采用核主成分分析和随机森林算法的变压器油纸绝缘评估方法

Oil － paper Insulation Evaluation Method of Transformer Based on Kernel Principal Component Analysis and Ｒandom Forest Algorithm

摘要:

为了实现对电力变压器绝缘状态的智能评估，提出了一种融合核主成分分析和集成学习理论的电力变压器油纸 绝缘评估方法。在特征提取方面，通过回复电压法( recovery voltage method，ＲVM) 提取特征量，并对特征量进行核主成 分分析( kernel principle component analysis，KPCA) ，将低维度的样本空间映射到高维度的核空间，并按照重要性进行排 序; 在评估识别方面，利用集成学习的思想建立分类器群模型，克服了单分类器的局限性，并提高了分类器的分类预测能 力。通过实例论证，融合核主成分分析和集成学习的分类模型在变压器油纸绝缘评估中具有很高的准确性。

Abstract:

In order to realize the intelligent evaluation of power transformer insulation status，an evaluation method for oil － paper insulation of transformer is proposed based on kernel principal component analysis and random forest algorithm． Ｒecovery voltage method ( ＲVM) is to study insulation aging status of power transformer． In order to overcome the limitations of single classifier，random forest ( ＲF) is introduced to get classifier groups． The randomness of training samples and feature selection can avoid the problem of overfitting． The differences are small among classifiers built with features when there are few features available． The original samples are analyzed by using kernel principle component analysis ( KPCA) so as to increase the number of features． Then random forests are constructed to get classifier groups in high dimensional kernel space． Finally，the voting results show that，the differences among classifier groups are improved effectively by KPCA． And results of the state diagnosis based on KPCA and random forests have a higher accuracy．