Reinforcement Learning for Emergency Materials Scheduling Based on Pointer Policy Network and Transformer Value Estimation

Sudden disasters are difficult to predict and inevitable, often leading to huge economic losses and casualties. Once disasters occur, there is an urgent need to promptly distribute emergency materials to affected areas. Therefore, the ability to make rapid decisions in emergency materials scheduling significantly impacts the effectiveness of rescue operations. In this paper, we consider a three-level scheduling network for emergency materials scheduling, and use deep reinforcement learning to design a PPO algorithm with a pointer network as the policy network and a transformer model as the value network, and the trained agent is verified to have excellent effect on the test set, which can make reasonable decisions for emergency materials scheduling quickly.