A Real-Time Series Arc Fault Detection Method for Residential Buildings Based on the Arc-FcaTCN and AIBC

Abstract Series arc faults in residential low-voltage distribution systems are significant early-stage hazards of electrical fires. Due to the inherently weak and unstable nature of the current signals associated with such arcs, they are easily masked by harmonics from nonlinear household appliances and current interference from multiple loads, making it extremely difficult to achieve reliable real-time detection of series arc faults. Conventional threshold-based arc detection methods suffer from low recognition accuracy, whereas existing artificial intelligence-based algorithms fail to adequately address the challenges of arc detection in complex real-world scenarios. To overcome these limitations, this paper proposes a real-time series arc fault detection method for residential buildings based on the temporal convolutional network with frequency-channel attention (Arc-FcaTCN) and artificial intelligence circuit breaker (AIBC). The proposed Arc-FcaTCN method utilizes single-cycle current data as input and employs causal dilated temporal modeling to capture short-term disturbances caused by unstable conduction, arc extinction and reignition, and local waveform distortions near current zero-crossing points. Meanwhile, an adaptive frequency-basis representation is adapted to characterize the sensitivity of different channels to arc-related frequency responses, thereby enhancing weak frequency disturbances with fault-discriminative significance while suppressing normal load harmonics and periodic background components. Furthermore, an arc fault experimental platform designed with reference to the relevant test requirements of IEC 62606 and an AIBC prototype system are developed. Based on this, residential datasets are constructed for single-load, dual-load, and six-load scenarios. Experimental results from both offline validation and AIBC online testing indicate that the proposed Arc-FcaTCN method exhibits strong robustness under complex load-interference conditions, providing an effective and deployable solution for real-time arc fault detection and early warning of electrical fire risks in residential buildings.