基于NSGAII的多目标随机加工时间柔性车间作业调度
Multi-Objective Flexible Job Shop Scheduling with Stochastic Processing Times Based on NSGAII

新兴信息技术的飞速发展与全球经济一体化格局加速国际强国间新一轮的激烈角逐，智能制造成为我国应对瞬息万变的市场需求，向数字化、自动化、智能化、低碳化转型升级的重大战略决策。柔性车间作业调度是智能制造车间生产计划管理的重要问题之一，因其NP-hard本质而备受关注，此外实际运营环境存在的不确定因素使现有解决方案面临巨大的挑战。文章以最小化最大完工时间、最小化误工–提前成本、最小化机器总负载为优化目标，结合作业加工时间随机变化的特性，围绕多目标随机加工时间的柔性车间作业调度问题展开研究，提出一种改进NSGAII的多目标随机加工时间柔性车间作业调度方法(N-MOSFJ)。该方法依据各目标偏好设计种群初始化方法，提升初始解质量和多样性，同时，通过构造参考下界在具有随机加工时间动态变化的解空间中为算法迭代引导可参考的搜索方向。为深入挖掘局部最优解，生成感知绩效因子，并设置动态定位因子以加速全局靶向探索，基于此建立个体综合适应度值评估机制确保算法客观量化个体质量。此外，采用差异优先与质量优先混合选择模式，协同质量约束的“优生多生”交叉变异策略，有效平衡多目标间冲突。通过在仿真数据集和领域公开数据集上对算法进行比较验证，实验结果表明提出的算法具有较好的有效性与稳定性。
The rapid development of emerging information technologies and the acceleration of global economic integration have intensified competition among international powers. Intelligent manufacturing has become a critical strategy for China to address dynamic market demands and achieve transformation toward digitization, automation, intelligence, and low-carbon development. Flexible job shop scheduling is a vital problem in production planning within intelligent manufacturing systems. Its NP-hard nature and the uncertainties in real-world operational environments present significant challenges to existing solutions. This study investigates a multi-objective flexible job shop scheduling problem with stochastic processing times. The objectives are to minimize the makespan, tardiness-early cost, and total machine load. A novel method based on an improved NSGAII, called N-MOSFJ, is proposed. The method designs a preference-based population initialization strategy to enhance the quality and diversity of initial solutions. Reference lower bounds are constructed to guide the algorithm’s iterations in a dynamically changing solution space with stochastic processing times. A performance-aware factor is introduced to improve local exploitation, while a dynamic positioning factor accelerates global exploration. An integrated fitness evaluation mechanism ensures an objective assessment of individual solution quality. To balance conflicts among multiple objectives, a hybrid selection strategy combining diversity and quality priorities is adopted. This is integrated with a quality-constrained “elite reproduction and multiple reproduction” crossover and mutation strategy. By comparing and validating the algorithm on simulation datasets and publicly available domain datasets, experimental results show that the proposed algorithm demonstrates good effectiveness and stability.