基于Spark的自适应差分进化极限学习机研究  

Self-Adaptive Differential Evolution Extreme Learning Machine Based on Spark

作　　者：杨敏 刘黎志[1] 邓开巍 刘杰[1] YANG Min;LIU Lizhi;DENG Kaiwei;LIU Jie(Hubei Key Laboratory of Intelligent Robot(Wuhan Institute of Technology),Wuhan 430205,China)

机构地区：[1]智能机器人湖北省重点实验室(武汉工程大学),湖北武汉430205

出　　处：《武汉工程大学学报》2021年第3期318-323,共6页Journal of Wuhan Institute of Technology

基　　金：湖北省教育厅科学研究计划指导性项目(B2017051)。

摘　　要：为了解决传统自适应差分进化极限学习机(SaDE-ELM)在单机环境下运行效率低下的问题,本文提出了基于Spark平台的并行化自适应差分进化极限学习机算法(PSaDE-ELM)。该算法的主要思想是:将差分进化算法中的原始种群均匀地分割为几个子种群,每个子种群均占有RDD的一个分区,在每个分区中使用SaDE-ELM算法独立进化,并且周期性地将各个子种群中的最优个体按照一定的拓扑结构替换掉其他子种群的最差个体,以此来达到各个子种群共同进化的目的。实验结果表明:PSaDE-ELM算法的预测准确率与SaDE-ELM算法相比基本没有丢失,且随着数据集样本数或子种群数量的增加,算法的运行效率至少提升了1.5倍,在一定程度上证明了本文提出的并行化算法的有效性。To solve the problem of low running performance of the traditional self-adaptive differential evolution extreme learning machine(SaDE-ELM)in stand-alone application environment,this paper proposes a parallel self-adaptive differential evolution extreme learning machine(PSaDE-ELM)based on the spark platform.The main principle behind the algorithm is as follows.First,the original population in the differential evolution algorithm is divided into several sub-populations evenly.Each sub-population occupies a partition of the RDD and evolves independently based onthe SaDE-ELM algorithm.Then,the optimal individuals in each sub-population are selected and replace the worst one speriodically according to a certain topological structure of the population which can achieves the purpose of coevolution of all sub-populations.The experimental results show that the prediction accuracy of the PSaDE-ELM algorithm is nearly the same with the SaDE-ELM algorithm.However,as the number of samples in the data set or the number of subpopulations increases,the running performance of the proposed algorithm is improved by at least 1.5 times,which to a certain extent validates the effectiveness of the proposed parallelized algorithm.

关 键 词：差分进化 极限学习机 并行化 SPARK 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]