conservative industry.
Of course, there is room for pessimism alongside optimism; these systems operate
in retrenched bureaucracies that are often tied up in politics. The strength of the
resulting regulations will be tested by this challenging environment. In either case,
it will provide ample fodder for studying and learning from the success and failure of
the different regulations implemented in each state.
The technology available to off-grid systems is also undergoing rapid innovation. The
drop in solar panel costs is well known and its impacts will continue to be felt in the
increasing attractiveness of off-grid solar-powered solutions. As well, innovation in
communications, power electronics and meters, and the development of these products
explicitly for a cost-conscious, developing-world context is enabling more complex and
interesting billing methods and business models. This business model innovation can
create new opportunities for microgrids to be profitable businesses.
Finally, this regulatory, technological, and business model innovation is occurring at
a time of growing awareness of the climate impacts of electricity generation, and the
need to reduce emissions growth in the developing world as well as the developed
world. Although expanding access to electricity accounted for just 3-4% of India's
carbon emissions increases in the last three decades, reducing the carbon intensity
of the energy supply is an imperative that cannot be overlooked [421. Electricity
use in India accounted for 11-25% of the increase in carbon emission since 1981,
and reducing the carbon intensity of generation should be a priority [42]. While
a basic level of electricity access should not be sacrificed in the name of carbon
131
emissions, as the emissions are small and the human dignity implications are large, it
is increasingly possible that this is a false tradeoff. It is possible to expand electricity
access while using low-carbon generation methods, as evidenced by the proliferation
of solar-powered microgrids and home systems in India and Africa, and the significant
regulatory and financial support from India's government for renewable-powered on-
and off-grid generation. Incorporating that off-grid generation into the central grid
as it expands-will not only do a favor for microgrid investors, but also for the climate.
132
Bibliography
[11 ABPS Infrastructure Advisory Pvt. LTD. Final Report On Policy and Regulatory
Interventions to Support Community Level Off-Grid Projects. Technical Report
November, 2011.
[21 ABPS Infrastructure Advisory Pvt. LTD. Model Draft Guidelines for Off Grid
Distributed Generation and Supply Framework, 2012.
[31 Venu Aggarwal, Alison Fahey, Hana Scheetz Freymiller, Simonne Li, C.C. Huang,
and Stephen Moilanen. Rural Energy Alternatives in India: Opportunities in Fi-
nancing and Community Engagement for Renewable Energy Microgrid Projects.
2014.
[41 Ashden India and Shakti Sustainable Energy Foundation. DRE-Based Micro
Grids: Motivation for Developing Grid-Interactive Systems. Technical report,
Shakti Sustainable Energy Foundation, 2014.
[5] Shishir Asthana. UDAY is a revival plan for discoms rather than a bailout
package, nov 2015.
[6] Pepukaye Bardouille. From Gap to Opportunity : Business Models for Scaling
Up Energy Access. International Finance Corporation, 2012.
[7] Thushyanthan Baskaran, Brian Min, and Yogesh Uppal. Election cycles and
electricity provision : Evidence from a quasi-experiment with Indian special
elections. Journal of Public Economics, 126:64-73, 2015.
133
[8] Subhes Bhattacharyya. Rural Electrification Through Decentralised Off-grid Sys-
tems in Developing Countries. Bentham Science Publishers, jul 2013.
[9] Subhes C. Bhattacharyya. Review of the Electricity Act 2003 of India. Center
for Energy Petroeum and Mineral Law and Policy, (November), 2003.
[10] Nicola U. Blum, Ratri Sryantoro Wakeling, and Tobias S. Schmidt. Rural elec-
trification through village grids - Assessing the cost competitiveness of isolated
renewable energy technologies in Indonesia. Renewable and Sustainable Energy
Reviews, 22:482-496, 2013.
[11] D Bornstein. How to change the world: Social entrepreneurs and the power of
new ideas. Oxford University Press, 2007.
[121 Yael Borofsky. Towards a Transdisciplinary Approach to Rural Electrification
Planning for Universal Access in India. PhD thesis, Massachusetts Institute of
Technology, 2015.
[131 Andrew Campanella. An Analysis of the Viability and Competitiveness of DC
Microgrids in Northern India. PhD thesis, Massachusetts Insitute of Technology,
2013.
[14] Stephen D Comello, Stefan J Reichelstein, Anshuman Sahoo, and Tobias S.
Schmidt. Enabling Mini-grid Development in Rural India. 2015.
[151 Concept Economics and Center for Advanced Engineering New Zealand. Inves-
tigation of the Value of Unserved Energy. Stage 1 Report for the Electricity
Commission, (December), 2008.
[16] Ranjit Deshmukh, Juan Pablo Carvalho, and Ashwin Gambhir. Sustainable
Development of Renewable Energy Minigrids for Energy Access : A Framework
for Policy Design National Laboratory. page 28, 2013.
[17] Douglas Ellman. The Reference Electrification Model : A Computer Model for
Planning Rural Electricity Access. PhD thesis, Massachusetts Institute of Tech-
nology, 2015.
134
[18] Forum of Regulators. Strategy for Providing 24x7 Power Supply. 2014.
[19] Forum of Regulators. Report on India's Renewable Electricity Roadmap 2030.
Renewable Electricity Roadmap 2030, 2015.
1201 Govt. of India Ministry of commerce & Industry. Cabinet approves amendments
in Power Tariff Policy to ensure 24X7 affordable Power for all, jan 2014.
[21] Chris Greacen, Richard Engel, and Thomas Quetchenbach. A Guidebook on
Grid Interconnection and Islanded Operation of MiniGrid Power Systems Up to
200 kW. Lawrence Berkeley National Laboratory, (April), 2013.
[22] Charles Moonga Haanyika. Rural electrification policy and institutional linkages.
Energy Policy, 34(17):2977-2993, 2006.
[23] Santosh M Harish. Do rural residential electricity consumers cross-subside their
urban counterparts ? Exploring the inequity in supply in the Indian power sector.
(August), 2014.
[24] lans. Nitish Kumar orders free power connection to all, nov 2015.
[25] F. S. Javadi, B. Rismanchi, M. Sarraf, 0. Afshar, R. Saidur, H. W. Ping, and
N. a. Rahim. Global policy of rural electrification. Renewable and Sustainable
Energy Reviews, 19:402-416, 2013.
[261 Deepak Paramashivan Kaundinya, P. Balachandra, and N. H. Ravindranath.
Grid-connected versus stand-alone energy systems for decentralized power-A re-
view of literature. Renewable and Sustainable Energy Reviews, 13(8):2041-2050,
2009.
