Deep reinforcement learning for large-scale flexible Job-shop dynamic scheduling problem

Aiming at the dynamic scheduling problem of a large-scale flexible job shop, a dynamic scheduling model of large-scale flexible job shop based on Double Deep Q-Network (DDQN) is constructed by taking the minimum expected completion time as the optimization goal and considering two dynamic factors of new job arrival and stochastic processing time. Firstly, the Markov decision process model for large-scale flexible job-shop dynamic scheduling is constructed, and the state space, action space and reward function are designed reasonably. Secondly, to overcome the problems of long optimization time, slow response to dynamic interference and insufficient real-time optimization ability of traditional metaheuristic algorithms, a real-time scheduling method framework based on the DDQN algorithm is designed to solve the dynamic scheduling problem of large-scale flexible job shop, so as to improve the solution efficiency. The experimental results show that the proposed method is superior to the traditional method based on composite scheduling rules in solving large-scale flexible job-shop dynamic scheduling problem, and has strong adaptability and effectiveness.