Research On Fault Line Selection Based On Machine Learning And Wavelet Packet Transform

The low accuracy of grounding fault line selection in distribution network is a difficult problem that has troubled power supply companies for a long time.It is of great significance to find out the fault line quickly and accurately and reduce the outage time for the improvement of power supply reliability of distribution network.In this paper,single-phase grounding faults of a small current grounding system are taken as the research object,and based on measured recorded data and simulated data,machine learning method is applied to fault line selection from the perspectives of data migration,binary classification and multiple classification,etc.The work is as follows:(1)The wavelet packet transform method is used to extract the zero-sequence current fault features from the actual fault recording and the simulated fault waveform,and the data sample set for machine learning is obtained.The fault recording data of a110 k V substation with a 10 k V off-line for 3 years were collected and sorted as the real data samples;The model of small current grounding system was established,By setting different grounding resistance,initial phase Angle and distance,the sampling frequency was the same as the actual fault recording data,and the simulated data samples were generated.In order to reduce the dimension of the data and extract fault features effectively,the wavelet packet transform is used to extract the fault eigenvalues of zero-sequence current,and the sample sets used in the multi-classification and binary classification technology routes are generated.(2)Taking real data sample set and simulated data sample set as objects,the generalization performance of common machine learning methods is studied.Taking sample data as input,the machine learning method was trained,and the line selection results of simulated data samples and real data samples were compared by using multiple classification and binary classification methods respectively.The results show that all the machine learning methods have better effect on line selection in the case of simulated data samples,while the accuracy of line selection decreases in the case of real data samples.For the two technical routes of multi-classification and binary classification,the multi-classification method is better in the case of real data(3)samples,and the binary classification method is better in the case of simulated data.It provides a valuable reference for machine learning algorithm selection in line selection method.(3)Based on the above analysis results,a multi-classification fault line selection method based on random forest is proposed.This method makes full use of the advantages of machine learning and multi-classification in line selection accuracy,and combines the two methods to form a fault line selection scheme with high line selection accuracy,which can be realized by using computer resources in substation.The results show that this scheme has high line selection accuracy.