基于强化学习的考虑运输资源分布式调度问题  

Reinforcement Learning Based Distributed Scheduling Problem for Considering Transportation Resources

作　　者：唐红涛[1,2] 廖义峰 TANG Hongtao;LIAO Yifeng(School of Mechanical and Electronic Engineering,Wuhan University of Technology,Wuhan Hubei 430070,China;Hubei Provincial Engineering Research Center of Robotics and Intelligent Manufacturing,Wuhan Hubei 430070,China)

机构地区：[1]武汉理工大学机电工程学院,湖北武汉430070 [2]机器人与智能制造湖北省工程研究中心,湖北武汉430070

出　　处：《机床与液压》2025年第1期43-51,共9页Machine Tool & Hydraulics

基　　金：国家自然科学基金项目(51705384;52075401)。

摘　　要：分布式车间调度问题在传统制造业中具有重要意义。然而,在实际生产过程中,各种加工资源通常是有限的,且在以往分布式调度问题中未考虑到。针对分布式工厂运输机器有限问题,提出一种基于强化学习的RLDE算法。分析并建立RLDE的自学习模型,同时设计了3种初始化策略来获得高质量种群。设计基于Q-learning的变异和变邻域策略,使种群能够选择合适的变异和邻域搜索方法。在多个著名的基准实例上对所提RLDE算法进行性能评估,并与4种先进算法进行了比较。结果表明:RLDE算法在解决分布式车间调度资源约束问题上表现出非凡的优越性。The distributed workshop scheduling problem is of great importance in traditional manufacturing industries.However,in the actual production process,various processing resources are usually limited and not considered in previous distributed scheduling.Aiming at the problem of limited transportation machines in distributed factories,a reinforcement learning-based RLDE algorithm was proposed.A self-learning model of RLDE was analyzed and established,while three initialization strategies were designed to obtain a high-quality population.Q-learning-based variation and variation-neighborhood strategies were designed to enable the population to select appropriate variation and neighborhood search methods.The performance of the proposed RLDE algorithm was evaluated on several well-known benchmark instances and compared with four state-of-the-art algorithms.The results show that the RLDE algorithm exhibits besides extraordinary superiority in solving the distributed workshop scheduling resource constraint problem.

关 键 词：分布式作业车间调度 Q-LEARNING 资源约束 多目标优化 

分 类 号：TB497[一般工业技术] TP278[自动化与计算机技术—检测技术与自动化装置]