DQL-assisted competitive evolutionary algorithm for energy-aware robust flexible job shop scheduling under unexpected disruptions

Abstract

Energy-aware scheduling has emerged as a well-defined research topic for achieving sustainable development. However, unexpected disruptions in workshop pose several challenges to the manufacturing process, causing the original schedules to become non-optimal or even infeasible, resulting in significant energy waste. Consequently, it is necessary to investigate the robust scheduling problem with energy-awareness. The energy-efficient robust flexible job shop scheduling problem (ERFJSP) aims to simultaneously optimize scheduling efficiency, total energy consumption, and robustness. A two-stage mixed-integer linear programming model considering machine breakdown and job insertion is formulated in this paper for the first time. To solve this problem, a double Q-learning (DQL)-assisted competitive evolutionary algorithm (DQCEA) is proposed. In DQCEA, a heuristic initialization strategy is first designed, which allows it to obtain high-quality and diverse solutions. Subsequently, a multi-objective competitive mechanism is proposed to classify search individuals into inferiors and superiors. Moreover, an inferior-pushing and superior-pulling-based crossover is designed for inferior members, facilitating knowledge transfer within the population and enhancing global diversification capability. Meanwhile, a hyper-mutation operator is devised for superior members, which incorporates eight search strategies to improve local intensification ability. Furthermore, DQL is employed to learn and recommend the most suitable strategy for each superior individual. Finally, extensive experiments are carried out to evaluate the correctness of the formulated MILP model and the performance of DQCEA. Experimental analysis confirms that DQCEA effectively provides promising schedules under different scenarios, demonstrating its reliability in addressing unexpected disruption challenges.