[27] Francis Kemausuor, Edwin Adkins, Isaac Adu-Poku, Abeeku Brew-Hammond,
and Vijay Modi. Electrification planning using Network Planner tool: The case
of Ghana. Energy for Sustainable Development, 19(1):92-101, 2014.
[281 Alok Kumar aiid Sushanta K. Chatterjee. Electricity Sector in India: Policy and
Regulation. Oxford University Press, 2012.
135
[29] Sadhan Mahapatra and S. Dasappa. Rural electrification: Optimising the choice
between decentralised renewable energy sources and grid extension. Energy for
Sustainable Development, 16(2):146-154, jun 2012.
[301 P.C Maithani and Deepak Gupta. Achieving Universal Energy Access in India:
Challenges and the Way Forward. Sage Publications India Pvt Ltd, 2015.
[311 David Manetsgruber and Dernard Wagemann. Risk Mitigation in Mini-Grids.
Institute ID-EEE and Hochschule Neu-ulm University of Applied Sciences.
[321 V Manglik. New Report Identifies Need for Investment in Indias Beyond- The-
Grid Sector, 2015.
[33] Carlos Mateo Domingo, Tomss G6mez San Roman, Alvaro SAnchez-Miralles,
Jesds Pascual Peco Gonzailez, and Antonio Candela Martinez. A reference net-
work model for large-scale distribution planning with automatic street map gen-
eration. IEEE Transactions on Power Systems, 26(1):190-197, 2011.
[34J Dimitrios Mentis, Manuel Welsch, Francesco Fuso, Oliver Broad, Mark Howells,
Morgan Bazilian, and Holger Rogner. Energy for Sustainable Development A
GIS-based approach for electri fi cation planning - A case study on Nigeria.
Energy for Sustainable Development, 29:142-150, 2015.
[35] E Mills. Technical and economic performance analysis of kerosene lamps and
alternative approaches to illumination in developing countries. Lawrence Berkeley
National Laboratory Report, 2003.
[36] Ministry of Power Government of India. Office Memorandum: Deendayal Upad-
hyaya Gram Jyoti Yojana (DDUGJY), 2014.
[37] Mohua Mukherjee. Lessons Learned from Two Decades of Experience with Pri-
vate Sector Participation in the Indian Power Sector. Technical report, India
Power Sector Diagnostic Review, World Bank, Washington D.C., 2013.
136
[38] Francesco Nerini, Oliver Broad, Dimitris Mentis, Manuel Welsch, Morgan Bazil-
ian, and Mark Howells. A cost comparison of technology approaches for improv-
ing access to electricity services. Energy2, 95:255-265, 2016.
[39] Alexandro Niez and International Energy Agency. COMPARATIVE STUDY
ON RURAL ELECTRIFICATION POLICIES IN EMERGING ECONOMIES:
Keys to successful policies 2010. Technical report, International Energy Agency,
2010.
[40] Hisaya Oda and Yuko Tsujita. The determinants of rural electrification: The
case of Bihar, India. Energy Policy, 39(6):3086-3095, jun 2011.
[41] Sanusi Ohiare. Expanding electricity access to all in Nigeria: a spatial planning
and cost analysis. Energy, Sustainability and Society, 5(1):8, 2015.
[42] Shonali Pachauri. Household electricity access a trivial contributor to C02 emis-
sions growth in India. Nature Climate Change, 4:1073-1076, 2014.
[431 Lily Parshall, Dana Pillai, Shashank Mohan, Aly Sanoh, and Vijay Modi. Na-
tional electricity planning in settings with low pre-existing grid coverage: Devel-
opment of a spatial model and case study of Kenya. Energy Policy, 37(6):2395-
2410, 2009.
[44] Jesus Peco. A Reference Network Model: the PECO Model. Instituto de In-
vestigacion Tecnologica Universidad Pontifica Comillas Working Paper, (June),
2004.
[45] Press Information Bureau Government of India. UDAY (Ujwal DISCOM As-
surance Yojana) for financial turnaround of Power Distribution Companies, nov
2015.
[46] Ana Pueyo, Francisco Gonzalez, Chris Dent, and Samantha DeMartino. The
Evidence of Benefits for Poor People of Increased Renewable Electricity Capacity:
Literature Review. Institute of Development Studies, (Sept), 2013.
137
1471 Manish Ram, Ramapati Kumar, Sven Teske, and Greenpeace. "e[r] cluster" for
a smart energy access: the role of microgrids in promoting the integration of
renewable energy in india. Technical report, Greenpeace, 2012.
1481 Aly Sanoh, Lily Parshall, Ousmane Fall Sarr, Susan Kum, and Vijay Modi. Local
and national electricity planning in Senegal: Scenarios and policies. Energy for
Sustainable Development, 16(1):13-25, 2012.
[491 Daniel Schnitzer, Deepa Shinde Lounsbury, Juan Pablo Carvallo, Ranjit Desh-
mukh, Jay Apt, and Daniel M. Kammen. Microgrids for Rural Electrification: A
critical review of best practices based on seven case studies Microgrids for Rural
Electrification : A critical review of best practices. Technical report, Berkeley,
CA: LBNL, 2014.
[501 Gireesh Shrimali and David Nelson. Solving India ' s Renewable Energy Fi-
nancing Challenge : Which Federal Policies can be Most Effective ? A CPI-ISB
Series. (March), 2014.
[51] Sudheer Pal Singh. States raise tariffs, but discoms still in doldrums, apr 2013.
[521 S Szab6, K B6dis, T Huld, and M Moner-Girona. Energy solutions in rural
Africa: mapping electrification costs of distributed solar and diesel generation
versus grid extension. Environmental Research Letters, 6(3):034002, jul 2011.
[531 S. Szab6, K. B6dis, T. Huld, and M. Moner-Girona. Sustainable energy plan-
ning: Leapfrogging the energy poverty gap in Africa. Renewable and Sustainable
Energy Reviews, 28:500-509, dec 2013.
[541 Bernard Tenenbaum, Chris Greacen, Tilak Siyambalapitiya, James Knuckles,
and The World Bank. From the Bottom Up: How Small Power Producers and
Mini-Grids Can Deliver Electrification and Renewable Energy in Africa. Tech-
nical report, The World Bank, 2014.
[55] The World Bank. Sustainable Energy for All Global Tracking Framework 2015.
2015.
138
[56] UTTAR PRADESH NEW AND RENEWABLE ENERGY DEVELOPMENT
AGENCY. Mini Grid Policy Uttar Pradesh 2016. pages 1-6, 2016.
[57] Gayathri Vaidyanathan. Coal Trumps Solar in India, 2015.
[581 Wartsila India Ltd. The Real Cost of Power. 2009.
[59] Adriaan Van Der Welle and Bob Van Der Zwaan. An Overview of Selected
Studies on the Value of Lost Load ( VOLL ). 2007.
