MRLM: A meta-reinforcement learning-based metaheuristic for hybrid flow-shop scheduling problem with learning and forgetting effects

• Published in: Swarm and evolutionary computation 2024-03, Vol.85, p.101479, Article 101479
• Main Authors: Zhang, Zeyu, Shao, Zhongshi, Shao, Weishi, Chen, Jianrui, Pi, Dechang
• Format: Article
• Language: English
• Subjects: Hybrid flow-shop scheduling; Learning effect and forgetting effect; Meta-learning; Metaheuristic; Q-learning
• ISSN: 2210-6502
• DOI: 10.1016/j.swevo.2024.101479

In the real-world production environment, the employee skills affect production efficiency. Especially, the learning and forgetting effects largely influence the processing time. This paper investigates a hybrid flow-shop scheduling problem with learning and forgetting effects (HFSP-LF). Two learning and forgetting effects models are constructed. The sequence-dependent setup time (SDST) is also considered. The objective is to minimize the makespan of all jobs. To solve such problem, a mixed integer linear programming (MILP) model is built to formulate HFSP-LF. Then, a meta-reinforcement learning-based metaheuristic (MRLM) is proposed. In MRLM, a constructive heuristic is employed to generate the initial solution. Several problem-specific search operators are developed to explore and exploit the solution space. The search framework of MRLM comprises a meta-training phase and a Q-learning-driven search phase. In the meta-training phase, the search operators are trained to obtain prior knowledge of their selection and an initial learning model is constructed. In the Q-learning-driven search phase, Q-learning is employed to implement automatic selection of search operators through continuously perfecting the learning model and absorbing the feedback information of searching. Finally, we conduct a comprehensive experiment. The experimental results demonstrate that the designs of MRLM are effective, and MRLM significantly outperforms several well-performing methods on solving HFSP-LF. •A hybrid flow-shop scheduling problem with learning and forgetting effects is studied.•A mathematical model is presented to format the considered problem.•A meta-reinforcement-learning search framework is proposed.•The proposed methods obtain better results.