基于免疫优化多分类SVM的变压器故障诊断新方法

摘要

针对支持向量机中参数设置对支持向量机分类精确度影响较大及传统支持向量机不能直接用于多分类问题的状况,提出了一种基于免疫优化多分类支持向量机的变压器故障诊断新方法,该方法利用免疫算法优化支持向量机分类参数.以一类分类算法为基础建立多分类算法模型,在高维特征空间求出超球体中心,然后计算样本与中心最小距离,以此判定该点所属故障类型.该算法充分发挥了支持向量机高泛化能力的优势,大大减少了对支持向量机参数选择的盲目性.仿真计算结果表明,在有限样本情况下,该方法能够达到较高的变压器故障诊断率,从而证实了该方法的正确性和有效性.%Considering the fact that the parameter setting for support vector machine (SVM) impacts on the classification accuracy and the traditional SVM can not deal with multi-class classification directly, a novel approach for transformer fault diagnosis based on multi-class support vector machine of immune optimization is presented, in which the parameters in SVM are optimized by immune algorithm. Multi-class algorithm model is established on the basis of one-category classification algorithm, hypersphere centers are obtained in high-dimensional feature space, and then the minimum distances are calculated between the sample and the center in order to determine the fault type the sample belongs to. The superiority of SVM in processing finite samples is fully brought into play, and blindness is greatly reduced about parameter selection of SVM. Simulation results show that the algorithm can detect transformer faults with a higher diagnosis rate in the case of limited samples, and prove the correctness and effectiveness of the method.