[601 Nathaniel J Williams, Paulina Jaramillo, Jay Taneja, and Taha Selim. Enabling
private sector investment in microgrid-based rural electri fi cation in developing
countries : A review. Renewable and Sustainable Energy Reviews, 52:1268-1281,
2015.
[61] Nancy Wimmer. Green Energy for a Billion Poor: How Grameen Shakti Created
a Winning Model for Social Business. MCRE Verlag UG, 2012.
[621 Chi-Keung Woo and Roger L. Pupp. Costs of Service Disruptions to Electricy
Consumers. Energy, 17(2):109-126, 1992.
[631 Marianne Zeyringer, Shonali Pachauri, Erwin Schmid, Johannes Schmidt, Ernst
Worrell, and Ulrich B. Morawetz. Analyzing grid extension and stand-alone
photovoltaic systems for the cost-effective electrification of Kenya. Energy for
Sustainable Development, 25:75-86, 2015.
[641 Alex Zvoleff, Ayse Selin Kocaman, Woonghee Tim Huh, and Vijay Modi. The
impact of geography on energy infrastructure costs. Energy Policy, 37(10):4066-
4078, 2009.
139
140
Appendix A
Inputs for REM: Definition and
Values Used
The most interesting of these inputs are also covered in Chapter 3. This appendix
contains a full explanation of the definition of the vast majority of the inputs to REM,
and the values used in the Vaishali case study.
A.1 Discount rates
Definition
These figures (along with the lifetime of the investment) are used to calculate the
annuity (i.e. the cost per year) from the upfront costs. They are critical inputs
because the annuity is the basis on which REM compares different options. There
is one for grid extension, for microgrid, and for isolated systems. They represent
the time value of money for the relevant actor; typically this would come from the
financing terms (i.e. debt or equity) that they receive, but more generally it reflects
the opportunity cost of putting this money towards the system in question, rather
than investing it elsewhere (for example to purchase other equipment that could be
141
used to start a business, or to purchase medicine or food). Thus, the actor who
is providing the money to build the electricity infrastructure influences what the
discount rate is, because they will have different opportunity costs for their money
depending on who they are. For example, for a bank, that opportunity cost may be
not investing in stocks or other projects, and for an individual consumer it may be
not purchasing medicine, food, or other necessities; clearly for these two actors the
opportunity costs, and hence the discount rates, would be different. The discount
rate would also be expected to vary based on government regulation, business model,
and financing arrangements.
The annuity value is what is used to compare different options of network and gen-
eration. The annuity is the sum of annual costs (like O&M costs, and non-served
energy costs) and the annualized upfront capital cost. The discount rate is used to
calculate the annuity with the following equation (where r = discount rate and L
lifetime of upfront capital cost):
Annuity = Up!ront capital cost * rAnnuity = (1-(1+r)-L )
The discount rate depends significantly on the regulatory and business scenario we are
considering. What discount rate we use depends on who will be funding the various
modes of electrification and what the overall goals of the planner are. The discount
rate encapsulates two main ideas; first is the opportunity cost of capital, and second
is the risk associated with waiting. The opportunity cost reflects the returns that one
could get from putting that capital towards something else; what that cost is depends
closely on what opportunities are available to the agent holding the capital. For a
poor rural family, those opportunities may include buying medicine, a new piece of
farming machinery, or sending a child to school; for an investor who might fund a
microgrid, they may be comparing the microgrid to opportunities like investing in the
stock market, or in a different local business.
The risk portion of the discount rate refers to the idea that the future is unpredictable;
there is uncertainty about market conditions, about the opportunities that might be
142
available in several years, and about the likelihood of an investment to yield the
expected returns. Different businesses have very different levels of risk; a utility may
be relative low-risk, since it has a regulated rate of return, while a microgrid could
be significantly more risky, contending with issues like local acceptance, maintenance
quality, demand growth, and the arrival of the central grid, among others.
The Capital Asset Pricing Model (CAPM), which is commonly used to determine the
rate of return that investors should require from a project, exemplifies the dual nature
of the discount rate. (Note that discount rate and interest rates/rates of return can
be two sides of the same coin.) CAPM calculates the rate of return as the sum of the
'risk free' rate of investing, plus an added rate to account for the risk inherent in the
specific project and the returns that are expected above 'risk free' returns.
Whose discount rate? The resources used to purchase electricity and build infras-
tructure have different opportunity costs depending on whose wallet they are taken
from, as well as the type of opportunity costs that are considered. For this reason,
we have allowed REM to use a different discount rate for each mode of electrification;
different sorts of actors may own them. The function of the discount rate is to incor-
porate the time value of money and risk into the cost comparisons made within REM.
Who bears these costs and what types of costs we allow our analysis to consider both
affect the discount rate the user should use in REM.
With regards to who bears the cost, the actor or actors who actually invest their
money in electricity infrastructure have different opportunity costs and hence dis-
count rates. Even if REM is run from a 'global social planner' point of view, where
the user is interested in finding a design that maximizes overall societal welfare, it
might need different discount rates for different types of systems. The ownership
structure and risks of different business models are important because by choosing to
invest resources in electricity infrastructure, the actor in question would be forgoing
investment in other opportunities that could also create value for society. For exam-
ple, if we consider that a solar home system may be owned by a single household, the
143
discount rate used will be different than if we consider that the system is owned and
leased by a larger company; in the latter scenario, we would use lower discount rates
because they have less uncertainty and different alternative opportunities.
The perspective of the analysis we wish to conduct with REM also affects what types
of opportunity costs we wish to incorporate. There may be opportunity costs than an
individual actor may not consider but that a social planner (for example in the central
government) would want to incorporate. Examples of these opportunity costs include
the GDP, development, and/or political impact of other investments the government
could make with the money needed to fund or subsidize an electrification project. A
user of REM who is more socially-minded would want to incorporate these broader
benefits, Conversely, a user of REM who is more market-driven (like the owner of a
business who wants to see where it is most likely that their off-grid technology would
be preferable to customers) would not want to include these social opportunity costs
in their analysis.
It is likely that a user of REM would be interested in funding or subsidizing to
an affordable level all modes of electrification; such users include government and
international development banks. In this case, since the source of the capital would
be the same across all modes of electrification, the opportunity costs would be the
same. This perspective, then, would dictate that the opportunity cost portion of
the discount rate be the same for all modes. However, different modes would still
have different risks associated with them, so this user may still wish to differentiate
between the modes slightly based on risk level.
Sensitivity to discount rate Discount rates typically vary within a rather tight
window, and small changes in the rate can have significant impacts on the annuity
calculated for different modes and, hence, the ultimate recommendation on electri-
fication mode from REM. The sensitivity to discount rate also hinges significantly
on the lifetime used for computing the annuity (which varies between components to
reflect their different useful lifetimes). The longer the lifetime, the more sensitive the
144
annuity is to the discount rate used. For high values of the discount rate, e.g. above
15%, extending the lifetime does not have a significant impact.
