摘要
针对传统演化优化算法难以在动态环境下有效地持续跟踪最优解的问题,本文提出了一种自适应反向扩散演化算法(ARDEA)对动态环境进行寻优.该算法采用多种群策略对最优解进行跟踪,通过设置子种群全局动态监哨点监测环境变化;并引入一种差分粒子群速度更新公式引导个体在搜索空间内不断寻找最优值;同时,本文提出了一种新的排斥策略以确保种群多样性,以及种群持续跟踪最优解的搜索能力.该策略包含两个新方法:其一,采用新提出的群间平均马氏距离判断种群间距离,对于群间距过小的两个子种群进一步通过hill-valley函数判定它们的搜索空间是否重叠,其二,将重叠搜索空间中的劣势子种群通过反向扩散操作(RD)重新初始化.新算法与当前性能较优的动态优化算法同时作用于移动峰测试问题,结果表明,ARDEA算法在动态环境中能更加有效地跟踪最优解,与其它比较算法而言,表现出较强的鲁棒性和适应性.
To solve the problem that traditional evolutionary optimization algorithm is difficult to effectively keep track of the optimal solution in dynamic environment,this paper proposes an adaptively reversed diffuse evolutionary algorithm(ARDEA).The new algorithm adopts the multi-population strategy to track the optimal solution and monitors the environmental changes by setting the global dynamic sentryin each subgroup.A differential particle swarm velocity update formula is introduced to guide individuals to search for the optimal points in the search space.Meanwhile,in order to ensure the diversity of the population and the search efficiency of the sub-population,a new exclusion strategy is proposed in this paper.This strategy includes two method.Firstly,it uses between-swarms average Mahalanobis distance to judge the inter-population distance.If the distance is too small between two sub-populations,hill-valley function is used to further determine whether they tracked the same peak or not.Secondly,the subpopulations with poor performance in the search overlap will be reinitialized by reverse diffusion operation(RD).The new algorithm is compared with several state-of-artdynamic optimization algorithms on moving peak problem.The results show that the ARDEA algorithm can track the optimal solution more effectively in the dynamic environment.Compared with other algorithms,the ARDEA algorithm shows strong robustness and adaptability.
作者
曹文梁
康岚兰
王石
Cao Wenliang;Kang Lanlan;Wang Shi(Department of Computer Engineering,Dongguan Polytechnic,Dongguan 523808,China;College of Applied Science,Jiangxi Lniversity of Science and Technology,Ganzhou 341000,China)
出处
《南京师大学报（自然科学版）》
CAS
CSCD
北大核心
2020年第4期119-128,共10页
Journal of Nanjing Normal University(Natural Science Edition)
基金
广东省普通高校特色创新(自然科学)项目(2019GKTSCX142、2017GKTSCX101)
东莞职业技术学院示范建设专项资金项目(政201803)
江西省科技厅自然科学基金面上项目(20202BABL202032)
江西省教育厅科技项目(GJJ181511)。
关键词
动态优化
粒子群优化
反向扩散
群间平均马氏距离
dynamic optimization
particle swarm optimization
reversed diffuse
between-swarms average Mahalanobis distance
