Kernel PCA与BP神经网络相结合的变压器故障诊断*

为了提高变压器故障诊断的准确率和抗干扰能力，提出一种基于核特征量的BP神经网络故障诊断模型。通过核主成分分析将故障样本从低维的特征空间非线性地映射到高维的核空间，提高了样本的可分性，然后以核特征量作为BP神经网络的输入特征量，建立变压器故障诊断模型。实验对比了结构相似、输入量不同的BP神经网络，结果表明采用核特征量的诊断模型具有更好的诊断效果和抗干扰能力。

Transformer fault diagnosis based on kernel PCA and BPNN

To improve the accuracy and anti-noise ability for transformer fault diagnosis,this paper proposed a novel BP neural network based on kernel features.Samples were nonlinearly mapped from the low-dimensional feature space into the high-dimensional kernel space by kernel PCA,which improved samples'separability,then built BP neural network in kernel space.Experiments compared performances of BP neural networks with silimar constructures,different input features,the results show that diagnostic models with kern...