Values used for Vaishali
* Grid extension: 13.3% for NBPCDL
9 Microgrid: 10% at minimum
* Isolated system: 14% based on SELCO financing rates
Grid: For the extension of the grid, we could use 10% in the absence of other data.
This number is based on a paper from Climate Policy Initiative "Solving India's
Renewable Energy Financing Challenge" [501 which bases their discount rate on the
government cost of borrowing (7.83% at the time of writing, 7.77% on Jan '16, based
on Indian 10 year bonds), plus a risk premium of 2%. This risk premium is meant
for large-scale renewable energy, so it may not be representative for a utility. Indeed,
this number may be too low as evidenced by regulatory documents in India.
The regulatory documents for the North Bihar Power Distribution Company Ltd
(NBPDCL) for Fiscal Year 2015-2016 from the Bihar Electricity Regulatory Com-
mission (BERC) grant NBPDCL a rate of return on their equity of 14%, and an
interest rate on their loans of 13%1. The rate of return on equity is established by
Regulation 73 (2) (c) of the Bihar Electricity Regulatory Commission regulations of
2007. These regulatory documents assume that the debt:equity ratio for capital in-
vestments will be 70:30, which gives us a weighted average cost of capital of 13.3%
2
Microgrids: This number will vary based on the regulatory context that we consider.
The risk premium will be lower if we consider a regulatory structure that accounts
for grid-compatible microgrids in a reliable and trustworthy way, but it will be higher
'Taken from Table 6.57 and Table 4.62, found at http://berg.co.in/media/Tariff-
Order/Tariff%20Order%20NBPDCL%20FY%202015-16.pdf Accessed April 6, 2016
2However, these regulatory documents note that NBPDCL "has projected funding of capitaliza-
tion through Grants at 94% and through Loans at 6%," meaning that the capital investments of
NBPDCL could be more likely to be dictated by the preferences of the central government grant
maker than by NBPDCL and their discount rate.
145
if the regulation does not provide investors with sufficient reassurance that their
investment will be properly valued and compensated when they connect to the main
grid. The risk premium will again be higher when we consider non-grid-compatible
microgrids, as the companies planning these will lose their whole investment when
the grid arrives; additionally, previous conversations with them have revealed that
they typically look for a relatively short payback period (on the order of 5 years).
TaraUrja uses a discount rate of 10% in the calculations that they sent to us, and
the World Bank cites 10.3% as the interest rate "that usually meets the short and
medium-term financing needs of the private sector." In light of the justification for
10% as an appropriate interest rate for a utility or large project (in [50J), this rate
does seem optimistic. [501 cites 12.3% as the commercial rate of interest applied
to renewable projects in India, though it is meant for larger utility-scale generation
installations, which could be perceived as lower risk than a microgrid. [31 and [231
use an interest rate of 12% for their calculations regarding solar microgrids (though
neither specifies a source for this figure). It seems that 10% is a reasonable lower
bound for microgrids. One would expect microgrids to have a higher discount rate
than utilities as they are often riskier investments, so in that light a higher rate may
be justified.
Individual home systems: Individual customers are known to have very high
individual discount rates (on the order of 20+%), likely due to high uncertainty
surrounding the future. [23] observes: "Ekholm et al. (2010) use discount rates of
62-74% for rural households and 53-70% for urban. Reddy and Reddy (1994) estimate
an internal rate of return of 28% for a switch from kerosene lamps to electricity- which
could be a lower bound on the discount rate". However, a lower discount rate would
be appropriate when financing is available to households.
Grameen Shakti is a highly-successful social enterprise, which supplies solar home
systems with financing through the microcredit scheme supported by the Grameen
Bank in Bangladesh. Households wishing to purchase a system can have a 3-year
loan with a 12% interest rate, 2 years for 10% interest, and 1 year for 9% interest .
146
In India, SELCO works to provide purchasers of solar home lighting with financing;
they state interest rates ranging 5%-14% over 3 to 5 years with a 10-25% upfront
deposit (25% is required by the Reserve Bank of India) 3.
Very high discount rates could make individual home systems very unattractive rela-
tive to other solutions within REM. The appropriate discount rate likely depends on
a variety of factors, especially the availability of and access to financial institutions
and the reliability of the household's income.
A.2 Cost of Nonserved Energy (CNSE)
Definition
This value refers to the penalty added to the cost of supply in REM when projected
demand is not met, on a per kWh basis. Critical values are higher than normal
values, and are used for loads that are higher-value to the consumer. For example, a
few lights at night would be considered critical, while a fan probably would not be.
The basis of CNSE should be in the value placed (either by the purchaser or by
society) on the supply of electricity and the services provided by it. The perspective
used will depend on the motivations of the planner and the anticipated business model
and subsidy structure. For example, if the model is used with the assumption that
the government will subsidize access to and quality of electricity, it may be more
appropriate to use a 'social' CNSE. This more global cost would capture values like
lost health from kerosene pollution, reduced hours for businesses and students to
work, and reduced economic growth.
If instead the planner anticipates that the government will not be providing significant
support to such projects, it would be preferable for the user to use a CNSE that
represents the value to the consumer themselves, as they will be the ones footing the
3http://www.selco-india.com/finance.html Accessed April 6, 2016
147
bill for higher microgrid or isolated system costs. I have assumed the latter situation
in my calculations and have been choosing lower bounds when given the option, as a
result.
Values used for Vaishali
" Critical: $2, based on [591. This is higher than estimates from substitution
analysis and from the Power Corp. of India's Value of Lost Load assessment
(discussed below), but these numbers are known to be left-skewed and to not
capture important effects like the GDP impacts of lack of electricity access.
" Noncritical: $0.51 is a lower bound, taken from the Power Corp. of India's
Value of Lost Load assessment. [581
There are several possible ways of calculating the Cost of Nonserved Energy, but
none of them are likely to ever find the 'true' value. By examining several methods
and taking into account what they miss, we can assess a reasonable estimate for the
purposes of running REM.
Via substitution: See Appendix B for more detail on how to calculate CNSE by
looking at substitutes for electric lighting. In summary, we use kerosene lighting
costs as an estimate for CNSE based on the units of 'cost to provide the service
of illuminating a house'. This is a reasonable basis for a CNSE because people's
expenditure on kerosene is an indicator of the lower bound of the value they place on
energy services - particularly on the service of lighting. Lighting is one of the highest
value uses for electricity (with agricultural water pumps and cell phone charging the
likely competitors), so a CNSE based on this use is likely to be 'a lower bound on the
upper bound' of an individual's valuation of electricity services.
