Research On Workshop Scheduling Based On Genetic Algorithm With Reinforcement Learning

Since 2013,with the "Industrial 4.0" in Germany and "China Intelligent Manufacturing 2025" successively putting forward,manufacture has ushered in a new turning point of development.The workshop scheduling is one of the important components of intelligent manufacturing.The rapid development of manufacturing industry is inseparable from the advancement of intelligent scheduling methods.Genetic algorithm has important value for researching workshop scheduling problems,but its performance is susceptible to parameters.Therefore,we study the genetic algorithm to solve the shop scheduling problem,and design a genetic algorithm based on reinforcement learning to improve the performance of the algorithm in this paper,and solve the shop scheduling problem efficiently.Firstly,a genetic algorithm based on reinforcement learning is proposed.At the beginning,a parallel strategy is designed.K-means is adopted to cluster the population according to individual chromosome,maximizes the separation of individuals,and evenly divides the population into several sub-populations.Then,a parameter self-learning strategy is designed.By using Q-Learning of reinforcement learning,self-learning the crossover probability of genetic algorithm,and the better crossover probability is adapted to the evolution of contemporary population.After that,to realize the communication among subpopulations,a communication mechanism is set up,and multi-strategy idea is combined to improve the efficiency of algorithm operation.Finally,nine functions are selected to test the algorithm.Compared with the traditional genetic algorithm and the general parallel genetic algorithm,the results show that the proposed algorithm has a significant effect on function optimization.Secondly,the proposed algorithm is applied to permutation flow shop scheduling problem.According to the characteristics of the problem,the job coding is adopted.Because the coding method of flow shop scheduling problem is relatively simple,the communication rule is replaced by dynamic updating of sub-populations.According to the characteristics of the population,Dynamic updating subpopulation strategy is implemented,followed by the crossover probability learning process based on reinforcement learning.Then,some standard instances are used to verify the algorithm.Compared with other algorithms such as standard GA,the results show that the algorithm has significant advantages in solving flow shop scheduling problems.Then,according to the characteristics of job shop scheduling problem,the proposed algorithm is improved,and the process-based gene string coding is adopted.Because the job shop scheduling problem is more complex than the flow shop scheduling problem,the initial sub-population communication mechanism is adopted to improve the time efficiency.Finally,the experiments are carried out based on standard examples and compared with standard GA and general parallel genetic algorithm,the results show that the experimental results are effective and better.Finally,the main work of this paper is summarized,and the future directions worthy of further study are prospected.