DHL: Deep reinforcement learning-based approach for emergency supply distribution in humanitarian logistics

Alleviating human suffering in disasters is one of the main objectives of humanitarian logistics. The lack of emergency rescue materials is the root cause of this suffering and must be considered when making emergency supply distribution decision. As large-scale disasters often cause varying degrees of damage to different influenced areas, which will cause differences in both human suffering and the demand for emergency supply in influenced areas. This paper considers a novel emergency supply distribution scenario in humanitarian logistics, which takes into account these differences. In the scenario, besides the economic goals such as minimizing costs, the humanitarian goal of alleviating the suffering of survivors is treated as one of the main bases of emergency supply distribution decision making. We first apply Markov Decision Process to establish the formulation of the emergency supply distribution problem. Then, to acquire the optimal resource allocation policy that can reduce the economic cost while decreasing the suffering of survivors, a Deep Q-Network-based approach for emergency supply distribution in Humanitarian Logistics (DHL) is developed. Numerical results demonstrate DHL has better performance and lower time complexity to solve the problem by comparing with other baselines.