组合核支持向量机在放电模式识别中的优化策略

发布时间：2018-05-01 08:31

本文选题：局部放电 + 组合核　；参考：《电工技术学报》2015年02期

【摘要】：传统的单一核函数SVM无法实现局部放电中的多特征空间向量的映射分类,且目前SVM大都采用不同尺度的径向基函数作为核函数,核的调整空间较为有限,无法针对不同特征空间的放电参数达到普适的效果。针对上述问题,本文提出一种基于多分组特征的组合核多分类SVM的局部放电识别方法。该方法首先完成多类特征空间在不同类型SVM核函数中的映射分类,再采用骨干粒子群(BBPSO)优化方法选取最佳核参数,并求解核函数权值系数,最终形成最优核函数组合分类模型。实验结果表明,该方法对多个特征空间数据具有普适性,且融合效果理想,分类精度高于误差反向传播神经网络(BPNN)和SVM识别方法。
[Abstract]:The traditional single kernel function SVM can not realize the mapping classification of multi-feature space vectors in partial discharge. At present, most of SVM uses radial basis function of different scales as kernel function, and the adjustment space of kernel is relatively limited. The discharge parameters in different feature spaces can not achieve universal effect. In order to solve the above problems, this paper presents a method of partial discharge recognition based on multi-packet feature combined kernel multi-classification SVM. In this method, the mapping classification of multi-class feature spaces in different types of SVM kernel functions is completed, and then the optimal kernel parameters are selected by using the backbone particle swarm optimization method, and the weight coefficients of the kernel functions are solved, finally the optimal kernel function combination classification model is formed. The experimental results show that this method is universal for multiple feature space data, and the fusion effect is ideal, and the classification accuracy is higher than that of the error back-propagation neural network (BPN) and the SVM recognition method.
【作者单位】：华北电力大学电气与电子工程学院;东北电力大学电气工程学院;
【基金】：中央高校基本科研专项资金(2014ZD32,13MS69) 国家自然科学基金(61204027)资助项目
【分类号】：TM855;TP18

【参考文献】