Research On Recognition Method Of Transformer Magnetizing Inrush Current Based On EO Optimized ELM

Transformers are vitally important electrical equipment in the power system.The protection of transformers is also one of the key concerns of the power sector.Among them,the transformer differential protection is the most important.When a fault occurs in the protection zone,the differential protection performs a protection action and quickly removes the fault;and the magnetizing inrush current generated when the voltage is restored after the no-load closing of the transformer or the external fault is removed will cause the transformer differential protection to malfunction and cause the power system Abnormal power failure;therefore,the accurate distinction between the inrush current of the transformer and the fault current can provide an important reference for the setting of the differential protection.Based on abundant domestic and foreign data,this paper summarizes the current common methods of identifying inrush current and fault current and analyzes their advantages and disadvantages.Aiming at the shortcomings of commonly used methods of identifying inrush current,this paper proposes "signal decomposition-feature extraction-establishment of recognition the research idea of "model-model verification".First,a simulation model of transformer magnetizing inrush current and fault current based on Matlab/Simulink platform is established,and then the simulation waveforms of magnetizing inrush current signal and fault current signal are obtained;the signal decomposition model evaluation method of modal aliasing analysis is adopted to compare and analyze the Empirical Mode Decomposition(EMD),Ensemble Empirical Mode Decomposition(EEMD)and Variational Modal Decomposition(VMD)are three modal decomposition methods.VMD is preferably used for modal decomposition of the current signal waveform obtained by simulation to obtain eigenvalues of different scales Modal function:Aiming at the difference of the entropy values of the two signal samples,a sample entropy feature extraction model is established to extract the feature vector of the eigenmode function and form a feature matrix as the input data of the recognition model.Then,through theoretical analysis of the three recognition algorithms of Probabilistic Neural Network(PNN),Support Vector Machine(SVM)and Extreme Learning Machine(ELM),ELM is preferred as the recognition model of this article;the initial weight and threshold of ELM may be randomly assigned in order to reduce the recognition accuracy,an algorithm for optimizing ELM using the Equilibrium Optimizer(EO)is proposed.The optimal solution obtained from optimization is assigned to the initial weight and threshold of the ELM,and then the current signal is trained and recognized to improve Recognition accuracy rate.Eventually,the established model is verified through experimental comparison.The results show that the EO optimized ELM model built in the article has the advantage of high recognition accuracy,which provides a technical reference for the identification of inrush current and fault current and the setting of transformer protection.