为解决加工任务动态到达的跨单元调度问题,使其能够在智能车间复杂多变的环境中实现自适应调度,提出一种基于深度Q网络的调度方法。构建以单元为节点,工件跨单元加工路径为有向边的复杂网络,引入度值定义了具有跨单元调度特征的状态空...为解决加工任务动态到达的跨单元调度问题,使其能够在智能车间复杂多变的环境中实现自适应调度,提出一种基于深度Q网络的调度方法。构建以单元为节点,工件跨单元加工路径为有向边的复杂网络,引入度值定义了具有跨单元调度特征的状态空间。设计了由工件层、单元层和机器层组成的复合调度规则,分层优化使调度方案更加全局化。针对DDQN(double deep Q networks)在训练后期还会选择次优动作的问题,提出了以指数函数为主体的搜索策略。通过不同规模的仿真实验,验证了所提方法能够应对多变的动态环境,快速生成较优的调度方案。展开更多
Inter-cell scheduling is analyzed where there are various machine types and exceptional jobs that need to visit machines in multiple job shop cells in a cellular manufacturing system(CMS).A hybrid heuristic approach c...Inter-cell scheduling is analyzed where there are various machine types and exceptional jobs that need to visit machines in multiple job shop cells in a cellular manufacturing system(CMS).A hybrid heuristic approach combining hyper-heuristic and ant colony optimization is developed.Different structures are designed for different types of machines under the framework of ACO.The pheromone trails are updated integratedly after the whole solution is constructed to achieve cooperative optimization.Experimental results show that the hybrid heuristic approach has significant advantages over ACO and CPLEX with respect to the makespan and utilization rate of the batch processing machine,and our approach is especially suitable for large dimension problems.展开更多
文摘为解决加工任务动态到达的跨单元调度问题,使其能够在智能车间复杂多变的环境中实现自适应调度,提出一种基于深度Q网络的调度方法。构建以单元为节点,工件跨单元加工路径为有向边的复杂网络,引入度值定义了具有跨单元调度特征的状态空间。设计了由工件层、单元层和机器层组成的复合调度规则,分层优化使调度方案更加全局化。针对DDQN(double deep Q networks)在训练后期还会选择次优动作的问题,提出了以指数函数为主体的搜索策略。通过不同规模的仿真实验,验证了所提方法能够应对多变的动态环境,快速生成较优的调度方案。
文摘Inter-cell scheduling is analyzed where there are various machine types and exceptional jobs that need to visit machines in multiple job shop cells in a cellular manufacturing system(CMS).A hybrid heuristic approach combining hyper-heuristic and ant colony optimization is developed.Different structures are designed for different types of machines under the framework of ACO.The pheromone trails are updated integratedly after the whole solution is constructed to achieve cooperative optimization.Experimental results show that the hybrid heuristic approach has significant advantages over ACO and CPLEX with respect to the makespan and utilization rate of the batch processing machine,and our approach is especially suitable for large dimension problems.
