基于随机森林的变压器局部放电模式识别

Transformer Partial Discharge Pattern Recognition Based on Random Forest

针对目前在局部放电模式识别领域中常用的分类器算法的缺陷,本文研究随机森林(random forest,RF)算法在局部电放模式识别领域的应用。首先对局部放电试验数据提取统计特征量,构建放电的学习样本。利用十折法对算法分类性能进行评判,并比较常见分类算法BP神经网络、支持向量机(support vector machine,SVM)、KNN、分类回归树算法(classification and regression tree,CART)以及RF算法的识别准确率。结果表明:利用RF算法构建放电模式分类器的识别准确率最高。此外,利用组成RF的基分类算法CART可分析不同放电模式间的主要区别。

As the deficiencies of the classifier algorithm commonly used in the partial discharge pattern recognition,this paper studies the application of RF in transformer partial discharge pattern recognition.Firstly,extracting the statistical characteristics from partial discharge test data to establish the discharge samples;Then,using ten folds method to judge the algorithm performance and compare the recognition accuracy of BP neural network,support vector machine,KNN,CART and RF algorithm.The results show that the accuracy of the discharging pattern classifier basic on RF algorithm was the highest.The main differences between the different discharge modes were discussed by CART algorithm which composes RF.