Research On Transformer Fault Diagnosis Method Based On Improved Whale Algorithm Optimized BiLSTM

Power transformer is the key equipment of the power grid,and its operation status directly affects the safety and stability of the whole power system,so it is of great significance for the research of transformer fault diagnosis.Based on the backgroun of dissolved gas analysis(DGA),a transformer fault diagnosis method for eigenmetric parameter dimensionality reduction and improved whale algorithm optimized bidirectional long short term memory network(Bi LSTM)is proposed in this thesis.Firstly,data for high-dimensional redundancy is not conducive to fault diagnosis,the transformer fault diagnosis model of L-Isomap and Bi LSTM is proposed.Based on the Bi LSTM fault diagnosis model,L-Isomap is used to reduce the dimensionality of fault characteristic parameters,and it is compared the diagnostic accuracy and computational efficiency with the original feature quantity and Isomap dimensionality reduction characteristic quantity.The effectiveness of L-Isomap is verified.The eigenvalues of L-Isomap dimensionality reduction are used as inputs to Support Vector Machines(SVM),Extreme Learning Machine(ELM),Back Propagation Neural Networks(BPNN)and Bi LSTM fault diagnosis models and these models are trained repeatedly multiple times.The advantages and good stability of the Bi LSTM model are verified.Secondly,aiming at the fact that the Bi LSTM hyperparameters will affect the accuracy of fault diagnosis,the Mixed-strategy Improved Whale Algorithm(MIWOA)is proposed to optimize the transformer fault diagnosis model of Bi LSTM.Adaptive weights and nonlinear convergence factors are introduced,local search mechanisms of bat algorithms is introduced to perturb local optimal solutions,and Levy flight strategy is introduced to improve the logarithmic spiral update position.10 standard functions are compared with the traditional optimization algorithm to prove that MIWOA has the best performance.Using MIWOA to find the optimal parameters of the Bi LSTM model,the MIWOA-Bi LSTM fault diagnosis model is established.Thirdly,in order to verify the effectiveness of the MIWOA-Bi LSTM fault diagnosis model,the feature parameters of L-Isomap dimensionality reduction are entered into the PSO-Bi LSTM,GWO-Bi LSTM,WOA-Bi LSTM,and MIWOA-Bi LSTM models,respectively.The fault diagnosis accuracy is 84.2%,86.7%,89.2%,and 95%,respectively,and the results showed that the MIWOA-Bi LSTM model has outstanding advantages.Therefore,the method proposed in this thesis can effectively improve the accuracy and reliability of fault diagnosis,and provide a new idea for transformer fault diagnosis research.The thesis has 52 pictures,24 tables and 71 references.