基于深度强化学习的作业车间节能调度研究

Research on energy-saving scheduling of job shop based on deep reinforcement learning

针对绿色制造背景下的作业车间调度问题,提出一种基于析取图的调度框架,该框架可以应对复杂多变的生产调度环境,并实时反映车间生产状态和机床能耗。在将调度问题转化为马尔可夫决策过程中,定义2个静态矩阵和5个动态矩阵作为状态空间,设计有关节能策略的组合调度规则,通过全局和局部两种方式描述奖励函数。最后,使用竞争深度Q网络训练模型。通过与调度规则、遗传算法等其他优化算法测试对比,证明了文章所提方法能够有效缩短完工时间和降低车间总能耗。

Aiming at the job shop scheduling problem in the context of green manufacturing,a scheduling framework based on disjunctive graph is proposed,which can cope with the complex and changeable production scheduling environment and respond to the production state of the shop and the energy consumption of machine tools in real time.In transforming the scheduling problem into a Markov decision process,two static matrices and five dynamic matrices are defined as the state space,the combined scheduling rules about energy-saving strategies are designed,and the reward function is described by both global and local ways.Finally,the model is trained using a dueling deep Q-network.By testing and comparing with other optimization algorithms such as scheduling rules and genetic algorithms,it is proved that the method proposed in this paper can effectively reduce the completion time and the total energy consumption of the job shop.