A HYBRID APPROACH BASED ON META-HEURISTICS AND MACHINE LEARNING TO IMPROVE COMPUTATIONAL EFFICIENCY IN JOB SHOP SCHEDULING PROBLEMS

Recent advances show the success of meta-heuristics in solving job shop scheduling problems. Machine learning-based methods emerge as catalysts for traditional approaches. Thus, investigating the combination of these techniques has become relevant, aiming to improve the quality of the solutions and the computational efficiency. This paper proposes a hybrid algorithm called K-means Particle Swarm Optimization (KPSO) to optimize makespan in the classic flexible job shop problem. Initially, the chaotic Singer map is used to generate diverse and well-distributed initial solutions. Therefore, K-means bundles these solutions and uses centroids as a starting point for the PSO, with the aim of accelerating convergence and improving outcomes. Tests in ten instances of the literature indicate that, in six of them, the makespan was reduced and, in 70% of the cases, the average makespan was improved. It is concluded that the use of clustering adds value and suggests consistency in the proposed algorithm.