Research On Fault Diagnosis Method Of Oil-immersed Transformer Based On Deep Learning

Oil-immersed transformers are the most important equipment in the process of power transmission and distribution.The reliability of the power grid depends largely on the trouble-free operation of the equipment.At present,the most commonly used method for evaluating the operating conditions of oil-immersed power transformers is Dissolved Gasin-oil Analysis(DGA).Based on the analysis of the existing oil-immersed transformer fault diagnosis methods,this paper applies the convolutional neural network in deep learning to transformer fault diagnosis,establishes the mapping of input feature vectors and output fault categories,and then uses deep learning in transformer faults.Advantages in diagnosis,the main work done includes:(1)The oil-immersed transformer collects dissolved gas data through oil chromatography,but the actual collected fault data is less,and there is no label data.Aiming at the problem of insufficient expression ability of original data,a feature vector expansion method based on gray relational analysis is proposed,which scientifically and standardizedly extracts feature vectors that are more conducive to the discovery of power transformer faults.(2)The accuracy of the existing fault diagnosis methods is not high,and the generalization ability of the model is not strong.In response to this,a transformer fault diagnosis model based on convolutional neural network is constructed.And the BN layer is added to the network to prevent the problem of overfitting during model training.Finally,the number and size of the convolution kernel in the neural network,the pooling method of the pooling layer and the number of model iterations are analyzed and discussed.Experimental results show that the convolutional neural network has powerful feature extraction capabilities,which can further improve the accuracy of fault diagnosis.(3)The hyperparameters of the convolutional neural network have a greater impact on the experimental results.In response to this problem,a Convolutional Neural Network Based on Genetic Algorithm(GA-CNN)model is proposed.Introducing the genetic algorithm(GA)into the model to optimize the weights and bias factors of the neural network can effectively prevent the network from falling into the local optimal solution,and use K-fold cross-validation to solve the problem of the model Instability issues.(4)compare support vector machine,BP neural network and GA-CNN experimentally.The experimental results show that the overall performance of GA-CNN is better than that of traditional neural networks,the fault diagnosis error is reduced,the model is more robust,and the prediction accuracy fluctuates less.