基于IGA优化AdaBoost-SVM的变压器多级故障诊断方法研究

Research on Multilevel Fault Diagnosis Method for Transformers Based on IGA-Optimised AdaBoost-SVM

中文关键词:变压器  故障诊断  AdaBoost算法  改进遗传算法  支持向量机

keywords:transformer  fault diagnosis  AdaBoost algorithm  IGA  support vector machine

中文摘要:针对单个支持向量机（support vector machine，SVM）分类性能差以及现有AdaBoost算法中各弱分类器使用同一个参数等问题，提出基于改进遗传算法（improved genetic algorithm，IGA）优化AdaBoost-SVM的变压器多级故障诊断方法。首先，自适应调整各个SVM中的核参数σ，利用AdaBoost算法对带有调整后的核参数的SVM弱分类器进行集成；其次，构建三级集成故障分类器，并利用IGA算法优化每一级集成分类器的参数和输入特征；最后，将优化后的IGA-AdaBoost-SVM多级故障诊断模型用于变压器的故障分类中，并通过受试者工作特征曲线对不同输入特征和方法进行分析对比。仿真结果表明，所提方法能够有效改善单个SVM分类性能差的问题，且使各个弱分类器能够选择最优参数和最优输入特征，与其他方法相比具有较高的故障诊断准确率，增强了集成分类器的分类效果。

Abstract:Focusing on the problems of poor classification performance of a single support vector machine （SVM） and the fact that each weak classifier shares the same parameter in existing AdaBoost algorithms，a multilevel fault diagnosis method for transformers based on IGA-optimized AdaBoost-SVM is proposed in this paper. Firstly，the kernel parameterσ in each SVM is adaptively adjusted，and the SVM weak classifiers with adjusted kernel parameters are integrated using the AdaBoost algorithm. Secondly，a three-level integrated fault classifier is constructed，and the IGA algorithm is used to optimize the parameters and input features of each-level classifier. Finally，the optimized IGA-AdaBoost-SVM multilevel fault diagnosis model is applied to transformer fault classification，and different input features and methods are analyzed and compared using the receiver operating characteristic （ROC） curve. Simulation results show that the proposed method can effectively improve the classification performance of a single SVM，enable each weak classifier to select optimal parameters and input features，exhibit higher fault diagnosis accuracy compared with other methods，and thereby comprehensively enhance the classification performance of the integrated classifier.