| 引用本文: | 赵文清,严海,周震东,邵绪强.基于残差BP神经网络的变压器故障诊断[J].电力自动化设备,2020,40(2): |
| ZHAO Wenqing,YAN Hai,ZHOU Zhendong,SHAO Xuqiang.Fault diagnosis of transformer based on residual BP neural network[J].Electric Power Automation Equipment,2020,40(2): |
|
| 摘要: |
| 基于传统BP神经网络的变压器故障诊断方法，当网络模型达到一定的深度时，模型的诊断性能会趋向于饱和，无法进一步提升网络模型的诊断性能，此时加深网络模型的深度反而会导致模型的诊断性能有所下降。此外，在小样本数据下，传统BP神经网络仍无法取得较好的诊断准确率。因此，为了提高变压器故障诊断准确率以及在小样本数据下的诊断性能，提出了基于残差BP神经网络的变压器故障诊断方法。所提方法采用堆叠多个残差网络模块的方式加深BP神经网络的深度，将传统BP神经网络的恒等映射学习转化为残差BP神经网络中的残差学习。同时，在每个残差网络模块中，模块的输入信息可以在模块内跨层传输，使得每个模块的输入信息可以更好地向深层网络传递，从而在小样本数据下仍可以训练得到较好的诊断模型。实验结果表明，相较于传统深层BP神经网络和传统浅层BP神经网络，所提方法具有更高的诊断准确率，同时在小样本数据下也体现出较好的诊断性能。 |
| 关键词: 电力变压器 故障诊断 残差BP神经网络 恒等映射 残差网络模块 |
| DOI：10.16081/j.epae.201912021 |
| 分类号:TM41 |
| 基金项目:国家自然科学基金资助项目(61502168) |
|
| Fault diagnosis of transformer based on residual BP neural network |
|
ZHAO Wenqing, YAN Hai, ZHOU Zhendong, SHAO Xuqiang
|
|
School of Control and Computer Engineering, North China Electric Power University, Baoding 071003, China
|
| Abstract: |
| The diagnostic performance of the transformer fault diagnosis method based on traditional BP neural network tends to be saturated when the network model reaches a certain depth, so it cannot further improve the diagnostic performance of the network model. In this case, deepening the depth of the network model will lead to a decline in the diagnostic performance. In addition, under the condition of small sample data, the traditional BP neural network still cannot achieve a better diagnostic accuracy. Therefore, in order to improve the diagnostic accuracy of transformer fault diagnosis and the diagnostic performance under small sample data, a transformer fault diagnosis method based on residual BP neural network is proposed. The depth of BP neural network is deepened by stacking multiple residual network modules, and the identity mapping learning of traditional BP neural network is converted into the residual learning of BP neural network. At the same time, in each residual network module, the input information of the module can be transmitted across layers within the module, so that the input information of each module can be better transmitted to the deep network, and then a better diagnosis model can be trained under the condition of small sample data. Experimental results show that, compared with the traditional deep BP neural network and the traditional shallow BP neural network, the proposed method has higher diagnostic accuracy and better diagnostic performance under small sample data. |
| Key words: power transformers fault diagnosis residual BP neural network identity mapping residual network module |
