Learning for Dynamic Heterogeneous Vehicle Routing under Traffic Disruptions

Zhuohang Ying

Co-Presenters: Individual Presentation

College: College of Business and Public Management

Major: BS.FINANCE

Faculty Research Mentor: Dan Liu

Abstract:

Title: Deep Reinforcement Learning for Dynamic Heterogeneous Vehicle Routing under Traffic Disruptions
Author: Zhuohang Ying, Department of Finance, Kean University

Abstract:
Vehicle routing is an important problem in transportation and logistics systems. Many existing studies on the heterogeneous capacitated vehicle routing problem (HCVRP) assume that travel conditions are static and known in advance. However, in real-world traffic environments, travel times are often affected by dynamic factors such as traffic signals, congestion, accidents, and unexpected road closures. These factors can significantly change vehicle travel times and make routing decisions more difficult.

This study explores a dynamic heterogeneous vehicle routing problem under changing traffic conditions. The focus is on how traffic disruptions, including red lights, congestion, accidents, and road closures, influence routing performance. We model the problem using a dynamic decision-making framework and apply deep reinforcement learning to learn routing strategies that can adapt to time-varying traffic conditions. The model considers vehicles with different capacities and makes routing decisions step by step based on current traffic states.

A series of simulation experiments are conducted to evaluate the proposed approach under different traffic scenarios. The results show that routing methods that consider dynamic traffic conditions perform better than static routing strategies, especially in environments with frequent disruptions. Vehicles using the dynamic approach are able to adjust their routes more effectively and reduce overall travel time.

These findings suggest that incorporating dynamic traffic information into vehicle routing decisions can improve solution quality and robustness. This research highlights the importance of considering real-world traffic conditions in vehicle routing problems and shows the potential of deep reinforcement learning for solving dynamic routing problems in realistic transportation systems.

Keywords: Dynamic Vehicle Routing, Heterogeneous Fleet, Traffic Congestion, Deep Reinforcement Learning, Transportation Systems