Research On Location And Diagnosis Method Of Partial Discharge In Power Substation

Power equipment can transform electric energy,and it can effectively receive and distribute electric power in substation.Once the power equipment fails,the substation will not operate normally,and even lead to power failure,resulting in major accidents and property losses.More than 80% of power equipment faults in substations are due to insulation defects,and the cause of insulation defect fault is mainly caused by partial discharge(PD).Therefore,accurate positioning and diagnosis of partial discharge in substations can ensure the safe and stable operation of the power grid system,which has extremely important strategic significance to national production and national economy.The key technologies for partial discharge localization and fault diagnosis of power equipment in substations involve cutting-edge technologies such as signal detection and preprocessing,localization algorithm research,feature parameter extraction,and pattern recognition.The environment of substation power equipment is complex.How to effectively detect PD signal,how to establish a more appropriate partial discharge location equation,and how to analyze and solve it correctly is a problem worth study.What kind of characteristic parameters to extract in the fault diagnosis of partial discharge and how to design a fault diagnosis classifier to match them are all issues worthy of research.In view of the above problems,this paper deeply studies the key technologies related to partial discharge localization and fault diagnosis in substations,and proposes corresponding solutions for different problems.The main research results of this paper are mainly reflected in the following aspects:1)A PD source preprocessing method is proposed based on improved S transform.When detecting a PD source signal,the detected PD source signal contains a lot of noise due to the complex and changeable site environment,which makes it difficult to obtain an effective signal.To effectively filter out complex noise signals in PD source signals,a PD source preprocessing method is proposed based on improved S transform.First,in view of the problem that the parameter selection of generalized S-transform window cannot be adaptively adjusted,an adaptive selection adjustment factor of grid retrieval method is proposed and applied to the narrowband signal filtering of PD signals.Second,aiming at the problem that it is difficult to effectively filter Gaussian white noise in the frequency range of PD signal,an optimized compact singular value decomposition(CSVD)method is proposed.The singular value decomposition is organized in a more compact way and applied to the filtering of Gaussian noise.The optimization CSVD method uses a low rank matrix to approximate the optimal solution,and the corresponding compact singular value parameters are optimized by interpolation fitting derivation method.The comparative experimental results show that the proposed improved S-transform method can effectively filter all kinds of noise signals in partial discharge signals,and this method can have a good preprocessing effect on PD signals.2)A transformer PD source localization method is proposed based on TDOA and truncated singular value regularization.Aiming at the difficulty in solving the localization equation of small-scale partial discharge in transformers,a PD source localization method is proposed based on time difference of arrival(TDOA)and truncated singular value regularization.First,starting from the inversion problem of partial discharge localization equation,aiming at the problem that the nonlinear localized equation of transformer is difficult to solve,the spherical transformation method is proposed to linearly transform the nonlinear equation system,which reduces the difficulty of solving nonlinear equations.Second,aiming at the ill posed problem of linear positioning equation,the regularization strategy for solving linear localized equations is proposed by truncated singular value decomposition,and the optimization method of regularization parameters and the overall algorithm architecture are given.Then,to reduce the ill-conditioned degree of the location equation,the coefficient matrix is processed by the balance method.Finally,the proposed localization method is simulated and verified by measured data.Theoretical and experimental results show that the proposed partial discharge location method based on TDOA and truncated singular value regularization is more accurate.The test results verify the accuracy and effectiveness of the proposed transformer PD location method.3)A PD source location method for substation is proposed based on random combination TSVD and L2 normal form distance clustering.Aiming at the problem of large error and instability of partial discharge location in large-scale substation complex environment,a new method of substation partial discharge location is proposed based on random combination TSVD and L2 normal distance.Aiming at the large-scale scene of substation,the partial discharge detection technology is studied based on ultra high frequency(UHF)antenna.UHF antenna is used to detect partial discharge signal,and time difference information is obtained by energy accumulation method.The time difference and UHF coordinate information are used to form a nonlinear equation system,and a random combination spherical model transformation method is proposed to transform the nonlinear equation into an extended linear equation.To avoid the influence of the coordinate system selection on the coefficient matrix,the coefficient matrix is preprocessed by the centralized preprocessing method.The extended linear system of equations is solved using the TSVD regularization method.To improve the accuracy of partial discharge localization,an L2 paradigm distance clustering algorithm is proposed to optimize the regularization of TSVD.The simulation results show that the maximum location error of this method is 1.10 m when the error range is 8%～10%.The maximum location error of the proposed positioning method is 1.41 m in the field substation scene.4)A partial discharge fault diagnosis method is proposed based on combinational logic and deep learning.To accurately diagnose the types of partial discharge,a partial discharge fault diagnosis method is proposed by combinatorial logic and deep learning.Aiming at the problem that a single fault feature extraction parameter can not better represent the essential characteristics of fault,a singular entropy feature extraction method is proposed based on information entropy.To further strengthen the parameter features,the concepts of kurtosis and high-order cumulants of PD signals are introduced.The combinational logic feature parameter extraction method is proposed by fusing the four feature parameters.Aiming at the problem that the traditional neural network has a simple architecture and limited feature learning ability,a deep learning algorithm with powerful learning ability is introduced,and a partial discharge fault diagnosis method is proposed based on deep belief network and SOFTMAX.Combining the proposed combinational logic feature extraction method with the deep learning fault diagnosis method,it is applied to the fault diagnosis of PD,and finally the purpose of accurate diagnosis is realized of typical partial discharge faults.