Research On Key Technologies Of Partial Discharge Localization In Substation Based On Sensor Array

Substation is an important place for transformation of electric energy in the power system.The reliability of its operation directly impacts the safe and stable operation of the power system.Insulation failure accounts for up to 80% of the main failures in substations,and partial discharge(PD)is the most important cause of insulation failure.PD localization can detect equipment insulation defects in time and avoid insulation breakdown failures,which is of great significance to ensure the safe and stable operation of the power system.The electromagnetic environment of the substation is very complicated,and the PD signal detected on site is easily disturbed by noise,which makes it impossible to obtain a reliable PD signal.Time difference parameter is one of the key factors that affect the accuracy of PD localization.A small time difference error may result in a larger localization error.When solving PD localization equations,the algorithms are susceptible to time difference errors,and the stability and accuracy of the localization are hard to guarantee.Aiming at these difficulties,this dissertation expounds on the importance of PD localization in substations.The theory and key technologies of PD localization in substations based on sensor arrays data processing are studied from the aspects of PD signal denoising,time difference parameter extraction,and PD localization algorithm.The work of this dissertation is mainly in the following aspects:First,aiming at the problem that it is difficult to effectively denoise the PD signal in complex noise environment,this dissertation proposes a PD signal denoising algorithm based on multi-resolution generalized S-transform(GST).The S-transform(ST)is modified by introducing two adjustment factors ? and ?,and the time-frequency resolution characteristics of the PD signal are improved.A GST domain time-frequency filter is constructed,which suppresses periodic narrowband interference signals by designing an appropriate time-frequency filter function.By Monte Carlo experiments,the local power spectrum of white noise obeys the statistical characteristics of the chi-square distribution.The confidence interval is set by hypothesis testing method to identify and suppress white noise effectively.Simulation and experiment results show that the signalto-noise ratio of the PD signal after denoising is high,the waveform distortion is small,and the waveform transformation trend is closer to the real PD signal.Second,aiming at the difficulty in extracting the time difference of PD signals with high precision,this dissertation proposes two time difference parameter extraction methods.To extract the time difference parameters further accurately under the effective denoising situation,a time difference extraction method based on multi-resolution GST denoising and time-domain energy accumulation is proposed.Based on the PD signal denoising algorithm proposed in Chapter 3,the time difference is extracted by the energy accumulation method.To extract the time difference accurately even when the noise cannot be denoised effectively due to the same frequency interference,a new time difference extraction method based on fast S-transform(FST)and singular value decomposition(SVD)is proposed.The FST extracts the main frequency points of the signal to perform the ST,which greatly reduces the calculation amount of the ST.To eliminate the influence of white noise on the extraction time difference,SVD is performed on the FST matrix,and the singular value selection criterion is determined by the singular value difference spectrum.The time difference is extracted by finding the peak point on the FST time-domain accumulation curve.The experiment results show that the two proposed methods can accurately extract the time difference parameters under different noise environments respectively,and the time difference error does not exceed 1.01% and4.27%,respectively.Third,aiming at the ill-posed of PD localization in the large space of the substation and its low solution stability,a new method for PD localization in substations based on regularization is proposed.First,this paper discusses the inverse problem of PD localization,analyzes the factors that affect the stability of solving PD localization equations,and studies the regularization theory for solving ill-posed inverse problems.Secondly,to guarantee the uniqueness of the localization results and reduce the complexity of the solution,the nonlinear localization equations are transformed into linear localization equations by eliminating the second-order terms.Then,to reduce the ill-conditioned degree of the equations and improve the stability of the solution,the localization equations are optimized through the centralization method and the balance method.Finally,the regularization parameters are calculated by the L-curve,and the optimized localization equations are solved by the Tikhonov regularization method to obtain the position of PD source.Monte Carlo simulations show that the proposed algorithm has high stability.The experiment results show that the spatial localization error of the proposed localization algorithm is within 2.18 m,which can realize PD localization in the space of the entire substation.At last,aiming at the problem that the PD localization in a small space of the main equipment such as the transformer is sensitive to the time difference error,this dissertation presents a method for the PD localization in the transformer based on the density peak clustering(DPC).To avoid solving complex nonlinear localization equations,the localization equations after linear transformation are solved by the Gaussian elimination method.To reduce the influence of time difference error on the accuracy of PD localization,based on the research of spatial clustering analysis,multiple linear localization equations are set up to obtain multiple initial localization values and perform clustering optimization.Because of the limitations of the DPC algorithm that need to manually set the cutoff distance to calculate the local density and subjectively select the clustering center based on experience,a density peak clustering algorithm with automatic finding centers(AFC-DPC)is proposed.The cutoff distance sequence is used to calculate the local density and ? value is used to determine the cluster centers.The experiment results show that the proposed algorithm can reduce the sensitivity of the localization results to the accuracy of the time difference,with an average localization error of 5.3 cm,which achieves accurate localization of PD in the transformer.