本文提出多特征融合和IPSO-KNN的有载分接开关机械故障特征选择。首

Proposed feature fusion and IPSO-KNN-load tap mechanical fault feature points selected. First, mRMR principle constructed in time domain, the energy and multi-scale weighted permutation entropy high dimensional set for screening, fault feature subset sensitive; then, the improved PSO sensitive subset of features to optimize the most preferably feature subset using KNN optimal feature subset of the different types of classification; finally, select the normal OLTC contacts, contacts burned off contacts and the contact data of four loose state, with FFMWPE, FFMPE , SMWPE, FFWPE MFPE and features of the method compared by comparison of the results of analysis of experimental data to verify the validity of the method, while emphasizing the advantages of fusion based MWPE characterized in sensitive feature extraction, feature selection described mRMR necessity.

本文提出多特征融合和IPSO-KNN的有载分接开关机械故障特征选择。首先，利用mRMR原则对构建出时域、能量和多尺度加权排列熵的高维特征集进行筛选，得到故障敏感的特征子集；然后，利用改进的粒子群算法对敏感特征子集进行优化得到最优特征子集，并使用KNN对不同类型的最优特征子集进行分类；最后，选取OLTC触头正常、触头烧毁、触头脱落和触头松动4个状态的数据，并与FFMWPE、FFMPE、SMWPE、FFWPE和MFPE特征方法相比，通过实验数据的结果分析对比，验证了所提方法的有效性，同时强调了基于MWPE特征融合在提取敏感特征方面的优越性，说明了mRMR特征选择的必要性。

In this paper, multi feature fusion and ipso-knn are proposed to select mechanical fault features of on load tap changer. Firstly, the high Witt feature set of time-domain, energy and multi-scale weighted permutation entropy is screened by using the principle of mrmr to get the fault sensitive feature subset; secondly, the improved particle swarm optimization algorithm is used to optimize the sensitive feature subset to get the optimal feature subset, and KNN is used to classify the different types of optimal feature subset; finally, the OLTC contact is selected as normal and touching Compared with ffmwpe, ffmpe, smwpe, ffwpe and mfpe, the results of experimental data analysis and comparison verify the effectiveness of the proposed method. At the same time, the advantages of mwpe based feature fusion in extracting sensitive features are emphasized, which shows the necessity of feature selection of mrmr.