改进的Q学习算法在作业车间调度中的应用  

Application of improved Q learning algorithm to job shop problem

作　　者：王超[1] 郭静[1] 包振强[2] 

机构地区：[1]扬州工业职业技术学院电子信息工程系,江苏扬州225009 [2]扬州大学信息工程学院,江苏扬州225009

出　　处：《计算机应用》2008年第12期3268-3270,共3页journal of Computer Applications

基　　金：江苏省高校自然科学基础研究面上项目(07KJB520139);江苏省教育厅自然基金资助项目(05KJB120156)

摘　　要：在制造业系统中车间调度是一项关键技术,可以用强化学习中的Q学习实现对车间作业的动态调度。传统的Q学习存在收敛速度慢和容易导致局部收敛的矛盾,为此提出一种改进的Q学习算法。在行为动作上提出了一种双层动作合成的动作组,给出常规数学中"聚度"概念来衡量在某一状态动作组选择的均匀程度,达到既能加速收敛又能防止局部收敛的目的,能有效适应现今复杂多变的动态生产环境。实验表明,该方法运用于动态车间调度中有较好的效果。The Job Shop Problem （JSP） is a key technology in manufacturing system, and the Q learning was used to realize it. The improved Q learning algorithm was suggested because of the traditional algorithm has limitations of slow and partial constringeney. In this algorithm, a complex action group was suggested. And uniformity degrees of action choices were weighed by cohesion, and then the two limitations could be overcome effectively. It could be adapted to the complicated manufacturing environment. Experimental results show its good effect in the JSP.

关 键 词：作业车间度问题 强化学习 Q学习 聚度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]