China International Youth Conference on Electrical Engineering (CIYCEE 2020)

Accurately predicting the future fault level of an active distribution network(ADN) is important to the operation, maintenance, and improvement of the management level of the ADN, with higher requirement for the reliability of a power system. Considering severe weather is an important cause of ADN faults, an ADN fault levels prediction algorithm based on XGBoost for selection of fault features and prediction of fault levels of an ADN considering meteorological factors was proposed. Feature engineering was used to preprocess the ADN and original weather data and extract features; An improved recursive feature elimination algorithm was proposed to eliminate redundancy and obtain the optimal feature set through cross validation; Results of analysing calculation example showed that the proposed algorithm had an accuracy rate of 91.5% for future fault trends, which provided a theoretical basis for the follow-up fault warning and maintenance of the ADN.

Active distribution network,characteristic engineering,fault level prediction,XGBoost