A novel MOEA/D with Q-Learning initialization for dual-resource constrained flexible job shop scheduling problem with limited multi-skilled workers

Abstract

Multi-skilled workers are crucial in the manufacturing sector of customized production environment, which can quickly adjust their work tasks to meet the specific requirements of different products. However, frequently changes in work positions not only reduce production efficiency but also impose burdens on worker. Thus, this paper investigates a dual-resource constrained flexible job shop scheduling problem with limited multi-skilled workers. And a mathematic programing model is established aiming to minimize both completion time and the maximum workload of the workers. To address the proposed model, a novel multi-objective evolutionary algorithm based on decomposition with Q-Learning initialization (MOEA/D-QI) is proposed. In MOEA/D-QI, a Q-learning based adaptive population initialization strategy is introduced to improve the algorithm’s convergence speed. Additionally, a neighborhood micro-population hybrid cross search strategy is designed to enhance exploration and exploitation capabilities. An external archive is established to improve the utilization of historical solutions. After verifying the correctness of the DRCFJSP-LMWT model using the CPLEX solver, the optimal parameter combination of the MOEA/D-QI algorithm was determined through orthogonal experiments, and the effectiveness of each improved strategy was verified by ablation experiments. Finally, the method’s efficacy and superiority are affirmed through extensive experiments involving a significant number of test cases.