This method does not capture added benefits of using electric lighting over kerosene
such as health, increased quality of light, and the social value to development and
GDP of energy access. While an individual consumer may not take all these factors
into account, a social planner who has the interests of the whole country in mind
148
would wish to incorporate these values into a higher CNSE.
Via reported VOLL: Power companies regularly determine a 'Value of Lost Load'
(VOLL), representing the cost of an outage, to guide their investment decisions.
While the cost of an individual outage is likely to be lower than the cost of simply
having no supply at all, this is still a useful yardstick4 . A report by Wartsila India
reports that the Power Grid Corporation of India estimates the VOLL in India as
Rs.34/kWh to Rs.112/kWh, or $0.51 - $1.67/kWh [581. The range is used because
the cost of an outage is dependent on a variety of categories. Welle (2007) notes that
customer type (industry, service sector, households), perceived reliability level, time
of occurrence, duration, and advance warning can all affect the determination of the
VOLL [59].
[59] notes that for developing countries the literature strongly suggests a range of
VOLL that is 1-10 $/kWh, and likely 2-5$/kWh, higher than the Power Corp of
India's suggestions ([59] notes their values are likely to be left-skewed). The VOLL
for rural consumers in India would likely be on the low end for variation within
consumer type, low end for perceived reliability level, and the high end for variation
based on time of occurrence, duration, and advance warning. Without information
on the relative magnitude of these effects, it is difficult to say how these customers
are likely to compare to the average.
Summary: Substitution analysis suggests a minimum of $0.37 to $1.84/kWh (CFL vs
Incandescent) for critical CNSE. VOLL studies from India suggest a range of $0.51 -
$1.67/kWh and a literature review of VOLL studies in developing countries suggests
a range of $2 - $5/kWh. All of these values come with the caveat that they are likely
to be left-skewed (i.e. underestimates).
4 1n future version of REM, an improved minimum reliability requirement will eliminate the need
for accounting for the cost of having no supply at all.
149
A.3 Diesel cost
Definition
This cost is used for assessing the costs of operating a diesel generator in a microgrid.
Values used for Vaishali
I will use a price of $0.82/liter. In July 2015, we were told that the price of diesel was
$0.82/L (there is no longer any subsidy for diesel fuel in India). This is consistent with
data on mypetrolprice.com. Added onto this figure is a transport cost as described
below, taken from [53].
"The transport costs ($/l), Pt, for diesel are estimated using the following
equation:
Pt = 2Pdct/V
where Pa ($/I) is the national market price for diesel; c (1/h) is the diesel
consumption per hour; t (h) is the transport time, and V (1) is the volume
of diesel transported. The factor 2 is due to the fact that the vehicle has to
drive back to the origin point (assuming dedicated transport for the fuel in
the generators). For the calculations, we assumed average values of V=300
1 and c=12 1/h. These parameters assume the use of a standard small van
of a carrying capacity of 300 1. The transport cost is calculated as 0.01
Euro per kWh one way [9,22], with 12 1 consumption in 1 h (it can cover
different distances during the same period assuming approximately the
same consumption per hour depending on the roughness of the surface)."
[531
150
A.4 Network and component lifetimes
Definition
Each capital investment modeled by REM, from the network to the charge controller,
has an associated lifetime, which allows us to calculate an amortized value (the yearly
annuity) which we can use as a standard measure between options. The lifetime
represents how long until the component in question needs to be replaced, or how
long until the developer wishes to recoup their investment.
Values used for Vaishali
* Network: 40 years for grid, 20 years for grid-compatible microgrid.
- For grid extension, 40 years is a common value used in utility regulation.
- For a microgrid, a shorter period of time would be more appropriate as
a project developer does not have the kind of assurances that a utility
has regarding their long-term stability. TaraUrja, who builds non-grid-
compatible microgrids, annualizes their costs over 10 years in financial
data provided to us .
" Per customer investment: this represents the lifetime of the meter, switchboard,
lightbulb, etc need to be replaced.
- 20 years is used. Utilities may use a 40 year time horizon as they are
relatively stable businesses, while a microgrid developer may be working
on a 10-15 year time horizon. As a compromise, I have chosen 20 years;
after this point, increases in the lifetime have a negligible effect on the
annuity due to discounting.
" Generation components:
- Battery: lifetime given as throughput, and converted into years within
REM. Throughput is 103 kWh for the smallest (0.284 kWh) battery, and
845 kWh for the largest (1.38 kWh) battery
- Generator: lifetime given in hours (35,000), and converted into years within
151
REM.
- Solar Panel: 20 years
- AC/DC Converter: 15 years
- Charge Controller: 15 years
A.5 Per customer costs
Definition
For each mode of electrification, we define the per customer costs as the cost of the me-
ter or load limiters, if they are employed, any provided internal wiring, switchboards
and lighting. It does not include labor costs for installing these components.
This value is not the same as the connection charge that is charged by the utility or
microgrid provider. Sometimes the utility may inflate the connection charge in an
attempt to balance their books, and the charge may also be used as a commitment
mechanism to make customers invested in their new electricity infrastructure. Mera
Gao, for example, reports that they charge a $2 connection fee as a commitment
mechanism [13].
Values used for Vaishali
e Isolated System: No meter + home wiring = ~$10
* Microgrid: Cheap meter + home wiring = $10 + $12.50 = $25
* Grid Extension: Meter + home wiring = $25 + $12.50 = $35
Meter costs: Utility meters in Rwanda cost around $25'. Off-grid meters cost $20-30
6 in Rwanda. According to Taraurja, meters are available at Rs 700 from Bentex, i.e.
'Source: email communication with Paul Rugambwa (employed at Rwanda's Ministry of infras-
tructure), March 2015
6 Source: communication with Mobisol, Mesh Power, and 3E Power during workshop in Rwanda
2015. (All are energy service companies operating in Rwanda.)
152
USD $10.60 7. I will use the lower value from TaraUrja for off-grid meters, especially
since this price is likely to fall with innovation in this sector. Isolated systems would
probably not use a meter, as the user typically owns the whole thing, though there
are certainly 'solar as a service' business models for which this is not true.
Other costs:
TaraUrja, a microgrid provider in India, charges a $15 (1000 Rs) connection fee
covering the LED bulb, socket, switchboard, load limiter, and line leading to the
house (this does not include a meter). For an isolated system, we would expect lower
per household costs, since the load limiter and line are not necessary. I assume $10
for this value.
The labor cost of installing meters, switchboards and bulbs in houses is important to
capture. The experience of another MIT research group (the uLink project) shows
that the cost of installing in-house wiring for a basic Solar Home System in Jamshed-
pur, India, is $2.50/household.
