Research On Job-Shop Batching Scheduling Optimization Based On Deep Reinforcement Learning

Shop scheduling is the core technology of manufacturing production management.With the development of economy and the improvement of production technology,the production mode of multi-variety and small batch has gradually become the mainstream,which puts forward higher requirements for the workshop scheduling technology of manufacturing enterprises.The job shop batch scheduling problem breaks through the constraint of a single number of jobs in the traditional job shop scheduling,and is closer to the actual job shop production situation.Therefore,the research on this problem has important practical significance.The problem needs to solve two sub-problems: job batching and sub-batch scheduling,which has huge solution space.Traditional methods have defects such as low solution quality and poor efficiency when solving this problem.At present,deep reinforcement learning has been successfully applied to solve combinatorial optimization problems,so this paper tries to introduce deep reinforcement learning algorithm to solve this problem,and the specific work is as follows:(1)Aiming at the sub-batch scheduling problem,a deep reinforcement learning algorithm based on graph neural network is designed.According to the characteristics of the sub-batch scheduling problem,the Markov decision process of the problem is established.Aiming at the limitations of the workshop state representation,a graph node embedding method was proposed to effectively extract the state features of the disjunctive graph.Based on deep reinforcement learning,a scale-independent agent and training scheme are designed to deal with the sub-batch scheduling problem under different batching schemes.In order to verify the solution quality,efficiency and generalization performance of the algorithm,the model is tested on the generated data set and the public data set.The test results show that the solution quality of the proposed algorithm is better than that of heuristic scheduling rules on all problem instances.When the model trained on small and medium-sized problem instances is directly applied to solve large-scale problem instances,it can still obtain better solutions than heuristic scheduling rules,which verifies the excellent generalization performance of the algorithm.In terms of solving efficiency,the proposed algorithm is slightly slower than the heuristic scheduling rules,but it can give better scheduling strategies in seconds and has better time responsiveness.(2)Combined with the above deep reinforcement learning model,a hybrid genetic algorithm is designed to solve the job shop batch scheduling problem by means of hierarchical iterative optimization.According to the characteristics of the job batching problem,the chromosome coding method,crossover and mutation operators are designed,and the elite retention mechanism and restart mechanism are introduced to make up for the shortcomings of traditional genetic algorithm in local search and effectively avoid the problem of easy to fall into local optimum prematurely.Finally,an example was given to verify the solution quality of the algorithm,analyze the impact of different batching strategies on the scheduling results,and prove the effectiveness of the consistent batching strategy.(3)According to the functional requirements of job shop scheduling,a job shop scheduling system with production scheduling as the core is designed and implemented.The job shop batching scheduling algorithm based on deep reinforcement learning is integrated into the system,and a variety of scheduling schemes are provided,and the scheduling results are comparable,so that the job shop scheduler can schedule the job shop scheduling tasks and improve the production efficiency.