Identification Method Of Fundamental Frequency Ferromagnetic Resonance Overvoltage And Arc Grounding Overvoltage In Power Distribution Network Based On Time And Frequency Domain Feature Fusion

The generation of overvoltage in distribution networks can cause damage to the insulation of transmission lines and electrical equipment,thereby affecting the safety and reliability of power grid operation.Among the types of overvoltage,single-phase ground overvoltage,arc ground overvoltage,and fundamental frequency ferromagnetic resonance overvoltage have similar characteristics,which can easily lead to misjudgment when it occurs in practice,thereby affecting subsequent fault handling.Accurate and rapid identification of overvoltage types and timely adoption of relevant suppression measures for various types of overvoltage are of important engineering significance for the safe operation of power grids.This paper takes single-phase grounding overvoltage,arc grounding overvoltage,and fundamental frequency ferromagnetic resonance overvoltage in distribution networks as identification objects,and studies their feature extraction and recognition methods.Firstly,the principles of single-phase grounding,ferromagnetic resonance,and arc grounding faults were analyzed,and the overvoltage waveforms of various faults in the 10 k V distribution network were simulated using ATP-EMTP.At the same time,Fourier transform was performed on the generated zero sequence voltage.The comparison of time-domain waveforms and frequency spectra shows that the overvoltage waveforms of single-phase grounding,arc grounding,and fundamental frequency ferromagnetic resonance all exhibit a sine wave shape,and the main frequency component is the power frequency component.Secondly,the zero sequence voltage generated by the fault is selected as the feature extraction signal.Take the half to three and a half power frequency periods after an overvoltage fault occurs in the signal as the time domain feature extraction period,and extract the peak index,impact index,and kurtosis index of this period as the time domain feature for overvoltage identification.Subsequently,complementary set empirical mode decomposition is performed on the four power frequency periods after the occurrence of overvoltage faults in the signal.Then,the Pearson correlation coefficient weighted energy entropy of each component is calculated,and the weighted energy entropy of the components corresponding to the maximum three correlation coefficients is selected as the frequency domain feature for overvoltage identification.Finally,the time-domain and frequency-domain feature quantities are respectively fed into the Back Propagation(BP)neural network optimized by the corresponding improved Particle Swarm Optimization(PSO)algorithm for preliminary identification.The output results of the preliminary identification of the two BP neural networks are preprocessed and assigned as basic probability values.After fusion using Dempster-Shafer(DS)evidence theory,the overvoltage type identification results are finally obtained.The results of testing overvoltage samples with and without noise interference show that the improved recognition method combining PSO-BP neural network and DS evidence theory effectively improves the recognition accuracy of a single classifier.