An Effective Multi-population Grey Wolf Optimizer based on Reinforcement Learning for Flow Shop Scheduling Problem with Multi-machine Collaboration. (December 2021)

• Record Type: Journal Article
• Authors: Chen, Ronghua; Yang, Bo; Li, Shi; Wang, Shilong; Cheng, Qingqing
• Is Part Of: Computers & industrial engineering. Volume 162(2021)
• Publication Date: 2021-12
• Subjects: Scheduling -- Grey Wolf Optimizer (GWO) -- Multi-machine Collaboration -- Reinforcement learning (RL) -- Multi-Population; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2021.107738
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: The mathematical model of FSSP-MC is proposed. An adaptive Multi-Objective Multi-population GWO based on RL is developed. RL is applied to adaptively adjust the individual numbers of the subpopulations. An improved search mechanism is adopted. Abstract: This paper proposes the Flow Shop Scheduling Problem with Multi-machine Collaboration (FSSP-MC). In FSSP-MC, several machines can operate a single task simultaneously, so it is a coupling problem of resource composition and task sequencing, which is more difficult and has a larger scale solution space than the traditional Flow Shop Scheduling Problem (FSSP), therefore an optimization algorithm with higher efficient and accurate is demanded. However, most existing intelligent algorithms are easily trapped into local optima and have low precision on solving large-scale problems. To this end, an adaptive multi-objective Multi-population Grey Wolf Optimizer (AMPGWO) based on Reinforcement Learning (RL) is developed to address FSSP-MC with the goals of minimizing maximum completion time (makespan) and the total machine load. In AMPGWO, the whole population is divided into three subpopulations, and different search strategies are adopted in different subpopulations to enhance population diversity. Since the numbers of individuals in a subpopulations are pretty crucial for the performance of the algorithm, which needs to be reasonably determined and dynamically adjusted, so RL is applied to adaptively adjust the individual quantity of each subpopulation and strengthen the information exchange among different subpopulations. Finally, 20 instances of FSSP-MC with different sizes are used for three comparative experiments, in which the effectiveness of multi-population and RL mechanisms, effectiveness of mutation mechanism of AMPGWO are verified. Through results analysis, it can be seen that proposed AMPGWO is pretty effective and significantly outperforms its competitors in solving FSSP-MC.