Multi-objective cooperative co-evolution algorithm with hypervolume-based Q-learning for hybrid seru system

Abstract

The hybrid seru system (HSS), which is an innovative production pattern that emerges from real-world production situations, is practical because it includes both serus and a flow line, allowing temporary workers who are unable to complete all tasks to be assigned to the flow line. We focus on the HSS by minimising both makespan and total labour time. The HSS includes two complicated coupled NP-hard subproblems: hybrid seru formation and hybrid seru scheduling. Thus, we developed a multi-objective cooperative co-evolution algorithm with hypervolume-based Q-learning (MOCCHVQL) involving hybrid seru formation and scheduling subpopulations, evolved using a genetic algorithm. To achieve balance between exploration and exploitation, a hypervolume-based Q-learning mechanism is proposed to adaptively adjust the number of non-dominated hybrid seru formations/scheduling in co-evolution. To reduce computational time and enhance population diversity, a population partitioning mechanism is proposed. Extensive comparative results demonstrate that the MOCCHVQL outperforms state-of-the-art algorithms in terms of solution convergence and diversity, with the hypervolume metric increasing by 22 % and inverse generational distance metric decreasing by 76 %. Compared with a pure seru system (PSS), the HSS can significantly reduce training tasks, thereby conserving the training budget. In scenarios with fewer workers and more batches, a positive phenomenon, where the HSS significantly decreases the training tasks relative to PSS while only slightly increasing the makespan, was observed. In specific instances, the HSS reduced the number of training tasks by 50 %, while only increasing the makespan by 10.5 %.