基于Pareto的双群体多目标微分进化算法  

Multi-objective Differential Evolutionary Algorithm Based on Pareto’s Double Populations

作　　者：汤可宗[1] 丰建文[1] 柳炳祥[1] 徐星[1,2] 

机构地区：[1]景德镇陶瓷学院信息工程学院,江西333403 [2]武汉大学软件工程国家重点实验室,武汉430072

出　　处：《系统仿真学报》2013年第8期1860-1864,1870,共6页Journal of System Simulation

基　　金：国家自然科学基金(61202313);国家科技支撑计划基金(2012B A H25F02);江西省教育厅基金(GJJ12642,GJJ13637,GJJ13633);江西省自然科学基金(20132BAB211020,20122BAB201044);武汉大学软件工程国家重点实验室开放基金项目(SKLSE2012-09-35)

摘　　要：针对多目标优化问题,微分进化是一种简单、快速且具有鲁棒性的进化算法。提出一种基于Pareto的双群体多目标微分进化算法(DEPDP),DEPDP与传统微分进化区别在于:个体的变异操作和选择方式。DEPDP的变异过程类似于粒子群优化的粒子速度更新操作,即包括可行解个体,也有不可行解个体的参与;在个体的选择过程中,组合修正后的不可行解个体和可行解个体,并采用一种特殊的"非劣排序和等级选择过程"确定出新一代种群。仿真实验表明:相比其他比较算法,DEPDP获得的Pareto最优解有着良好的多样性均匀分布特点,接近真实的Pareto前沿,收敛性也较好。Differential Evolution（DE） is a simple,fast and robust evolutionary algorithm for multiobjective optimization problem（MOP）.A novel multi-objective differential evolutionary algorithm was proposed based on Pareto’s double populations（DEPDP）.There are some differences between DEPDP and traditional DE： the mutation operation and selection strategy of individuals.The former is similar to the velocity updating operation in particle swarm optimization,including the feasible and infeasible individuals.A special nondominated sorting and ranking selection scheme is incorporated in the selection of individual so as to generate a new population after combining the modified infeasible individuals and feasible individuals.Simulation results show that DEPDP obtains Pareto optimal solutions with better uniform distribution characteristics of diversity than other compared algorithms,and they are close to the true Pareto front.Also,its convergence is good.

关 键 词：微分进化 多目标优化 变异 选择 多样性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]