Research On The Application Of Cooperative Learning Scatter Search Algorithm In Distributed Shop Scheduling

Economic globalization has transformed the manufacturing industry from a single-factory model to a multi-factory collaborative production model.The heterogeneity of manufacturing resources and the diversity of customer demands greatly increase the difficulty of distributed scheduling for multi-factory systems.Distributed manufacturing is an inevitable trend in the development of the manufacturing industry.Distributed shop scheduling focuses on the rational distribution of workpieces among factories and the rational processing sequence within each factory to achieve the optimization of scheduling metrics in a distributed manufacturing environment.Distributed shop scheduling problem is an important class of combinatorial optimization problems in modern manufacturing systems,with higher complexity and greater difficulty in solving,and has significant academic significance and practical value.Scatter search algorithm(SS)adopts a populationbased global search strategy that makes less use of the randomness of the search process and focuses on a series of systematic methods to construct new solutions and improve the concentration and diversity of the search.In recent years,it has received extensive attention from academia and industry and has been widely applied in production scheduling.In this paper,a collaborative learning scatter search algorithm is designed,and the reinforcement learning driven scatter search algorithm is applied to solve the distributed flow shop scheduling problem with different constraints.The main research content and work of this paper are as follows:(1)A two-stage cooperative scatter search algorithm with multi-population hierarchical learning mechanism(TCSSMH)is proposed to overcome the weaknesses of the SS algorithm in weak local search ability and slow convergence speed.Firstly,TCSSMH adopts an adaptive bidirectional selection search strategy based on the elite reference set,which guides population evolution by focusing on elite individuals.Secondly,a multi-population hierarchical learning mechanism is embedded in the reference set updating process.The candidate populations are divided into three levels:dominant,moderate,and inferior,based on the function fitness value.These three sub-populations interactively learn and evolve to balance exploration and exploitation capabilities.Finally,the optimal individuals of each sub-population are obtained through pattern search optimization,which strengthens the ability of local search.The effectiveness of each component in the algorithm is verified through the method of constructing fitness terrain.Experimental results on the CEC2017 benchmark test set show that the TCSSMH algorithm outperforms the SS and its advanced variants.(2)A knowledge-driven cooperative scatter search algorithm(KCSS)with reinforcement learning is designed for the distributed blocking flowshop scheduling problem(DBFSP).Multiple neighborhood structures are designed according to the problem characteristics,and the Q-learning algorithm is combined to adaptively select neighborhood perturbation strategies throughout the entire search process to improve exploration ability and search efficiency.A local search method based on neighborhood reconstruction is proposed to perturb the currently found optimal solution and enhance the development ability of KCSS in the local region.In addition,a path relinking mechanism is introduced in the subset combination method to ensure the diversity of solutions during the optimization process.Finally,the performance of the KCSS algorithm is verified on a benchmark set based on Naderi and Ruiz,and the experimental results demonstrate the robustness and effectiveness of the KCSS algorithm,updating 518 known optimal solutions out of 720 benchmark instances.(3)A cooperative scatter search algorithm based on Q-learning algorithm(QCSS)is proposed to solve the distributed permutation flow shop scheduling problem with sequence-dependent setup times.In the diversified generation phase,two effective heuristic algorithms are designed to construct high-quality and diverse initial populations.In the improvement phase,multiple-population cooperative strategies are designed to enhance global exploration and local exploitation,including cooperative Q-learning algorithm and local enhancement search.The reference set is divided into two subpopulations that coevolve and update the population using Boltzmann learning and ε-greedy strategies,respectively.The interaction and competition between subpopulations improve the global optimization performance.In addition,a restart mechanism is proposed in the reference set updating phase to ensure solution diversity.The performance of the QCSS algorithm is verified on the benchmark set,and the simulation experiment results show that various design operations can effectively improve the performance of the algorithm,and the performance of the proposed algorithm is significantly better than that of the comparison algorithm.