基于神经网络的变压器故障诊断系统

Transformer Fault Diagnosis System Based on Neural Network

针对变压器故障征兆和故障类型的非线性特征,结合油中气体分析法,采用改进后的BP神经网络进行故障诊断。经统计分析和数据预处理确定网络输入,使用择优选取法优化网络结构参数,使训练速度和误差精度达到较满意的效果。并基于此算法用JAVA语言开发了小巧实用的故障诊断系统,能够较好地完成网络训练、故障诊断和数据维护等功能。

In view of non-linear characteristics between fault symptoms and fauh types of transformers, an improved BP neural network is used to diagnose transformer fault with the data of dissolved gas analysis （DGA）. After pretreatment of input data and optimization of parameters of network structure, a satisfactory result is obtained in training speed and diagnosis accuracy.Based on the algorithm, a fault diagnosis system is developed in JAVA language which is able to complete the functions of network training, fault diagnosis and data maintenance.