A.6 Demand growth rate, years of demand growth,
and k-factor
Definition
REM includes three factors that relate to the trajectory of demand growth and elec-
trical losses growth over the duration of the project. These inputs are (1) the demand
growth rate, (2) the final year to consider for generation and network design, and (3)
the level of losses accounted for in the later year of the project, termed 'k-factor'.
The Demand Growth Rate is the rate of demand growth per year on an individual
customer basis, and the Years of Demand Growth defines how many years of demand
growth to consider in the design of the network and generation. If Demand Growth
7 Source: Email Communication with Sreyashee Das, TaraUrja, Jan 27 2016
153
is 0%, the Years of Demand Growth does not affect the output.
Demand grows for the number of years specified by Years of Demand Growth; the
demand in the last year of demand growth is the 'maximum demand.' The generation
will be designed for the year of maximum demand, regardless of the path taken to get
there (e.g. regardless of whether the level of demand is a result of 0% demand growth
or 50% demand growth and a lower initial demand). The network will be designed so
that line constraints are met in the year of maximum demand, and it will also take
into account the net present value of losses over the whole lifetime of the network.
The network will be sized to account for losses most effectively, and since losses are
assumed to be quadratically proportional to demand growth, this rate may affect the
network cost and hence network line choice.
For REM, the Years of Demand Growth is used in two ways: it is used as the 'years of
demand growth' and also the 'years of losses growth'; the latter represents the period
over which losses are expected to grow, given the demand growth rate specified.
REM currently treats these two values as the same, though they are separate inputs
to RNM. If the years of losses growth is longer than the years of demand growth,
RNM would assume that the losses continued growing as if demand kept growing at
the 'demand growth rate.' After this year, RNM assumes that the losses drop to a
flat value for the rest of the lifetime of the network, simulating the event of further
demand growth requiring reinforcement of the lines; that value of losses is determined
by the k-factor. The value of the losses for the rest of the lifetime of the project will
be the product of the initial losses and the k-factor. The years of losses growth and
the k-factor control the weight that power losses in the later part of the network's life
have in the NPV of a network.
Values used for Vaishali
Although the proper estimation of demand growth is an important issue for properly
building sizing the generation for microgrids, this analysis is not concerned with its
154
impact. I will assume that demand and losses are constant throughout the project's
lifetime. As a result, I will leave the demand growth rate at 0% and the final year for
network and generation design as year 1. The k-factor will be 1.
Minimum reliability threshold
Definition
This input defines the minimum percent of kWh demanded that the REM-proposed
solution must provide. The percentage is calculated over the timeframe of the total
period simulated by REM, which typically is three weeks. Its main function is to en-
sure that all customers are given some basic amount of electricity supply and to avoid
recommendations in which consumers are not provided any electricity. Thus, its main
function is not to enforce high-quality service, since it cannot differentiate between
service in different time periods; it can be used to force higher-quality minimum pro-
vision of electricity, but it is a crude way of doing so. The fault of using a similar
metric to assess quality of service (rather than presence of service) is illustrated in
the case of Indian utilities that are required to provide 6-8 hours/day of service to
villages in order to receive subsidy; the utilities may simply supply that electricity in
off-peak hours like the middle of the night when it is not needed much.
The preference for useful provision of electricity is reflected instead in the demand
profiles, the cost of non-served energy used for that demand, and which portions of
that demand are assigned a 'critical' or 'non-critical' cost of non-served energy.
Values used for Vaishali
This metric is not currently useful for requiring a certain quality of electricity service,
but rather for specifying that no households are left without a minimum supply of
155
electricity in REM's design'. The quality of supply should be modified through the
demand profiles and the cost of nonserved energy. Thus, the precise value of this
threshold is not critical.
The Indian Government has made its subsidization of grid extension to rural villages
contingent on them receiving at least 6-8h of service per day 123]. If demand is
evenly spaced throughout the day, for example in an area with robust residential
and commercial loads, this would suggest a value of 25%-33%. However, the study
period over which this metric is calculated is on the order of several weeks, and we
are looking a predominantly residential load that is clustered in the evening. This
would argue for a higher threshold to use for specifying a bare minimum of service,
so I use 33%.
A.7 Annual management costs
Definition
This category includes the definition of the number of customers in a 'small' and
'medium' microgrid (the number for a large microgrid is assumed to approach infin-
ity), as well as the annual management costs for a small, medium, and large microgrid.
These figures are used to define the costs in administration and revenue collection as-
sociated with each additional household (or building) added to a microgrid. The
distinction between small/medium/large is used to simulate economies of scale, and
these figures are used to fit a smooth exponential curve that is used to define costs
for all sizes of microgrids.
The values used for the small microgrid are also used for the management costs for
an isolated system. The costs for a 'large' microgrid are also used for grid extension.
8 This parameter will be improved in future versions of REM, so that it will be more specific to
the time of electricity service provision.
156
Values used for Vaishali
" Number of customers for small/medium microgrid: 1/100
" Annual Management cost small/medium/large microgrid per customer: $15/11/7
per year
Campanella's thesis reports the management costs and scale factors for Mera Gao;
when operating at full capacity, management and collection costs are approximately
$7/household/year [13]. This number is based on a regional office with a manager and
fixed costs that coordinates 150 microgrids of 30 customers each, and 20 employees
(10 technical maintenance staff, 10 collections agents) that each serve 15 microgrids.
We can use that figure to define the low end of the cost, as Mera Gao operates very
cheaply and this number assumes that the business has enough microgrids that they
have reached full economies of scale.
It is worth noting that the economies of scale in Campanella's figures for Mera Gao
come from having more microgrids of a fixed size, not larger microgrids, which is what
this REM input is meant to represent. However, it remains the best and most detailed
data I have found regarding the scaling of management costs. If the regional office
is operating at half capacity, the management cost goes up to $11/household/year,
which in the absence of better data is used as an estimate for the medium management
cost.
TaraUrja reports to us' that the salary of one person s 6000-10,000 Rs/mo, or $100-150
USD. If one of these people were fully employed administering a medium microgrid
(or 100 individual systems), the cost would be $15/household at the upper limit;
this would be a very inefficient arrangement. I use this cost as the upper limit for
management costs.
9Meeting, January 2016, New Delhi, India.
157
Grid electricity cost
Definition
This is the value used to calculate the costs of the electricity delivered through grid
extensions.
Value Used for Vaishali
$0.09/kWh is used.
The Forum of Regulators report, "Report on Road Map for Reduction in Cross Sub-
sidy", Chapter 6, Reports the cost of supply for domestic customers is 6 Rs/kWh
($0.09/kWh), for agricultural it is 9 Rs/kWh ($0.13/kWh) [191. However these figures
are all-India and do not seem to include losses. It also reports that total (technical
and nontechnical) losses for Bihar are 24.54%.
