Methods for Parameter Estimation with Devices in Microgrids

Abstract

Microgrids may be described as miniaturized, independent, islanded, autonomous electrical networks. Before deploying or building a microgrid, it is informative to simulate its operation. In such a simulation, one must assign parameter values to the device models, and these parameter values may not be easily known for any number of reasons: lack of manufacturing data, inaccessibility to directly measure or infer such parameters, or safety concerns. Using non-invasive measurements, this work seeks to estimate these parameter values in devices which may exist within a low voltage microgrid. Constant power loads, diesel gensets, and solar inverters can all be found in low voltage microgrids. This thesis will discuss each model and seek to find optimal parameters for each device’s operation.

First, two different dynamic constant power loads (DCPLs) are considered. An appropriate model structure is established, and a hybrid algorithm for parameter estimation (HAPE) is introduced to estimate defining parameters in the model. In order to verify the load model and the HAPE, two experiments are conducted with different DCPLs using a Power-Hardware-in-the-Loop (PHiL) testbed. The PHiL testbed consists of a real-time computer working with a programmable power amplifier in order to perturb the input voltage’s amplitude and frequency. The experimental waveforms are used to inform the HAPE. The resulting parameter estimates are used to define simulation models, and the performance of the HAPE is discussed.

Second, a similar approach will be taken to estimate parameters in a model for a diesel genset, not a load. Unlike the first part of this thesis, this second part will implement a similar HAPE, but with some important differences. The HAPE used here will proceed in generations, consider a parameter sensitivity analysis, and be implemented across multiple computing nodes on a supercomputing platform: MIT Supercloud.

Third, a small system is considered: a grid-connected home with rooftop solar power. Unlike the previous two parts of this thesis, this third part will discuss an approach for choosing parameter values in a solar inverter’s simulation model. The solar inverter includes active power filtering functionality in its control strategy to mitigate current distortion at the home’s point of common coupling. Waveforms captured from experimental non-linear loads are included to show how a solar inverter would operate alongside such loads while connected to the utility grid. A Monte Carlo method, implemented on MIT Supercloud in a massively parallel fashion, is used to survey a wide range of parameter values. With results from thousands of simulations, a set of parameters is selected which minimize component size.

In all three parts of this thesis, the models with parameters to be estimated may be described as grey box models. With the model’s structure established, a hybrid algorithm for parameter estimation (HAPE) can be used to repeatedly simulate the model with a candidate set of parameters. The HAPE borrows from some established approaches (Simulated Annealing, Tabu Search, Particle Swarm Optimization) and offers new features. Recent advances in computing have allowed for algorithms to be implemented in a massively parallel fashion. Heuristic approaches are sometimes preferred in simulations that may contain a large number of non-linearities, exhibit non-smoothness, or contain event-based phenomena. Also, heuristic approaches may need many more iterations to reach convergence, however, so the algorithms in this work are implemented on MIT Supercloud. With appropriate device models established, they can be included as part of a larger simulation of a microgrid to more accurately demonstrate its operation.