Multi-manned disassembly line balancing problems for retired power batteries based on hyper-heuristic reinforcement

Abstract

Considering the complexity and uncertainty of retired power batteries, worker disassembly can reduce the cost and increase the flexibility of the disassembly process. To address the problem of considerable waste products, we investigate a multi-manned disassembly line balancing problem with different work modes (MWM-DLBP). In addition, we explore the impact of variability in penalizing factors (earliness and tardiness) on costs when products of different scales need to be disassembled. Given the NP-hard of the problem and unpredictable situations, we propose a reinforcement learning-based Hyper-Heuristics (RLHHs) algorithm to solve the practical issues. A Markov model is constructed to generate a sequence of low-level metaheuristic algorithms, and a multi-armed bandit (MAB) method is designed to tune different critical parameters intelligently. The superiority of RLHHs is verified by comparing the optimization results with other algorithms. Further, results reveal that the new disassembly scheme can save an average cost of 9.81% compared to the original system design. Finally, we provide manufacturers with the specific workforce planning and working mode for disassembly tasks under different scenarios.