Strategizing Equitable Transit Evacuations: A Data-Driven Reinforcement Learning Approach

TL;DR

This work tackles equitable transit evacuations under natural disasters by formulating an MDP for bus-based evacuations and solving it with a PPO-based controller that incorporates an equity penalty. The equity term relies on a point-biserial index to prioritize equity-priority communities while minimizing total evacuation time, enabling a data-driven balance between efficiency and fairness. Extensive simulations on a San Francisco Bay Area network, using GTFS and OpenStreetMap data, demonstrate that the Equity-RL framework achieves substantially lower inequity (lower |r_pb|) and more equitable service distribution with competitive evacuation times compared to stochastic and rule-based baselines. The results suggest a scalable, real-time capable approach to emergency management that can inform urban resilience planning and improve equitable access to evacuation resources.

Abstract

As natural disasters become increasingly frequent, the need for efficient and equitable evacuation planning has become more critical. This paper proposes a data-driven, reinforcement learning-based framework to optimize bus-based evacuations with an emphasis on improving both efficiency and equity. We model the evacuation problem as a Markov Decision Process solved by reinforcement learning, using real-time transit data from General Transit Feed Specification and transportation networks extracted from OpenStreetMap. The reinforcement learning agent dynamically reroutes buses from their scheduled location to minimize total passengers' evacuation time while prioritizing equity-priority communities. Simulations on the San Francisco Bay Area transportation network indicate that the proposed framework achieves significant improvements in both evacuation efficiency and equitable service distribution compared to traditional rule-based and random strategies. These results highlight the potential of reinforcement learning to enhance system performance and urban resilience during emergency evacuations, offering a scalable solution for real-world applications in intelligent transportation systems.

Figures (9)

• Figure 1: The framework of the proposed methodology using reinforcement learning. The RL agent (controller) interacts with the environment, which represents the transportation network. At each time step $t$, the controller takes an action $a_t$ based on the current state $S_t$. The environment then transitions to a new state $S_{t+1}$, providing feedback to the controller.
• Figure 2: Illustrative example of a six-node network. This figure presents a network comprising six nodes interconnected by directed links, illustrated by the lines with arrows indicating direction. Nodes are denoted as circles, while buses are depicted as square blocks positioned on the links. The numbers on the links represent travel times. Red numerals indicate the demand of evacuees at origin nodes, and blue numerals represent the capacity of shelters.
• Figure 3: Space-time trajectory for feasible solutions. In this graph, the trajectories of Bus 1 are illustrated with orange lines and those of Bus 2 with blue lines. Different trips are represented by different shades of the same color. Each vector in the diagram is annotated with the status of all buses, represented as $s_b = \{c(b), p(b), d(b)\}$, where each parameter denotes the capacity of the bus, the number of evacuees on board and the next destination, respectively.
• Figure 4: Optimal solution for six-node network
• Figure 5: Visualization of the San Francisco Bay Area transportation network from OpenStreetMap2 .