基于深度强化学习的云制造产品配置  

Cloud Manufacturing Product Configuration Based on Deep Reinforcement Learning

作　　者：童晓薇[1] 刘艳斌[2] TONG Xiaowei;LIU Yanbin(School of Mechanical and Intelligent Manufacturing,Fujian Chuanzheng Communications College,Fuzhou Fujian 350007;Fujian College Association Instrumental Analysis Center,Fuzhou University,Fuzhou Fujian 350116)

机构地区：[1]福建船政交通职业学院机械与智能制造学院,福建福州350007 [2]福州大学福建省高校测试中心,福建福州350116

出　　处：《湖北理工学院学报》2023年第6期46-51,共6页Journal of Hubei Polytechnic University

基　　金：福建省教育厅科技项目(项目编号:JZ180373)。

摘　　要：针对云制造环境下供应商加入退出、制造成本变动等因素导致的传统产品配置方法不适用的问题,提出了一种以全过程服务质量最大化为目标的产品配置建模方法,建立了适应动态变化的云制造环境的产品配置模型。文章将云制造产品配置问题视为马尔可夫决策过程,阐明了面向产品配置优化问题的智能体、环境、动作、奖励、转移等强化学习关键概念,设计了基于深度Q网络的产品配置优化求解算法。轮式装载机产品配置仿真实验表明,基于深度强化学习的云制造产品配置方法的模型全局奖励值接近且收敛于理论奖励值上限,很好地适应了云制造环境的动态变化特征。Aiming at the problem that the traditional product configuration method is not applicable due to factors such as the entry and exit of suppliers and the change of manufacturing cost in the cloud manufacturing environment,a product configuration modeling method aiming at maximizing the quality of the whole process service is proposed,and a product configuration model adapted to the dynamically changing cloud manufacturing environment is established.This paper regards the cloud manufacturing product configuration problem as a Markov decision-making process,clarifies the key concepts of reinforcement learning such as intelligences,environments,actions,rewards,transfers,etc.,oriented to the product configuration optimization problem and a deep Q-network-based algorithm is designed for solving the product configuration optimization.Simulation experiments based on wheel loader product configuration show that the global reward value of the model of the cloud manufacturing product configuration method based on deep reinforcement learning proposed in this paper is close to and converges to the upper limit of the theoretical reward value,which is well adapted to the dynamic changes of the cloud manufacturing environment feature.

关 键 词：产品配置 云制造 强化学习 深度Q网络算法 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]