基于AdaBoost多分类算法变压器故障诊断

Fault diagnosis of transformer based on multi-classification algorithm AdaBoost

油中溶解气体分析(Dissolved Gas Analysis,DGA)技术可以有效发现充油变压器内部的早期潜伏性故障,是对充油设备进行诊断的常用方法.k-最近邻(k-Nearest Neighbor,kNN)算法是一种惰性分类算法.为了满足实际工程中对变压器故障模式分类精度的要求,AdaBoost.M2作为AdaBoost二分类算法的延伸,可将多个略好于随机猜测的弱分类器组合提升为分类精度更高的强分类器,完成多分类任务.针对单一算法往往不能满足实际工程对分类精度的需求且高精度算法难以获得的问题,利用AdaBoost的扩展算法AdaBoost.M2对每个kNN分类器的权重根据误差不断调整,再通过加权投票将其组合提升为强分类器,提高了故障诊断精度.实验结果显示,运用该模型结合DGA技术对变压器故障进行诊断,相比于单一kNN算法,诊断准确率整体提高了27.8%,表明该方法是可行的.

The internal early potential fault of oil‐immersed transformer can be found effectively by the technology of Dissolved Gas Analysis ,w hich is a common detection method for diagno‐sis of oil filled equipment .kNN is an inert classification method .To meet the precision require‐ments of transformer failure classification in the practical engineerings ,as the extension of two classification algorithms ,AdaBoost .M2 completes multi‐class mission ,by improving weak classifier combinations better than random guessing to the strong classifier with better classsifica‐tion accurary .Concerning that a single algorithm can′t meet the actual engineering require‐ments of classification accuracy ,and high accuracy algorithm is difficult to get ,using AdaBoost‐M2 to adjust the weight of each kNN classifier according to the error .Then ,these weak learn‐ers are promoted to a strong classifier by weighted voting .The accuracy of fault diagnosis is improved in this way .The results of simulation show that this model is available in improving the accuracy rate of faults diagnosis of transformer .The diagnostic accuracy rate is proved to be increased by 27.8% .Therefore ,this method is workable .