Detection Algorithms of Parallel Arc Fault on AC Power Lines Based on Deep Learning Techniques

Several studies on arc fault detection have been recently conducted. The arc fault is detected by analyzing the frequency- and time-domain current characteristics of the arc in the AC parallel arc fault. In this study, the focus was on detecting AC arc faults using artificial intelligence concepts. The detection performance was analyzed by comparing different combinations of input feature parameters and neural networks. In particular, the performances of the input parameters were compared and analyzed, including the frequency average, instantaneous frequency, entropy, fast Fourier transform and the maximum slip difference combination, and the FFT and frequency average combination. Different combinations of parameters and neural network structures were applied to the respective parallel AC enclosed case and unenclosed case, and the performances were compared. It was determined that the combinations of two input parameters should be applied to achieve high performance in both enclosed and unenclosed cases. In addition, the detection rate with respect to the amount of training data was analyzed. The combination of two input parameters improves the robustness and reliability of arc fault detection.

Automated summary

This study focuses on detecting AC arc faults using artificial intelligence concepts. The detection performance is analyzed by comparing different combinations of input feature parameters and neural networks. It is found that the combination of two input parameters improves the robustness and reliability of arc fault detection in both enclosed and unenclosed cases.