基于非线性主分量分析-多层感知器网络的变压器油中溶解气体数据分析

采用人工神经网络进行变压器DGA数据的分析与诊断。为全面评价变压器的实际运行状况，综合利用了各特征气体含量及其比值信息，并借鉴模糊数据处理思想构造初始输入特征集合。借助一个特殊的复合神经网络进行数据分析与故障诊断。其中，非线性主分量分析网络执行多元输入特征信息的融合及主特征选择，形成待识别故障类的敏感特征量；随后的多层感知器执行故障模式识别。试验结果表明，在DGA分析的基础上，应用非线性主分量分析－多层感知器复合神经网络可有效实现变压器不同故障模式的智能化识别，获得较好的诊断结果。

NPCA-MLP Network for Analysis and Diagnosis of DGA Data From Power Transformer

Artificial neural network (ANN) is applied to analysis and diagnosis of DGA data from power transformers. Both the quantum of characteristic gases and their ratios are used for health condition evaluation of the oiled transformers. Also, such principle as fuzzy data analysis is used for constructing multivariate input sets with original data from DGA. Finally, data analysis and fault diagnosis of the transformer are implemented by a special compound neural network, consisting of a neural principal component analysis (NPCA) and a multi-layer perceptron(MLP). The NPCA network execute information fusion and feature selection of the multivariate inputs, with which feature vectors sensitive to fault patterns to be classified are found. Subsequently, a MLP network is applied to fault patterns recognition. Results of experiment show that the proposed compound NPCA-MLP network has better performance and higher intelligence in fault diagnosis of power transformer.