Series Arc Fault Identification Method Based on Multi-Feature Fusion

With the increase of various loads connected to the low-voltage distribution system, the difficulty of identifying low-voltage series fault arcs has greatly increased, which seriously threatens the electricity safety. Aiming at such problems, a neural network algorithm based on multi-feature fusion is proposed. The fault current has the characteristics of randomness, high frequency noise, and singularity. A GA-BP neural network model is built, and the wavelet analysis method (based on singularity), Fourier transform method (based on high frequency noise), current cycle difference method (based on randomness), and current cycle similarity derivation method (based on randomness) are used for feature extraction and can more comprehensively reflect the characteristics of arc faults. Simulation results show that the multi-feature fusion algorithm has a higher recognition rate than other algorithms. Moreover, compared with the support vector machine model, logistic regression model, and AlexNet model, the GA-BP neural network model has a higher recognition accuracy than the other three models, which can reach 99%.

Automated summary

A neural network algorithm based on multi-feature fusion is proposed to identify low-voltage series fault arcs, with simulation results showing a higher recognition rate than other algorithms. Compared to support vector machine, logistic regression, and AlexNet models, the GA-BP neural network model achieves a recognition accuracy of 99%.