Optimizing Mobility Infrastructure, Services, and Subsidies in Multimodal Transportation Networks

Cities aiming to improve their transportation networks are integrating emerging mobility options at a rapid pace. These travel modes can provide individuals with greater flexibility to access additional points of interest. However, computing transportation accessibility metrics in networks characterized by multiple travel modes, multiple traveler costs, and individual preferences remains a challenge. Addressing this challenge is critical for optimally designing public policies related to the provision of mobility infrastructure, services, and subsidies. This dissertation describes the development of a network model that facilitates a comprehensive calculation of accessibility. Then, it presents optimization models for the allocation of need-based mobility subsidies and the coordinated design of a multimodal transportation network with shared mobility. Both optimization models are formulated to balance system-level accessibility improvements with fairness considerations.

Chapter 2 introduces NOMAD: Network Optimization for Multimodal Accessibility Decision-making. NOMAD integrates the personal vehicle, transportation network company (e.g., Uber, Lyft), carshare, public transit, personal bike, bikeshare, scooter, walking, and feeder micro-transit modes into a unified network model by way of a “supernetwork” graph topology. A time-dependent generalized travel cost function assigned to each edge in the graph incorporates the following disutility factors: monetary cost, day-to-day mean travel time, (un)reliability as represented by day-to-day 95th percentile travel time, crash risk, and physical discomfort. Movement-based node costs are imposed to either penalize or benefit movement from one edge to another edge via a specific node. NOMAD and its open-source implementation can be used to create multimodal travel cost matrices, which may immediately serve as an input for accessibility analysis and other policy decisions related to shared mobility. Examples of such policy decisions are explored in Chapters 3 and 4.

Chapter 3 presents a mixed-integer linear program to optimize the allocation of need-based mobility subsidies to eligible recipients. This study was inspired by recently-piloted Universal Basic Mobility (UBM) programs that provided all eligible individuals with equal-valued mobility subsidies. The proposed optimization model is framed such that all eligible residents receive a transit pass, with the remaining program budget distributed differentially based on an individual’s geographic residence. Subsidies may be used on transportation network companies, shared scooters, and bikeshare. The optimization model maximizes system-level accessibility to jobs while ensuring a fair distribution procedure. A case study in Allegheny County, PA indicates that mobility subsidies can serve to compensate residents in communities with inferior transit service and network infrastructure. Empirical results reveal that the proposed allocation procedure improves aggregate accessibility and fairness metrics by 27% and 25-fold, respectively, relative to the equal subsidy allocation procedure of prior UBM implementations. Another important result is that the optimal solution can be approximated with high accuracy by an interpretable surrogate regression model based on underlying features of the transportation network.

Chapter 4 describes a bi-level, bi-objective mixed-integer program to jointly select the locations of bike-share stations and microtransit service zones. The upper level objective is a weighted sum of a system-level accessibility metric and the minimum value of the accessibility metric computed for any origin. The lower level solves the combined mode split and traffic assignment problem for a multimodal system with shared mobility. Since the relationship between the lower and upper level decision variables cannot be expressed in closed form, an iterated local search heuristic is used as the solution method. A case study in Allegheny County, PA illustrates the benefit of joint optimization. When bikeshare and microtransit investments are optimized together for a specific set of model parameters, system-level accessibility increases by 4.0 percentage points relative to the transit-only baseline scenario, and minimum-level accessibility rises substantially by 8.7 percentage points. A key insight from these results is that bikeshare and microtransit individually provide limited improvements for the most disadvantaged area, but their impact magnifies when deployed together. Experiments also reveal diminishing marginal returns of the bikeshare budget on the combined objective value due in large part to diminishing returns on the minimum-level accessibility metric. These findings suggest that the constraints of the existing public transit network limit the extent to which shared mobility can improve accessibility in the worst-off residential areas. Finally, Pareto optimization illustrates the importance of balancing system and fairness objectives to ensure that the most disadvantaged areas are not left behind in pursuit of broader system-wide gains