Research On Deep Reinforcement Learning Methods For Solving Flowshop Scheduling Problem

As one of the most widely studied combinatorial optimization problems,job scheduling problem is an abstract model of practical problems from the fields of transportation,logistics,and factories.In combinatorial optimization problems,the complex constraints and the huge solution space increase the difficulty of theirs solving.Traditional solution methods have some limitations.Reinforcement learning,as the most concerned field of machine learning recently,has successful applications in many fields.Therefore,in order to explore the application of artificial intelligence algorithms in practical production,this paper focuses on deep reinforcement learning algorithms to solve the permutation flowshop scheduling problem,a typical combinatorial optimization problem.The main work of the paper is as follows:(1)The actor-critic model is established to solve the permutation flow shop scheduling problem.Firstly,the mathematical model of the permutation flow shop scheduling problem is established according to the constraints and objective function.The essence of the problem is to determine the scheduling order of the jobs,which can be abstracted as the sequential decision problem.In order to solve the scheduling problem,this paper introduces two neural network models to help make the sequence decision.One is a pointer network,which is a special encoder-decoder model.The other is an attention network which combines the Transformer and pointer network.In order to solve the discussed problem,a deep reinforcement learning framework of actor-critic is designed based on these two sequential decision models.(2)The heterogeneous actor-critic model is proposed.According to the characteristics of the permutation flow shop scheduling problem and the task characteristics of different modules in the model,this paper analyzes the functional characteristics of the existing network,and designs a heterogeneous actor-critic algorithm.In the part of model training,the ?-greedy strategy is used to improve the exploration ability of the model while keeping the training effect of the model.Finally,this paper combines the 2-opt algorithm into the model to further improve the solution ability of the model.Experiments show that the actor-critic algorithm proposed in this paper is efficient and superior in solving the permutation flow shop scheduling problem.Compared with the traditional methods,especially the metaheuristic algorithm,the solution of this proposed model is very close and even exceeds the metaheuristic algorithm in some cases.