The Forum of Regulators report "Strategy for Providing 24x7 Power Supply" [18], cites
all-India Transmission and Distribution losses of 23.04% in 2012-2013. For Bihar this
number is 49.42%, yet the Forum's 2015 report cites a substantially reduced rate
(below), so improvement must have happened in the intervening years, though there
is some discrepancy with the numbers provided by Bihar distribution company.
Financial Statements for the Financial Year 2013-2014 from the.North Bihar Power
Distribution Companyl0 report an average power purchase cost of 4.2 Rs/kWh, and a
distribution loss of 35%. Including distribution losses, this results in a power cost of
~6.5Rs/kWh, or USD$0.097/kWh, although these losses may included 'non-technical'
losses like theft and non-collection. Based on this range of data, 9 cents/kWh is
justifiable, based on rounding off the figures from NBPDCL as well as the Forum's
figures for cost of supply.
'
0Acquired in hardcopy during a 2013 visit to Bihar.
158
A.8 Operations and maintenance (O&M) costs
Definition
REM breaks down O&M costs into 'fixed' and 'variable' components for each item.
The fixed component is represented in terms of man-hours per year, while the variable
component is represented as a fraction of capital cost. This breakdown captures the
idea that there are some maintenance tasks that do not change regardless of system
size, while others are highly dependent on system size.
Values Used for Vaishali
The value used for cost of labor per hour could be a range of values; we have heard
monthly salaries for technicians that range from $100 (from TaraUrja) to $2400 (from
Mera Gao). I will use a figure of $0.90/hr, which is in line with TaraUrja's upper
estimate of $150/month for their technicians, assuming a 40-hour workweek. The
values used here are drawn from those used in [171.
Fixed costs (man hours per year)
e Battery: 5 hrs
* Generator: 25 hrs
* Solar Panel: 5 hrs
o AC/DC Converter: 2 hrs
* Charge controller: 2 hrs
Variable costs (fraction of capital cost)
" Battery: 1%
" Generator: 5%
"Future versions of REM will add an additional level of detail by describing the fixed component
of the O&M as a per-system cost, allowing for economies of scale in maintaining several co-located
generation resources and also more accurately capturing the expected travel time between systems
for a technician. As the data-gathering for this variable has not yet been conducted, I will leave it
to future users of REM.
159
" Solar Panel: 1%
" AC/DC Converter: 1%
" Charge Controller: 1%
A.9 Installation costs
Definition
Installation costs for per customer components (meters, internal wiring) are included
in the cost of components. Installation costs for the network are included in the costs
as well. However, for the generation components, installation cost is given separately.
The values used here are drawn from those used in [171.
Values Used for Vaishali
" Battery: 20% of capital cost
" Generator: 80% of capital cost
" Solar Panel: 40% of capital cost for large (0.25 kW) panel, 20% for small
(0.02kW) panel.
" AC/DC converter and Charge Controller: 0 (their costs are included in the cost
of installing the solar panel)
A.10 Existing Network data
The location of the existing network and transformers was obtained directly from the
North Bihar Power Distribution Company Ltd, which provided us with a physical
map of the network in 2014. That physical map was geocoded (as described in [17])
with the help of Tata Consultancy. The location of MV/LV transformers was given
to us in Jan 2016. This data was used to extrapolate the location of consumers who
160
were already connected to the grid.
A.11 Demand & Customer Type
Definition The demand profiles used are the same as those described in Ellman's
thesis, and much of this description is paraphrased from it - please see section 2.6 of
[171 for a more detailed description. The profiles are constructed in a semi-random
way based on a set of activities that the electricity might be used for and a set of
parameters for each activity. These parameters are as follows:
* Whether the activity is 'critical' or 'noncritical', to assign the proper cost of
nonserved energy
o The energy used by this activity.
o The hours of the day in which this activity might happen.
o Restrictions (such as temperature or sunset times) on the usage of this activity.
o Average daily duration of activity.
o Variability in daily duration of activity.
From these inputs, a set of demand profiles is generated for every hour of the year.
Using a set of randomly generated demand profiles allows REM to account for benefits
that arise from aggregating customers with peak demand at different times.
Values used All customers were assumed to be domestic customers. While REM is
capable of handling more than one demand type, there is very little data regarding the
location of commercial loads available. The possible load types that these domestic
consumers used in the base case are as follows. The ownership probabilities are taken
from a National Sample Survey of India, see [171 for more details).
161
Usage Critical? Avg. Probability Avg daily Restrictions
# used of ownership duration
Critical light y 2 1 5 Irradiance less than
0.05
Extra light n 2 1 5 Irradiance less than
0.05
Night light n 2 1 5 Irradiance less than
0.05
Fan n 1 0.26 - Temp greater than
32.5 degrees C
TV - night n 1 0.26 5
TV - day n 1 0.26 1
TV - Standby n 1 0.26 -
A.12 Network and Generation Catalog
An important input to REM is the equipment available for generation and distribu-
tion. The equipment used is based on that used in [171, and the reader should refer
to section 2.5 for a complete description of the inputs and their sources. The network
catalog has been modified somewhat since the publication of 1171 by our collaborators
at IIT Comillas. These modifications are as follows:
o Remove the LV single phase wires, because their costs were not coherent with
the rest of the catalog.
o Adjust the costs of lines to be coherent with Indian values, and so that they
were monotonically increasing with capacity.
o Adjust resistance of each line to be monotonic with capacity.
These modifications were necessary to prevent the clustering step from getting stuck
at very small clusters.
162
Appendix B
Calculation of the Cost of Nonserved
Energy
In calculating the CNSE, there are a few different methods available to us. Woo and
Pupp (1992) identify three different methods - proxy-based, consumer surplus, and
contingent valuation [62]. The last two have very high data requirements and high
uncertainty.
" Consumer surplus methods involve obtaining a demand curve, which may be dif-
ficult, especially as the shape of the demand curve may depend on the timescale
involved.
* Contingent valuation relies on surveys, which can be unreliable, as villagers are
prone to overestimating their willingness to pay; additionally, their answers will
be constrained by their available budget.
[231 uses a combination of proxy-based and consumer surplus methods to calculate
a social cost of power interruptions. For the latter, they use cross-sectional data on
electricity consumption as it varies with different levels of tariffs charged in different
regions of India; from this data they construct a demand curve, which they use to
calculate consumer surplus forgone due to restricted supply. They add to this lost
consumer surplus the cost of the backup used (specifically, the costs that are in surplus
163
'_- , .. I- , - - -411_- .41 L. % .11, 1
of what they would have spent for electricity), in this case kerosene for lighting.
