Research On The Fault Diagnosis Of Transformer Based On Generating Adversarial Network Data Enhancement And Vibration Signal

As the pivotal equipment of the power system,the safe and stable operation of the transformer has an important influence on the high-quality and effective power supply of the system.However,in actual production,the probability of transformer fault events is low,resulting in a lack of fault data,which brings difficulties to fault diagnosis research based on deep learning.In view of this,this paper studies the data augmentation based on generative adversarial network,and uses auxiliary classification generative adversarial network to expand the transformer imbalance data set to improve the accuracy of fault diagnosis.The main research work is as follows:(1)The occurrence mechanism and types of main faults(thermal faults and electrical faults)of transformers are expounded in detail;the aging factors of transformers and the existing diagnostic methods are analyzed,and the status quo of various detection methods are expounded,including offline detection and online detection;Transformer vibration analysis test is studied,on this basis,combined with actual data,the transformer operating conditions are analyzed,the basic principle of generative adversarial network and its improved variants are explained,the advantages and disadvantages are analyzed,that is,the application is summarized.(2)A data augmentation method based on a two-dimensional auxiliary classification generative adversarial network(ACGAN)is proposed.A dedicated zero-padding layer is added to the discriminator and generator to expand the feature map extracted by convolution and improve the The ability to mine deep features;the Wasserstein distance is used as the loss function,and gradient penalty is introduced to replace weight clipping to improve training stability;the end of the discriminator uses a fully connected layer to reduce the impact of feature positions on classification,and introduce maximum Pooling layer and average pooling layer to improve the adaptability of the discriminator in different scenarios.The original unbalanced dataset is augmented with the data generated by the model,in order to improve the accuracy of transformer fault diagnosis.The simulation results show that,compared with the ACGAN framework before the improvement,the training process of the improved model is more stable,the quality of the generated data is further improved,and the accuracy of fault diagnosis can be improved.(3)On the basis of improving the ACGAN framework,the self-attention mechanism is introduced into the generator and the discriminator to fully capture the global features of the data,improve the learning ability of the model,introduce gradient penalties,overcome problems such as gradient disappearance,and speed up the model.convergence.It is verified on the transformer vibration test data set.The results show that the improved model has a faster convergence speed,and the generation ability of the generator can be further improved.Using the generated data of the improved model to expand the unbalanced sample set can significantly improve the fault diagnosis accuracy.