Learning-based multi-objective hyper-heuristic algorithm for reconfigurable assembly line scheduling problems

Reconfigurable assembly lines have emerged as a vital manufacturing paradigm to meet the growing demand for customized and multi-variety products. This study considers the reconfigurable assembly line scheduling problem, involving product sequencing optimization, to minimize reconfiguration cost, production workload equalization, and logistics leveling simultaneously. This study formulates a novel and linearized multi-objective mathematical model, which rectifies deficiencies in prior formulations. A novel Q-learning-based multi-objective hyper-heuristic algorithm is proposed. The algorithm integrates multiple metaheuristic operators, including particle swarm optimization, teaching-learning-based optimization, whale optimization algorithm, and grey wolf optimizer, within a unified search framework. Q-learning is employed to adaptively select the most promising operator at each search stage based on real-time performance feedback. Moreover, the proposed algorithm incorporates a new density-aware leader selection strategy with a survival-time decay factor to select the global best solution for population evolution, favoring superior solutions in sparse regions and increasing selection pressure on high-quality individuals. A numerical case study demonstrates that the models with the ε-constraint method could achieve a set of Pareto solutions. A computational study on 120 generated benchmark instances demonstrates that the proposed methodology outperforms nine other high-performing multi-objective algorithms.