Fault Arc Detection Based on Channel Attention Mechanism and Lightweight Residual Network

An arc fault is the leading cause of electrical fire. Aiming at problems difficulty in manually extracting features, poor generalization ability models and low prediction accuracy traditional detection algorithms, this paper proposes a method based on fusion channel attention mechanism residual network model. This to perform global average pooling information from each feature map assigned by block while ignoring local spatial data enhance recognition rate arc. introduces one-dimensional depth separable convolution (1D-DS) module reduce model parameters shorten time single samples. The experimental results show that F1 score for under mixed load conditions 98.07%, parameter amount reduced 46.06%. proposed dramatically reduces quantity, floating-point number complexity structure ensuring high rate, which improves real-time response detect fault. It has guiding significance applying edge side.