Job-shop排序问题的遗传强化学习算法  

Genetic Reinforcement Learning Algorithm for Job-shop Scheduling Problem

作　　者：潘燕春[1] 周泓[2] 

机构地区：[1]深圳大学管理学院,深圳518060 [2]北京航空航天大学经济管理学院,北京100083

出　　处：《计算机工程》2009年第16期25-28,共4页Computer Engineering

基　　金：国家自然科学基金资助项目(70771003;70521001)

摘　　要：针对Job-shop排序问题的复杂性,提出一种遗传强化学习算法对其求解。通过引入多个随机变量,把Job-shop排序问题转换成多阶段决策问题,通过仿真手段构建作业排序问题模型环境,求取系统性能指标并保证解的可行性。设计一个多智能体Q-Learning算法和遗传算法相结合的算法用于解决Job-shop排序问题。仿真优化实验结果验证了该算法的有效性。Aiming at Considering the complexity of Job-shop scheduling problem, this paper proposes Genetic Reinforcement Leaning（GRL） algorithm to solve it. Several stochastic variables are introduced to transform the Job-shop scheduling problem into sequential decision-making problem, and the model environment is built through simulation to obtain system performance indices and ensure the feasibility of solution. An algorithm integrating multi-Agent Q-Learning algorithm with Genetic Algorithm（GA） is designed to solve Job-shop scheduling problem. GRL algorithm is validated by simulation and optimization experiment results.

关 键 词：遗传强化学习 Job—shop排序 多阶段决策 仿真 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]