基于改进的差异演化算法求解SVM反问题的研究

2018-03-13 22:18:35樊永生熊焰明余红英
现代电子技术 2018年6期

樊永生+熊焰明+余红英

摘 要: 针对求解支持向量机反问题的效率较低,算法复杂度高以及运用传统方法求解该问题容易陷入局部最优出现早熟收敛的问题,提出一种基于改进差异的差异演化算法。该算法在标准差异演化算法的基础上利用种群分类机制对算法进行改进,对改进后的算法与标准差异演化算法和K?means聚类算法进行实验设计,并对算法最终实验结果进行分析,改进的差异演化算法除在运行时间外,结果对比以及最大间隔次数比都有明显的提升,有效地保护处于最优解区域但是适应值低的个体,能够提高算法局部搜索能力,有助于算法实现全局收敛。实验结果表明,改进的差异演化算法在求解SVM反问题上能有明显的提升。

关键词: 支持向量机; 局部最优; 差异演化算法; 全局收敛; 种群分类机制; IRIS数据库

中图分类号: TN911?34; TP301.6 文献标识码: A 文章编号: 1004?373X(2018)06?0141?04

Abstract: Since the traditional algorithm has the problems of low efficiency of support vector machine (SVM) inverse problem solving and high algorithm complexity, is easily to fall into local optimum, and is prone to the premature convergence, a new differential evolutionary algorithm based on improved difference is proposed. On the basis of normative differential evolution algorithm, the population classification mechanism is used to improve the algorithm. The experimental design was carried out for the improved algorithm, normative differential evolution algorithm and K?means clustering algorithm. The final experimental results of the algorithm are analyzed. The maximum interval numbers and average interval numbers of the changed differential evolution (CDE) algorithm beyond operation time are improved greatly. The algorithm can effectively protect the individual within the optimum solution region but with low adaptive value, improve the local search ability of the algorithm, and is conductive to the realization of global convergence. The experimental results show that the performance of the CDE algorithm is improved obviously for SVM inverse problem solving.

Keywords: support vector machine; local optimum; differential evolution algorithm; global convergence; population classification mechanism; IRIS database

0 引 言

支持向量机(Support Vector Machine,SVM)是机器学习领域中一个重要的研究热点,因其具有优秀的泛化能力和学习能力被广泛地应用在模式识别、文本分类、信号处理、回归分析等领域[1?3]。SVM是在统计学习理论的基础上,提出最优超平面间隔理论。该理论结合结构风险最小化原理,将原始训练样本进行提取压缩,得到支持向量集合,通过这些支持向量子集学习新的知识。SVM出发点是根据已经给定的正负标签的类别信息,寻找两类样本之间的最优超平面。在寻找过程中,靠近当前最优超平面最近的样本点到超平面的距离尽可能的大,不同类别的样本之间的间距也会越大,分类器的泛化能力也就越强。……

登录APP查看全文