基于深度学习神经网络和量子遗传算法的柔性作业车间动态调度  

Dynamic scheduling of flexible job shop based on deep Q-learning neural network and quantum genetic algorithm

作　　者：陈亮[1] 阎春平[1] 陈建霖 侯跃辉 CHEN Liang;YAN Chunping;CHEN Jianlin;HOU Yuehui(College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing 400044,P.R.China)

机构地区：[1]重庆大学机械与运载工程学院,重庆400044

出　　处：《重庆大学学报》2022年第6期40-54,共15页Journal of Chongqing University

基　　金：重庆市技术创新与应用示范项目(cstc2018jszx-cyzdX0163)。

摘　　要：针对柔性作业车间动态调度问题构建以平均延期惩罚、能耗、偏差度为目标的动态调度优化模型,提出一种基于深度Q学习神经网络的量子遗传算法。首先搭建基于动态事件扰动和周期性重调度的学习环境,利用深度Q学习神经网络算法,建立环境-行为评价神经网络模型作为优化模型的适应度函数。然后利用改进的量子遗传算法求解动态调度优化模型。该算法设计了基于工序编码和设备编码的多层编码解码方案;制定了基于适应度的动态调整旋转角策略,提高了种群的收敛速度;结合基于Tent映射的混沌搜索算法,以跳出局部最优解。最后通过测试算例验证了环境-行为评价神经网络模型的鲁棒性和对环境的适应性,以及优化算法的有效性。To deal with the problem of dynamic scheduling of flexible job shop,a dynamic scheduling optimization model was constructed to minimize average delay penalty,energy consumption and deviation,and an ameliorated quantum genetic algorithm based on deep Q-learning neural network was proposed.First,a learning environment based on dynamic event disturbance and periodic rescheduling was built,and an environment-behavior evaluation neural network model was established using deep Q-learning neural network algorithm as the fitness function of the optimization model.Then the dynamic scheduling optimization model was solved by using the improved quantum genetic algorithm which designed a multi-layer encoding and decoding scheme based on process encoding and equipment encoding.A strategy for dynamically adjusting the rotation angle based on fitness was developed to improve the convergence speed of the population and exclude local solutions by combining with chaos-based Tent mapping search.Finally,test cases verified the robustness and adaptability of the environment-behavior evaluation neural network model,as well as the effectiveness of the optimization algorithm.

关 键 词：柔性作业车间动态调度 能耗 平均延期惩罚 偏差度 深度Q学习神经网络 改进量子遗传算法 混沌搜索 

分 类 号：TH11[机械工程—机械设计及理论]