Abstract: For the batch scheduling problem of parallel machines under the condition of dynamic workpiece arrival,the existing researches seldom consider the impact of constraints such as product processing incompatibility,equipment maintenance and target completion on the actual production planning.Therefore,on the basis of considering the influence of product processing incompatibility and the maximum number of shutdowns on the integrated scheduling of production and maintenance for the batch-processing parallel machines,a multi-level imperfect equipment maintenance strategy with four different maintenance activities was proposed.After that,a multi-objective integrated scheduling mathematical optimization model with the target completion quantity of different checkpoints,the equipment maintenance and the maximum number of downtime equipment was constructed.To solve the problem,four kinds of local search operators and a clustering-based crossover strategy were designed,and a Q-learning-based Self-adaptive Multi-Objective Evolutionary Algorithm (QSMOEA) was constructed.A large number of simulation experiments found that:①four kinds of local search operators and a clustering-based crossover strategy played an active and important role in QSMOEA algorithm,and their contribution to the overall performance of proposed algorithm was not less than 11.91%;②QSMOEA algorithm was significantly better than other four advanced intelligent optimization algorithms,and the average relative percentage deviation was not less than －18.58%;③the proposed multi-level imperfect equipment maintenance strategy was significantly better than the traditional equipment maintenance strategy,and the optimal maintenance plans of the proposed maintenance strategy also had significant advantages in the integrated optimization research;④through sensitivity analysis,the maximum number of shutdowns had a significant impact on the integrated scheduling results of production and maintenance.

Key words: parallel batch-processing machine, multi-level imperfect maintenance, integrated scheduling of production and maintenance, Q-learning algorithm, self-adaptive multi-objective evolutionary algorithm

摘要： 针对工件动态到达情况下的并行机批调度问题,现有研究很少考虑产品加工不兼容、设备维护和目标完成量等约束对实际生产计划的影响。为此,在考虑产品加工不兼容和最大停机设备数量对并行机生产与维护集成调度影响的基础上,提出了具有4种不同维护活动的多级非完美性设备维护策略,进而构建了考虑不同见证点目标完成量、设备维护和最大停机设备数量的多目标集成调度数学优化模型。为求解该问题,设计了4种局部搜索算子和一种基于聚类的交叉策略,并以此构建了基于Q学习的自适应多目标进化算法(QSMOEA)。最后,通过大量仿真实验发现:① 4种局部搜索算子和基于聚类的交叉策略在QSMOEA算法中发挥着积极且重要的作用,且它们对算法整体性能的贡献度不低于11.91%;② QSMOEA算法显著优于其他4种先进的智能优化算法,且平均相对百分比偏差不低于-18.58%;③ 多级非完美性设备维护策略显著优于传统设备维护策略,且所提维护策略的最优维护计划在集成优化研究中也具有显著性优势;④ 通过敏感性分析发现,最大停机设备数量对生产与维护集成调度结果具有显著性影响。

关键词: 并行批处理机, 多级非完美性维护, 生产与维护集成调度, Q学习算法, 自适应多目标进化算法