Microgrid fault classification based on random forest feature selection

To ensure the stable operation of microgrid systems and to be able to quickly perform fault isolation and recovery in the event of short-circuit faults, fault detection is crucial. To address the problems of low amount of fault features and low classification accuracy in the current microgrid fault detection, this paper proposes a fault classification scheme based on MODWT and RF feature selection. The original feature set is obtained by doing MODWT on three-phase and zero-sequence currents and extracting multiple wavelet coefficient features; the original feature set is feature selected by MIC and RF, the obtained MIC and MDI are normalized and mul-tiplied to obtain the feature importance index M-R value, and the preferred feature set is obtained by feature analysis based on the M-R value. The proposed feature scheme is compared with three existing feature extraction methods in MATLAB/Simulink model-ling and simulation, and the results show that the zero-sequence current features are crucial for fault classification; the proposed method achieves higher fault classification accuracy with fewer features, which is significantly better than other feature schemes.