Multimodal transport path planning for New International Land-Sea Trade Corridor of China based on deep reinforcement learning

Abstract

The New International Land-Sea Trade Corridor is located in western China. It connects the Silk Road Economic Belt, the 21st Century Maritime Silk Road, and the Yangtze River Economic Belt, and holds an important strategic position in the regional coordinated development. Aiming at the high-dimensional decision-making problem of multimodal transport path planning for the new corridor, this paper proposes a path planning algorithm that combines a multi-hierarchy heterogeneous graph attention network with deep Q-network (MHGAN-DQN). The algorithm dynamically incorporates the multiple heterogeneous relationships between nodes and edges in the transportation network into the action selection process, significantly enhancing the model's ability to capture key paths. Additionally, the algorithm introduces a node decomposition method to ensure consistency in the reward of action selection, avoiding conflicts with the requirements of the Bellman equation, effectively simplifying the state space management in path planning. Finally, this paper thoroughly examines the multi-objective impact of cost, time, and carbon emission on multimodal transport path selection. It further explores the comprehensive impact of railway freight subsidies, carbon tax policies, and uncertainties in transportation time on the final transportation scheme. The research results significantly improve the accuracy and efficiency of path decision-making, deepen the application of deep reinforcement learning in multimodal transport, and provide decision-making support for policy development related to the sustainable development of logistics systems of the New International Land-Sea Trade Corridor.