一种基于网格服务的分布式GEP-BP分类算法

摘要

When distributed data is classified by traditional and single BP algorithm,training speed is low and the local optimum is immersed readily. To solve the problem, in the present research,it presents hybrid classification algorithm based upon GEP-BP (HCA-GB). In the HCA-GB,range of GEP population is adjusted dynamically by means of self-adaptive coefficient. On the basis of HCA-GB, this paper proposes classification of distributed GEP-BP on grid service (CDGB-GS), which combines grid service and HCA-GB to resolve classification of distributed data. By simulated experiment,it is showed that the average number of convergence of HCA-GB is improved about two times by means of adjusting self-adaptive coefficient. For very large and complex data sets, average consumptive time of CDGB-GS is less than traditional algorithms, and classification accuracy of CDGB-GS is improved by about 32.06% at most in proportion with traditional algorithms.%为了克服单一BP算法对分布式数据进行分类时具有训练速度慢、易陷入局部最优等缺陷,提出了基于GEP-BP的混合分类算法HCA-GB,同时结合网格服务的思想,提出了基于网格服务的分布式GEP-BP分类算法CDGB-GS,且在HCA-GB算法中,利用自适应系数的方法动态调整GEP种群的大小,从而有效地提高了HCA-GB的全局收敛性.比较仿真实验表明,通过动态调整自适应系数,HCA-GB的平均收敛次数提高了约2倍;对于大数据集而言,在实验室局域网环境下,CDGB-GS算法的平均耗时比传统算法要小,与传统算法相比,CDGB-GS算法的分类精度最大提高了约32.06%.