Research on Arc Fault Diagnosis Technique for Series Faults in Low Voltage DC Multi-Load Circuits

For the problem of series arc fault detection in low-voltage DC multi-load system, this paper builds a DC arc fault experimental platform according to UL1699B standard, simulates a multi-operating condition mixed load scenario, collects the system normal and fault state current signals, combines the wavelet threshold noise reduction with the improved adaptive noise-complete empirical modal decomposition (ICEEMDAN) to process the signals, and dynamically optimizes the decomposition parameters through the Gray Wolf Optimization (GWO) algorithm to extract the envelope entropy feature of the intrinsic modal function (IMF). Grey Wolf Optimization algorithm (GWO) is used to dynamically optimize the decomposition parameters and extract the envelope entropy features of the intrinsic modal function (IMF). The GWO-CNN-BiLSTM-Attention hybrid model is constructed, and repeated experiments show that the accuracy of this method in fault identification under mixed load conditions reaches $98.03 \%$, which is greatly improved compared with the traditional CNN and BiLSTM models, and it can effectively identify DC arc faults under mixed load line scenarios, and it provides a high-precision detection solution for arc protection in DC systems.