In this analysis we use the proxy method to determine the cost of non-served energy,
based on the cost of the alternative employed in the absence of electricity. The
consumer demonstrates that their valuation of the electricity service is at least the
cost of the alternative by spending money on an alternative source of a service. While
this method is imperfect in many ways, the data requirement is not as high as the
other options and the methodology more transparent, making it more tractable to
and clearer to future users of REM.
This method can provide a reasonable basis for our CNSE estimates, but it is im-
portant to understand what it misses. This valuation will not reveal the full cost of
non-served energy, because it cannot capture the additional value provided by the
advantages of the electricity-based service over the alternative. In the case of light-
bulbs vs. kerosene, this method misses advantages like better quality of light and
reduced pollution. It also will not capture macroeconomic effects like impacts on
GDP growth.
Here it is important to note that electricity itself is not a product - it is a way
of providing a service. Typically, people place value on the service that electricity
provides, not on the kWh themselves. Thus, determining the value of a non-served
kWh requires converting that kWh into units of service, whether that service is hours
of lighting, volume of processed agricultural goods, or water pumped. This conversion
will inherently make our estimation dependent on the technology used to provide
electricity-based services (just as the calculation of the cost of a kerosene backup
depends on the efficiency of the lantern).
When electricity is provided to those who did not have it, lighting (and cell-phone
charging) is often the first service used, indicating that it is the highest-value service
provided by electricity1 . We can identify a floor for the value of this electricity service
(in $/kWh) by identifying (1) how much money people spend on kerosene to replace
'It is possible that other services are more valuable, but the equipment required also has a very
high up-front cost and they are budget-constrained and/or cash-flow and debt-constrained.
164
Luminosity: Incandescent CFL Wattage LED Wattage
Wattage
415 Lumens 40W (10.4 1/W) 8W 4.6 W
(can provide 376-500 (5W bulb is 450 1)
lumens) (51.9 l/W) (90 1/W)
710 Lumens 60W (11.9 1/W) 11 W-14 W ~7.8 W
(50.7 - 64.5 1/w) (90 l/W)
Table B.1: Wattages of several types of bulbs
the lighting service and (2) how many kWh are needed to provide that service with
electricity.
Cost of kerosene lighting
The cost of kerosene lighting can be calculated in a straightforward way with a few
assumptions, laid out below and iin the calculations section.
Kerosene costs in India are subsidized, to Rs. 15/liter for the first 4 liters/month,
above which it costs Rs. 25/liter (this is, of course, ignoring the existence of any
black market, although we have been told that such a market is quite alive and well)
[23].
Efficiency of electric lighting
Answering this question is fairly straightforward; we can observe the wattage of the
bulbs used to provide light. However, as bulbs differ, our ultimate valuation of non-
served energy will now depend on the technology employed. Many LED bulbs avail-
able have efficiencies of approximately 90 lumens/watt, and come in many wattages
including 2W, 5W, 1OW and above. Table B.1 compares the wattage by CFLs 2 and
LEDs3 needed to match a standard 40W or 60W incandescent bulb.
It is important to note that CFLs and, especially, LEDs may have too high an upfront
cost to be desirable for many poor Indian families: incandescent commonly cost 10-
2CFL and incandescent data based on data from: CFL Bulbs and Fluorescent Tubes buy-
ing guide" by Bijili Bachao. https://www.bijlibachao.com/lights/cfl-bulbs-and-fluorescent-tubes-
buying-guide.html
3LED data based on data from Phocos indoor LED lamps and products available on earth-
LED.com
165
15 rupees, while CFLs and LEDs can be hundreds of rupees (though those costs are
coming down rapidly)[571. We will set aside the issue of upfront cost for now and
focus on running costs, though they could prove an important consideration as the
cash-flow limitations of poor Indians. Even though the more expensive bulbs have
commensurately long lifetimes, the time-horizon of consumers' decision-making may
be shorter than that lifetime (i.e. their discount rate is very high) due to their budget
limitations.
Calculating CNSE
As mentioned above, what consumers care about is not the kWh purchased but the
service provided. In order to use kerosene costs to value that service, we must decide
on which units to value that service in. Two candidates are lumens and lighting
hours. Since humans do not perceive lumens linearly (Home lighting is around 30-
300 lumens/sq meter, while direct sunlight is around 100,000 lumens/sq meter 4),
lighting hours is a more reliable indicator. This method will miss the benefits from
increased lighting quality and total output with an electric bulb, but more accurately
reflects the metrics on which Indian consumers will be experiencing the world and
making decisions. To calculate CNSE by using hours of lighting service, I use the
following equation:
CNS E ( $ __ 1000 Wh $ liters kerosene * # lamps used
\kWh) 1 kWh Liter hr watts, light bulb
Mills (2003) observes that kerosene lamps use anywhere from 0.005 - 0.42 liters/hr,
and uses 0.01 liters/hr as a midpoint for his calculations. He observes that they
produce 7.8 - 67 lumens, and uses 7.8 for his calculations [35]. The low end is what
he observed with a lamp purchased in a rural market in Vietnam, the high end is
from a hurricane lamp purchased in a US hardware store. I use both values to show
the possible range.
Assuming a family uses two kerosene lamps for 4 hours per day each month, this
4Table 16.1 from Schubert, Fred E., "Light-Emitting Diodes" 2006. 2nd ed.
https://www.ecse.rpi.edu/~schubert/Light-Emitting-Diodes-dot-org/Sample-Chapter.pdf Accessed
March 25, 2016
166
Incandescent CFL LED
$1.84 $0.37 $0.20
Table B.2: Calculated CNSE values
amounts to 2.4 liters/month, for 36 Rs/mo ($0.54). The possible range using Mills'
efficiencies of 0.005-0.42 liters/hr comes gives a range of 1.2 L/mo (Rs 18, $0.27) up
to 100.8 L/mo (1Rs. 2460, $37.05 ).
We have commonly heard that a family's expenditure on kerosene in India can be
around 200 Rs/mo, so I use kerosene lamp efficiencies that result in that expenditure
level. I also assume that 2 kerosene lamps are used in place of one light bulb, and
that lighting is required for 5 hours/day. The level of efficiency which achieves that
is 0.03 liters/hr, well within the range of common values observed by [35]. Table B.2
shows the resulting values of CNSE ($/kWh), for each of the types of light bulb that
might be used. In rupees, this is a range of 27.6-3.1 Rs/kWh. This is a very wide
range and it does not account for the improved lighting quality provided by electric
bulbs. For comparison, a household paying Rs120/mo for two CFLs which they use
for 6h/day is paying an effective rate of 42 Rs/kWh (these numbers are similar to
what we have heard for systems run by Mera Gao, without the cell phone charger).
167