Research On Line Selection Method For Single-Phase-to-Ground Fault In Distribution Network Based On CatBoost

With the increasingly complex structure of my country’s distribution network,the requirements for power supply reliability are also increasing.At present,in the low-current grounding system widely used in my country,the frequency of single-phase grounding faults is the highest,accounting for about 80% of the total number of faults,making the problem of fault line selection after a singlephase-to-ground fault in the distribution network become one of the research hotspots.Due to the continuous input of cable lines,the compensation function of arc suppression coils,and the changes in the operation mode and structure of the distribution network,the difficulty of fault line selection is greatly increased.At present,it is difficult to guarantee the reliability and accuracy of line selection under different working conditions by relying on the traditional fault line selection method of manually extracting fault features.Therefore,by introducing machine learning algorithms,this paper transforms the problem of fault line selection into a multi-classification problem in machine learning,and studies a single-phase-to-ground fault line selection method for distribution network based on Cat Boost.Firstly,this paper conducts a detailed theoretical analysis of the transient and steady-state fault characteristics after a single-phase ground fault occurs in a small-current grounding system by constructing an equivalent circuit.It is concluded that the transient process after the fault contains more abundant fault information.Research on fault line selection based on transient zero-sequence current.It is concluded that the transient process after the fault contains more abundant fault information,so the Research on fault line selection based on transient zero-sequence current.Use MATLAB/Simulink to build the distribution network model,set different fault conditions for fault simulation,compare the simulation results with the theoretical analysis,and verify the accuracy of the built distribution network model,providing a reliable source of data for subsequent research on line selection.Secondly,using the advantages of ensemble learning in solving classification problems,a fault line selection method based on the Cat Boost algorithm is studied,which takes the transient zerosequence current as the feature vector.Aiming at the problem of noise interference in the actual distribution network,the wavelet threshold denoising method is selected,and the appropriate wavelet basis function and decomposition layers are determined through denoising simulation experiments,and then the data preprocessing of the zero-sequence current signal is completed,and the establishment of The fault line selection model based on Cat Boost is established,and the parameter optimization method of the model is determined.Since Cat Boost has the advantage of automatic classification features,it solves the problems of high complexity of manual feature extraction and inaccurate line selection using a single fault feature during fault line selection.Finally,based on the distribution network simulation model built in this paper,the sample data sets under different fault conditions are obtained by writing MATLAB scripts for automatic simulation,and the accuracy of the Cat Boost line selection model established in this paper is verified.Random forest,The XGBoost and Light GBM algorithms respectively established line selection models for comparative analysis,using the accuracy rate,F1-score(Weighted)and kappa coefficient as the model evaluation indicators.Realize high-precision line selection,and have better model performance than other line selection models.At the same time,a number of different test sets were established to test the applicability of the Cat Boost line selection model for different arc suppression coil compensation degrees,distribution network structure changes,high-impedance grounding at the end of the line,and asynchronous sampling,and the Cat Boost line selection model was verified.While having high line selection accuracy,it also has strong generalization ability and robustness.