Research On Fault Diagnosis Methods For Oil-immersed Transformers

As the most important equipment in the process of power transmission and distribution,the reliability of power grid operation largely depends on the fault-free operation of the equipment.Therefore,the detection of transformer operation status is of great significance.Based on the current research status,dissolved gas analysis is the most commonly used method to measure the fault status of oil immersed transformers.In this paper,the dissolved gas content in oil is used to construct a transformer fault feature vector,and the long and short term memory network(LSTM)is introduced to establish a functional relationship between the transformer fault feature vector and the transformer fault type,thereby obtaining a transformer fault diagnosis method based on deep learning.Aiming at the abnormal values contained in the dissolved gas data in the oil of oil-immersed transformers,this paper establishes a data cleaning method based on Isolated Forests,which segments the original data and realizes the abnormal value cleaning process of the original data.The method of maximum normalization is used to standardize the cleaned data.The training data and test data required for this experiment are established.To solve the problem of difficult selection of super parameters for Long Short-Term Memory(LSTM)networks and difficulty in constructing models with high classification accuracy,an Improved Sparrow Search Algorithm optimized for LSTM networks(ISSA-LSTM)transformer fault diagnosis model is proposed.Firstly,the Sparrow Search Algorithm(SSA)is improved using a set of good points and adaptive weights to overcome its shortcomings of slow convergence speed and easy falling into local optimization.Then,multiple test functions are used to verify the effectiveness of the Improved Sparrow Search Algorithm(ISSA).Finally,the ISSA is used to optimize the LSTM network,and the improved model is obtained.The experimental results demonstrate that the established model can effectively diagnose faults in transformers,providing a guarantee for the healthy operation of transformers.Aiming at the shortcomings of LSTM that are insensitive to spatial features and incomplete in feature extraction,a dual attention transformer fault diagnosis model is proposed in this paper.Firstly,the original Convolutional Neural Network(CNN)is improved by using spatial attention mechanism and dilated convolution.Then,a temporal feature weighting mechanism is assigned to the temporal and spatial memory network to obtain the temporal attention weighted LSTM.Finally,the spatial feature extraction mechanism and temporal feature extraction mechanism are fused to obtain a dual attention transformer fault diagnosis model.The results show that the dual attention transformer fault diagnosis model considering both spatial and temporal characteristics has high diagnostic accuracy for transformer fault diagnosis.To obtain the transformer fault diagnosis model with higher diagnostic accuracy,the ISSA is introduced into the dual attention transformer fault diagnosis model,and a dual attention model optimized by the ISSA is established.The experimental results show that the hyperparametric optimization of the dual attention model can make more accurate and reliable diagnosis of transformer fault information,and has certain engineering practical value.