Research on a Dynamic Scheduling Algorithm for Intelligent Logistics Distribution Systems Based on Deep Reinforcement Learning
Dynamic scheduling in intelligent logistics distribution systems involves high-dimensional state representation, stochastic order arrivals, and complex route constraints, which make traditional scheduling methods less effective in real-time environments. To address this problem, this study proposes an improved deep reinforcement learning-based dynamic scheduling algorithm. The method integrates multi-source state information and incorporates an attention mechanism, prioritized experience replay, and graph neural network modeling to enhance feature representation, sample efficiency, and spatial dependency learning. Experiments were conducted on both simulated datasets and real-world delivery data to evaluate the proposed approach in terms of delivery efficiency, routing cost, service quality, and convergence performance. The results show that the proposed method reduces average delivery time to 29.8 minutes and total route length to 442.1 km, while increasing the on-time delivery rate to 91.3%. In addition, the model converges faster and exhibits better training stability than heuristic and traditional reinforcement learning methods. The ablation study further verifies the effectiveness of the key modules. These findings indicate that the proposed method provides an effective solution for dynamic logistics scheduling under complex constraints.