Scheduling Algorithm for Raw Material Transportation Via Deep Reinforcement Learning

Note that the traditional scheduling control algorithms (e.g., Scheduling control algorithm based on mixed integer programming.) depend on the specific scheduling system model, which leads to get the optimal solution difficultly. This paper proposes a novel scheduling algorithm based on deep reinforcement learning, where three parts of raw material transfer scheduling control are included in the algorithm, that is, raw materials are transferred from the cargo ship to the wharf, from the wharf to the factory, and from the factory to the processing location. The algorithm takes the system state as the input of Deep Q-Network, calculates the action value of each part of the scheduling action through the deep neural network, and finally selects the optimal scheduling action based on the action value, which effectively reduces the storage cost and the number of pipeline changes in the process of raw material transportation. In addition, this paper also proposes to optimize the input of neural network through state reuse, which further reduces the switching times of transmission pipeline. Simulation results show that the proposed method does not depend on the specific scheduling model, reduces the solution time, and effectively reduces the storage cost and pipeline switching cost in the scheduling process.