Research On Power Transformer Fault Diagnosis Method Based On Machine Learning

With the unceasing speed up of power grid construction in our country,higher request is put forward for the safe and stable operation of power grid.Transformer plays an irreplaceable role in power system and its running state directly affects the reliability of power supply.In order to detect the latent faults in transformer timely and accurately,and solve the hidden faults in the running process as soon as possible,the research of transformer fault diagnosis method should be strengthened.At present,dissolved gas analysis in oil is the main method to evaluate the operation status of oil-immersed transformer.On this basis,aiming to improve the fault diagnosis accuracy of oil-immersed transformer,this thesis carried out indepth research from two aspects of improving the classification and recognition performance of Support Vector Machine(SVM)and mining the deep characteristic information contained in dissolved gas data in oil:(1)Aiming at the problem of low accuracy in transformer fault diagnosis by SVM,a transformer fault diagnosis method based on HGWO-SVM is proposed.In order to improve the global optimization ability and generalization ability of the grey wolf search algorithm,the elite opposition-based learning,nonlinear control parameters and Levy flight strategy are used to improve the grey wolf optimization algorithm,and the position update idea of the adaptive weight particle swarm optimization algorithm is integrated to optimize the search method of the grey wolf algorithm.On this basis,a hybrid grey wolf optimization algorithm(HGWO)is proposed to efficiently optimize the penalty factor and kernel parameters of SVM to ensure that SVM has the best classification performance.Then,the HGWO-SVM fault diagnosis model is constructed,and the actual fault data is used as the input of the model,and compared with the commonly used transformer fault diagnosis methods.The results show that the proposed method can improve the fault recognition rate of the transformer,which is more suitable for the field of transformer fault diagnosis,and also shows the superiority of the improved algorithm.(2)Aiming at the problem that the redundant informatio n between fault data affects the diagnosis efficiency,a transformer fault diagnosis method based on KPCA-HGWO-SVM is proposed.The kernel principal component analysis method is used to eliminate the redundant information between variables,and the data af ter dimensionality reduction is input into the HGWO-SVM model for training and classification.Different transformer fault diagnosis models are established for comparative experiments.The results show that feature extraction processing of data can effectively improve diagnostic accuracy and computational efficiency,and provide theoretical support for subsequent research.(3)Aiming at the problems that unlabeled samples generated during transformer operation are difficult to learn and utilize,the classification performance of Softmax classifier is not as good as that of SVM,and in order to further improve the performance of transformer fault diagnosis,a transformer fault diagnosis method based on SDAE-HGWO-SVM is proposed.Firstly,a feature extraction model combining Stacked Denoising Autoencoders(SDAE)and Softmax classifier is established,and batch normalization strategy and Dropout mechanism are introduced to further optimize the network,so as to mine the deep features of fault data.Secondly,the HGWO-SVM multi-classifier is introduced to construct the SDAEHGWO-SVM transformer fault diagnosis model.Finally,through simulation analysis,it can be seen that the fault diagnosis accuracy of SDAE-HGWO-SVM model is the highest,which reflects that the feature extraction ability of SDAE is better than that of KPCA,and the classification effect of SVM is better than that of Softmax,which fully proves the effectiveness,feasibility and superiority of SDAE-HGWO-SVM fault diagnosis method.