Harmonic Fault Diagnosis Of Autotransformer Based On Compressive Sensing Theory

In recent years,the coupling oscillation problem of vehicle network caused by high harmonics has become more and more serious.As the core equipment of traction power supply system,the autotransformer(AT)produces a large number of harmonics in the operation process of AT,which seriously threatens the safe operation of high-speed train due to the phenomenon of"DC bias magnetism"and"leakage reactance".However,due to the defects of Nyquist sampling theorem,it is impossible to sample the higher harmonics generated by AT,and the data after sampling grows exponentially.Therefore,based on the compressed sensing(CS)theory,the acquisition,reconstruction of autotransformer harmonic signals and fault diagnosis are completed.Firstly,based on the current research situation of traction power supply systems,this paper analyzes AT theoretically and builds a simulation model.However,due to the magnetic coupling between windings,numerical oscillations occurs during simulation and causes the image distortion.Therefore,a new damping ladder algorithm is proposed to suppress the numerical oscillation.Firstly,the damping factor?is estimated,then the auxiliary damping factor?is added for correction,and finally the excitation current _ni is compensated for oscillation.The simulation results show that the established model can clearly reflect the harmonic waveform of AT when it is in operation,providing a theoretical basis for fault diagnosis when the AT appears harmonics.Secondly,by studying and analyzing the current situation of CS theory at home and abroad,and comparing the accuracy and time of existing reconstruction algorithms,a new type of spectral projected gradient with line interpolation correction(SPG-LIC)algorithms is proposed to reconstruct harmonic signals with inter-harmonics.This algorithm first for the short-range spectrum leakage,the windowed fast Fourier transform is used to suppress it and make the transformed signal have high sparsity,Then the multi-line interpolation correction algorithm is introduced to suppress the long spectrum leakage and gate effect.By predicting the sparsity of the signal,the original signal is reconstructed using the spectral projection gradient algorithm and compared with other reconstruction algorithms.The results show that SPG-LIC algorithm has higher SNR and accuracy,and shorter reconstruction time than other reconstruction algorithms.Finally,a harmonic fault diagnosis method based on the gradient information of least square method support vector machine(G-LS-SVM)is proposed for the harmonic signal reconstructed by SPG-LIC algorithm.This method first uses a double-sparse dictionary model to construct an over-complete dictionary containing harmonic signals in each frequency band,then builds a G-LS-SVM fault diagnosis model,and finally the feature vectors of harmonic signals in each frequency band are input into G-LS-SVM for experimental simulation.The experimental results show that when the over-complete dictionary contains the feature vectors of harmonic signals in each frequency band,the model can accurately identify the harmonic signals in each frequency range,and the G-LS-SVM model has strong anti-noise ability and strong robustness.