A Reinforcement Learning-Based Scheduling Algorithm for Special Material Transportation

Efficient scheduling of special materials is essential for improving the operational efficiency of material transportation systems. However, multiple processing stages, heterogeneous transportation resources, and complex transfer paths pose significant challenges to efficient scheduling. To address these challenges, this paper proposes an improved reinforcement learning (RL)-based scheduling algorithm. First, a scheduling optimization model is established with the objectives of minimizing makespan and balancing resource utilization. Second, an improved action-value update strategy is developed to reduce Q-value estimation bias, thereby improving policy convergence and scheduling performance. Experimental results show that the proposed Improved DQN outperforms greedy, genetic, and standard DQN algorithms across different task and resource scales. In particular, it achieves an average relative error rate of 27.48% with 500 tasks and a maximum error rate of only 6.51% under different resource configurations, demonstrating its effectiveness and scalability for complex special material scheduling.