求解大规模优化问题的云差分进化算法

Cloud computing based differential evolution algorithm for large-scale optimization problems

针对大规模优化问题求解难、差分进化算法运算时间长等问题,利用云计算MapReduce并行编程模型,结合差分进化算法隐含并行性,提出云差分进化算法。该算法利用Hadoop集群平台,采用多子群机制,并将子种群与Map任务形成一一对应关系;算法的各个子种群之间根据拓扑结构进行个体迁移,以增加其多样性,从而能搜索更大的范围,提高寻优的几率。仿真实验结果表明,云差分算法能有效地减少求解大规模优化问题的时间消耗,并且取得较好的精度。

Large scale op tim ization problem s are hard to be so lve d , and it u su ally takes difference evo lu tion algo rithm longtim e to ta ckle i t . W ith M apReduce m odel and the im p lic it p a ra lle l cha racte ristic o f d iffe re n tia l e v o lu tio n , this paper proposedthe cloud com puting based d iffe re n tia l evo lu tion a lg o rith m . T h is m ethod was based on Hadoop p la tfo rm ,a n d adopted m u lti popu la tio n m echanism . In order to pa ra lle lize d iffe re n tia l evo lu tion e ffe c tiv e ly , each Map task was responsible fo r each subpopulation . In order to increase the d ive rsity o f p o p u la tio n , it a p plied m igration operation between each subpopulation according to thetopology stru ctu re . The m igration between subpopulation could enlarge the search space and im proves the p ro b a b ility to locateop tim um . The sim ula tion experim ent results show that the cloud com puting based d iffe re n tia l evo lu tion can not only effectivelyreduce the tim e co n su m p tio n ,b u t also im proves the accuracy o f solutions.