Logistics Network Topology Optimization and Transportation Scheduling Using Graph Neural Networks

Abstract As the logistics industry faces growing challenges, effectively using advanced intelligent algorithms to optimize logistics network topology and transportation scheduling is crucial. This paper presents the Adaptive Attention Mechanism Deep Reinforcement Learning (AAM - DRL) algorithm. By integrating Graph Neural Networks (GNNs) and reinforcement learning, AAM - DRL efficiently models node - edge relationships in logistics networks. It dynamically adjusts the importance of state - action pairs via the attention mechanism, accelerating learning and enhancing policy performance. Experiments show AAM - DRL outperforms traditional algorithms (like DQN, DDPG, A3C, and PPO) in convergence speed, policy stability, robustness, and multi - objective optimization. Mathematically, it reveals the attention mechanism's key role in policy optimization, enriching the theoretical research on logistics network optimization and complex - network - environment reinforcement learning. In practice, AAM - DRL accurately optimizes logistics network topologies, efficiently schedules transportation, cuts costs, boosts efficiency, and copes well with complex logistics environments, offering high practical value.