摘要
分析变压器油中溶解气体含量进行变压器故障诊断的关键是找到油中溶解气体含量和故障之间的非线性关系。针对已有检测方法诊断准确性不高的问题,提出不基于Fourier变换,而是利用细分的方法构造一类新的具有加权性质的小波函数。将小波函数作为前馈神经网络的隐含层函数并优化网络的学习率,构造出加权小波神经网络处理变压器油中溶解气体含量数据。通过实际故障数据验证,此方法较已有的诊断方法准确性更高,在同等计算精度下速度更快,进而提高了变压器故障诊断的效率。
The key point of analyzing dissolved gas in transformer to diagnose transformer fault is to find out the non-linear relationship between dissolved gas content and the fault type. Using subdivision method, but not based on Fourier transformer to create a new wavelet function is put forward to deal with the problem which diagnosis accuracy is not high using traditional analysis method. The novel wavelet function combines with feed-forward neural network which learning rate is optimized to build a weighted wavelet neural network for fault diagnosis of transformers. Using weighted wavelet neural network to analyze actual fault data proves that novel neural network has higher diagnostic accuracy, and higher calculation speed under the same condition, thus can increase the efficiency of the transformer fault diagnosis.
出处
《电力系统保护与控制》
EI
CSCD
北大核心
2010年第18期19-23,共5页
Power System Protection and Control
关键词
加权小波
小波神经网络
油溶解气体
变压器故障检测
weighted wavelet
wavelet neural network： dissolved gas analysis
transformer fault diagnosis
