基于复合规则和强化学习的混流装配线调度方法  

Compound Rules and Reinforcement Learning Based Scheduling Method for Mixed Model Assembly Lines

作　　者：郭具涛[1,2] 吕佑龙 戴铮 张洁 郭宇[2] GUO Jutao;LYU Youlong;DAI Zheng;ZHANG Jie;GUO Yu(Shanghai Spaceflight Precision Machinery Institute,Shanghai,201600;School of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing,210016;Institute of Artificial Intelligence,Donghua University,Shanghai,201620)

机构地区：[1]上海航天精密机械研究所,上海201600 [2]南京航空航天大学机电学院,南京210016 [3]东华大学人工智能研究院,上海201620

出　　处：《中国机械工程》2023年第21期2600-2606,2614,共8页China Mechanical Engineering

基　　金：装发快速支持项目(JZX7Y20220163200201);科技创新行动计划启明星项目(ZZQB14042000)。

摘　　要：针对混流装配线的平衡与排序问题,提出了一种基于复合规则和强化学习的智能调度方法。根据数学模型,设计了平衡规则库与排序规则库,提出了规则加权组合的近端策略优化(PPO)算法,并利用具有Actor-Critic训练流程和优先经验回放机制的强化学习过程,实现了复合规则权值参数的调控优化,生成了平衡与排序方案。所提方法与PPO+单一规则算法、复合规则和遗传算法的对比实验验证了所提方法的有效性。A scheduling method was proposed based on compound rules and reinforcement learning for balancing and sequencing problems of mixed model assembly lines.A balancing rule set and a sequencing rule set were designed with the consideration of mathematical model,and a proximal policy optimization(PPO)algorithm featured with Actor-Critic training procedure and preferential experience learning mechanism was employed to regulate weighted parameters of these rules,in order to generate reasonable balancing and sequencing solutions.In comparative experiments,the proposed scheduling method demonstrates the effectiveness over other methods including PPO algorithm with single rule,compound rules,and a genetic algorithm.

关 键 词：混流装配线 平衡与排序 深度强化学习 复合规则 集成优化 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TP391.9[自动化与计算机技术—计算机科学与技术]