The Intelligent Detection Method for Arc Faults Based on Empirical Wavelet Transform and Machine Learning

Aiming at the problem that the series arc fault is not easy to accurately detect in the electric bicycle charging circuit, this paper proposed an AC series arc fault detection method based on the empirical wavelet transform method (Empirical Wavelet Transform, EWT) and the salp swarm optimization support vector machine (Salp Group Algorithm-Support Vector Machines, SSA-SVM). First, two different types of electric bicycle batteries were used as loads, and a series arc fault experiment was carried out in the electric bicycle charging circuit using an arc fault generator. Secondly, the EWT algorithm was used to decompose the loop current signal, and the sum of the harmonic amplitudes of the first four decomposed components in the frequency domain of 0-500 Hz were used as the fault features. Finally, an arc fault detection model was established using SSA-SVM, the fault features were classified. The results show that the arc fault detection accuracy of the proposed method can reach more than 90%, and the detection time is within 0.5ms. Compared with three existing arc fault feature extraction algorithms, the effectiveness of the proposed method was verified.