摘要
遗传算法求解多目标FJSP时,关键参数在计算过程中不能智能动态调整,从而影响算法效率和解的质量.本文基于改进的遗传算法和增强学习算法建立一种求解多目标的自学习遗传算法.遗传算法改进如下:首先提高全局和局部机器选择比重构造初始种群,然后依据快速非支配排序和拥挤距离计算适应度值,并设计选择算子,利用增强学习在种群迭代间动态调整交叉概率和变异概率,最后设计交叉和变异算子.实验部分以最大完工时间最小C_(max)、最大负荷机器最小W_(m)、总机器负荷最小W_(t)这3个目标为例,对多个算例进行了大量的测试和分析,证明了该方法的有效性和高效性.
When genetic algorithm is used to solve multi-objective FJSP,the key parameters can not be adjusted intelligently and dynamically in the process of calculation,which affects the efficiency and quality of the algorithm.Based on improved genetic algorithm and enhanced learning algorithm,a self-learning genetic algorithm for solving multi-objective is established.The improvements of GA are as follows:firstly,the proportion of global machine selection and local machine selection is increased to construct the initial population;then the fitness value is calculated according to the fast non dominated sorting and the crowding distance;and the selection operator is designed;finally,crossover probability and mutation probability are dynamically adjusted between population iterations by reinforcement learning,and crossover operation and mutation operation are designed accordingly.In the experimental part,the three targets of minimizing makespan time、maximum load machine and total machine load are taken as examples to test and analyze,which proves effectiveness and efficiency of the method.
作者
常镜洳
于东
CHANG Jing-ru;YU Dong(University of Chinese Academy of Sciences,Beijing 100049,China;Shenyang Institute of Computing Technology,Chinese Academy of Sciences,Shenyang 110168,China;Dalian Neusoft University of Information,Dalian 116023,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2022年第12期2465-2470,共6页
Journal of Chinese Computer Systems
基金
国家科技重大专项课题项目(2018ZX04032002)资助。
关键词
多目标FJSP
遗传算法
增强学习
快速非支配排序
multi-objective FJSP
genetic algorithm
reinforcement learning
fast non dominated sorting
