A novel shuffled frog-leaping algorithm with reinforcement learning for distributed assembly hybrid flow shop scheduling

Distributed hybrid flow shop scheduling (DHFS) problem has attracted much attention in recent years; however, DHFS with actual processing constraints like assembly is seldom considered and reinforcement learning is hardly embedded into meta-heuristic for DHFS. In this study, a distributed assembly hybrid flow shop scheduling (DAHFS) problem with fabrication, transportation and assembly is considered and a mathematic model is constructed. A new shuffled frog-learning algorithm with Q-learning (QSFLA) is proposed to minimise makespan. A three-string representation is used. A newly defined Q-learning process is embedded into QSFLA to select a search strategy dynamically for memeplex search. It is composed of four actions based on the combination of global search, neighbourhood search and solution acceptance rule, six states depicted by population evaluation on elite solution and diversity, and a newly defined reward function. A number of experiments are conducted. The computational results demonstrate that QSFLA can provide promising results on the considered DAHFS.

Automated summary

This study considers the distributed assembly hybrid flow shop scheduling (DAHFS) problem with actual processing constraints and proposes a new algorithm with reinforcement learning to minimize makespan.