基于ART-ACO-LVQ复合算法的变压器故障诊断研究  

Research on transformer fault diagnosis based on ART-ACO-LVQ compound algorithm

作　　者：高忠江 周有为 钟雨哲 童建林 余代海 GAO Zhongjiang;ZHOU Youwei;ZHONG Yuzhe;TONG Jianlin;YU Daihai(China Metallurgical CCID Electric Technology Co.,Ltd.,Chongqing 400010,China)

机构地区：[1]中冶赛迪电气技术有限公司,重庆400010

出　　处：《电气应用》2022年第10期31-37,共7页Electrotechnical Application

摘　　要：为了提高变压器故障诊断的效率,提出了一种基于ART-ACO-LVQ复合神经网络算法的变压器故障诊断方法。首先针对ART2神经网络算法基本原理进行分析;然后利用ACO算法对ART2神经网络算法进行结合,进而对其初始权值进行优化;接着再结合有监督学习特点的LVQ神经网络算法,并提出基于ART-ACO-LVQ复合算法的变压器故障诊断方法;最后利用该复合方法对不同的变压器故障类型进行仿真分析。其仿真结果表明:通过基于ART-ACO-LVQ复合神经网络算法能够有效识别不同类型的变压器故障,对25个测试样本识别正确率达到100%,而利用ART2神经网络算法和ART-ACO神经网络算法的诊断结果正确率仅分别为68%和84%。进而表明所提的一种基于ART-ACO-LVQ复合神经网络算法的变压器故障诊断方法具有较高的正确性以及可靠性,其结果具有一定的工程实际意义。In order to improve the efficiency of transformer fault diagnosis,a transformer fault diagnosis method based on ART-ACO-LVQ compound neural network algorithm was proposed.Firstly,the basic principle of ART2neural network algorithm was analyzed.Then ACO algorithm was used to combine ART2 neural network algorithm,and the initial weight was optimized.A transformer fault diagnosis method based on ART-ACO-LVQ compound algorithm was established by combining LVQ neural network algorithm with supervised learning characteristics.Finally,different transformer fault types were simulated and analyzed by this compound method.It can be obtained that the different types of transformer faults can be effectively identified by ART-ACO-LVQ compound neural network algorithm.The recognition accuracy of this method is 100%for 25 test samples.The diagnostic accuracy of ART2 neural network algorithm and ART-ACO neural network algorithm are only 68%and 84%,respectively.Therefore,the proposed transformer fault diagnosis method based on ART-ACO-LVQ compound neural network algorithm has high correctness and reliability.And the results have certain engineering practical significance.

关 键 词：故障诊断 自适应共振 蚁群算法 学习矢量量化算法 复合神经网络 

分 类 号：TM41[电气工程—电器] TP18[自动化与计算机技术—控制理论与控制工程]