模糊神经网络在变压器故障诊断中的应用

提出了与神经网络结合的模糊变压器故障诊断新方法 ,克服了一般模糊诊断学习困难的局限 ;通过与模糊判决矩阵的对应关系 ,发现神经网络系统的权值矩阵就是模糊诊断里面的判决矩阵。模糊神经网络、组合神经网络和判决树 3种方法对故障样本的正判率分别为 90 .4 %、75 .4 %、83.3% ,这表明模糊神经网络方法的有效性与可行性 ,它弥补了DGA试验相近故障识别率低的不足 ,克服了组合神经网络无“可塑性”的缺陷 ,避免了判决树对样本选择的强烈依赖 ,使故障诊断准确度大为提高 ;也说明了DGA和其它电气试验相结合综合分析的必要

APPLICATION OF FUZZY NEURAL NETWORK TO TRANSFORMER FAULT DIAGNOSIS

Based on fuzzy neural network, a new method of power transformer fault diagnosis is proposed in this paper. The method combines fuzzy diagnosis with neural network, which solves the difficulty of fuzzy diagnosis in self-studying and finds out the physical meaning of the weight matrix with the help of the relation between the neural network weight matrix and the fuzzy decision matrix. That is to say the neural network weight matrix is the fuzzy decision matrix. The respective accuracies of fuzzy neural network, combinatorial neural network and decision tree is 90.4%, 75.4% or 83.3%. The results have validated that the correctness, the validity and feasibility of the method. The contrast analysis has pointed out the advantages of the proposed method, compared to combinatorial neural network and decision tree. It has offset the insufficiency of DGA test, overcome the limitation of combinatorial neural network which possesses no plasticity, avoided the intensive dependence of decision tree to samples, and increased the accuracy of fault diagnosis greatly. It also explained the necessity of synthetic analysis that DGA test is combined with other electrical tests such as absorptance test, dielectric loss test and so on.