Abstract:

In view of the existence of redundant fault characteristic variables and non-strongly correlated variables in the initial feature subset of active distribution network risk, a feature selection method based on VarianceThreshold-GARFECV is proposed. The proposed method combines the variance threshold and the recursive feature cancellation cross-validation(RFECV) technology based on genetic algorithm, which can effectively select the optimal feature set. Experimental results show that the proposed method can screen and select the initial feature set of distribution network fault risk, and eliminate the characteristic variables with weak correlation and redundancy, so as to reduce the complexity of distribution network data, avoid overfitting, and increase the interpretability of the model, with high accuracy and stability.

Key words: feature selection, situational awareness, risk prediction, VarianceThreshold