Understanding and Optimizing Equity-Of-Travel in Demand-Responsive Transit Services: Exact Algorithm and Learning-Based Method

Demand-Responsive Transit (DRT) is a flexible on-demand transit solution that complements fixed-route transit services by serving door-to-door travel needs with flexible routes and schedules. The DRT commonly takes the form of paratransit and microtransit and is especially useful in lower-density areas, which support the travel needs of disadvantaged populations such as senior citizens, people with disabilities, and low-income households. Considering the diverse travel needs and different levels of service received among heterogeneous populations, this dissertation explores the overlooked aspect of travel equity--Equity of Travel (EoT)--within DRT operations, particularly focusing on static, reservation-based services and contributes to promoting equitable DRT services with mathematical models and solution algorithms. Specifically, this dissertation identifies and addresses equity issues in DRT operations by first mining historical trip data of DRT services to highlight the EoT issues in real-world operations, followed by the development of exact and learning-based optimization models validated by comprehensive real-world instances. The dissertation is organized into three major studies: (1) The first evaluates EoT metrics, using real-world data to analyze service equity during both normal operations and the COVID-19 pandemic. This analysis led to the development of pandemic-specific EoT metrics to better assess and address these exposure risks. (2) The second study introduces the Equitable Dial-a-Ride Problem (EDARP), formulated as a bi-objective optimization model that balances minimizing travel costs with EoT goals. Utilizing the Branch-Cut-and-Price algorithm and novel ride-time-related resource calibration methods, this study advances solution strategies for real-world DRT scenarios, enhancing service equity effectively. (3) The third study presents the Branch-and-Price with Neural Cuts (BP&NeuCuts) algorithm, integrating Graph Neural Networks with traditional Branch-and-Price techniques to accelerate the optimization process. This approach effectively narrows the decision space and improves scalability and efficiency in handling complex Dial-a-Ride problems within DRT systems, evidenced by computational improvements in test cases. The dissertation concludes by demonstrating the practical impact of these methodologies in refining DRT operations to better serve equitable access, integrating both theoretical advancements and operational applications. This consideration seamlessly connects both planning and operational dimensions, integrating quantitative and qualitative perspectives to pave the way for a more equitable DRT system.