| dc.description.abstract | Urban public transit is an important component of transportation systems and plays a critical role in providing mobility in many metropolitan areas. However, with aging systems, continuous expansion, and near-capacity operations, transit systems are susceptible to unplanned delays and service disruptions caused by equipment, weather, passengers, or other internal and external factors, resulting in great inconvenience for passengers and economic loss for operators. Ensuring good service provision during service disruptions is important for public transit management.
Resilience is an important concept related to incidents. It usually refers to the ability of an entity to return to its initial conditions after it is disturbed. Since monitoring, control, and planning are the three major tasks for public transit system management, we define the resilience of a public transit system as the ability to monitor, control, and plan for incidents (service disruptions) in ways to mitigate congestion, improve travel efficiency, and reduce safety risks.
This dissertation focuses on the first two tasks to improve the resilience of public transit operation in light of disruptions that regularly take places. Specifically, we aim to 1) understand the impact of unplanned incidents on PT systems (i.e., monitoring); and 2) design mitigating strategies to relieve incident impacts (i.e., control). The specific topics we cover in the thesis can be categorized by a two-by-two matrix. The first dimension considers short-term (e.g., less than a couple of minutes) v.s. long-term (e.g., more than 1 hour) incidents while the second dimension considers monitoring and control tasks. Five different studies under the umbrella of this two-by-two matrix are presented.
The first study evaluates a transit system performance under random short-term service suspensions using a bulk-service queue model. We prove that under random suspensions, headways can be represented as the difference between two compound Poisson exponential variables. Assuming no vehicle overtaking, we approximate the headway as a zero-inflated truncated normal distribution to obtain a closed-form moment generating function (MGF). Based on the MGF, we derive the system stability conditions and the mean and variance of queue length and waiting time at each station with analytical formulations. The second study provides an empirical analysis of the impact of service disruptions. We use a real-world train collision incident at the Chicago Transit Authority (CTA) system to analyze the impact of unplanned long-term incidents on the system's demand, supply, and passenger behavior. We also propose a redundancy index to quickly identify alternative capacity in CTA under service disruptions. The third study proposes a probabilistic method to infer passengers' behavior (e.g., waiting, switching to another line, transferring to a bus) under disruptions. The main contribution is a probabilistic model to recognize whether an observed smart card record (e.g., transfer to a bus stop) is a normal behavior or due to the incident. This model allows us to extract the actual behavioral responses and outperforms the typical rule-based methods. The fourth study proposes a station-based path recommendation model to reduce the total system travel time during disruptions. We use a robust optimization-based formulation to address the demand uncertainty. The closed-form robust counterpart is derived. To tackle the lack of an analytical formulation of travel times due to left behind, we propose a simulation-based first-order approximation to transform the original problem into a linear program and solve it iteratively with the method of successive average. The fifth study proposes an individual-based path recommendation model with the objective of minimizing total system travel time and respecting passengers’ path choice preferences. Passengers’ behavior uncertainty in path choices given recommendations and travel time equity are also considered in the formulation. We model the behavior uncertainty based on passenger’s prior preferences and the posterior path choice probability distribution with two new concepts: epsilon-feasibility and Gamma-concentration, which control the mean and variance of path flows in the optimization problem. We show that these two concepts can be transformed into linear constraints using Chebyshev’s inequality. The individual path recommendation problem with behavior uncertainty is efficiently solved using Benders decomposition. Finally, we use a post-adjustment heuristic to address equity requirements.
Future research directions and potential applications of the work are discussed in the last chapter. | |
