The Research On Diagnosis Method Of Power Transformer Based On Deep Learning And DGA

The safe and stable operation of power transformers is of great significance to the entire power system.Fault diagnosis technology is an important means to ensure the safety of power transformers.Dissolved gas in oil is closely related to the operating status of the transformer.Dissolved Gas Analysis(DGA)is also one of the important techniques for fault diagnosis of power transformers.At present,how to extract feature quantities from the original DGA data that are useful for characterizing various operating states of power transformers,and how to perform fast and accurate fault classification on the obtained feature quantities has become the main content of researching power transformer fault diagnosis technology.In recent years,with the development of computer technology and artificial intelligence,feature extraction and pattern recognition,mainly based on deep learning theory,have shined in the field of fault diagnosis of power field.Its core ideas mimic human brain thinking method,by constructing a deep neural network to mine and learn the deep-level features contained in the original data.In this context,this paper provides a new solution for power transformer fault diagnosis by studying the deep learning related theories and DGA data.Aiming at the problem that the original DGA data feature expression ability of power transformers is insufficient,the fault diagnosis effect is poor.Based on deep learning theory,a DGA data feature dimension enhancement method based on deep auto-encoders(DAE)is constructed to expand the feature dimension of various types of DGA fault data.Compared with the traditional method,the DAE method has less manual intervention,and the extracted high-dimensional features have stronger ability to express features.For the traditional intelligent diagnosis algorithms,a series of problems such as lack of accuracy are prone to occur.Based on deep learning theory,a powerful feature learning capability of convolutional neural network(CNN)is constructed to build a CNN fault diagnosis model for power transformers.Considering that CNN can be used to extract not only 1-dimensional time-domain features but also 2-dimensional space-domain features,this paper establishes 1D-CNN and 2D-CNN power transformer fault diagnosis models,and verifies the validity of the two models through examples.In order to further improve the accuracy and reliability of the diagnosis results,a two-stream CNN,combination of 1D-CNN and 2D-CNN,power transformer fault comprehensive diagnosis model,is constructed based on the combination of information fusion technology and deep learning theory.This model can use two-stream CNN to perform fusion training on feature data obtained by different feature extraction methods.During the training process,two-stream CNN continuously updates the parameters of two CNNs,which is more robust than traditional information fusion methods.The example proves that the two-stream CNN comprehensive diagnosis model can effectively improve the fault diagnosis accuracy of power transformers.