基于多核学习整合GIS局部放电多类特征的分类研究
Research on classification of multi-class partial discharge features in GIS based on multiple kernel learning

摘要 传统的单一核函数构成的支持向量机(Support Vector Machine, SVM)无法解决气体绝缘组合电器(Gas Insulated Switcher, GIS)所采集到的局部放电数据(Partial Discharge, PD)分布不规则、特征类别复杂和规模巨大等问题.针对这类问题，本文提出使用多个核的线性组合方式即简单多核学习方法（Simple Multiple Kernel Learning, SimpleMKL）对多类特征进行分类研究.通过赋予不同局部放电特征不同的核函数，以多个核函数的线性加权代替传统的单一核函数，利用梯度下降法不停迭代求解核函数的权值系数，最终达到整合局部放电多类特征并提高分类精度的目的.实验结果表明:该方法对局部放电多个特征空间具有普适性，分类精度高于单核SVM和融合SVM识别方法.
Abstract：The SVM formed by the traditional single kernel function cannot solve the practical problems such as irregular distribution of GIS partial discharge, complicated characteristic category and large scale. In view of the above problems, this paper proposes the use of multiple kernel linear combinations, SimpleMKL method, to study the classification of multiple features. By assigning different kernel functions with different partial discharge characteristics, linear weighting of multiple kernel functions replaces the traditional single kernel function, and the gradient descent method is used to iteratively solve the weight coefficients of the kernel, eventually achieving the integration of PD multiple types of characteristics and improving the classification accuracy. The experimental results show that the proposed method is universal for multiple feature spaces of partial discharge, and the classification accuracy is higher than the single kernel SVM and fusion SVM recognition method.