Human electrocution recognition method based on residual network

Human electrocution presents a substantial threat to public safety. Prevailing techniques for detecting human electrocution, however, suffer from limited accuracy and extended detection timeframes. To address these challenges, we introduce a method for human electrocution detection based on a Residual Network 18 (ResNet18). First and foremost, we analyze the transient amplitude disparities and steady-state phase distinctions in residual current in the case where human electrocution and resistive leakage occur, incorporating the International Electrotechnical Commission (IEC) human impedance model with the low-voltage distribution network topology. Subsequently, we investigate a feature extraction method for fault detection based on fault phase voltage and residual current within a 2-millisecond window after the fault occurrence, resulting in a trajectory-based human electrocution detection method rooted in voltage-current patterns. On these grounds, the sampled signals traversing fault initiation and fault phase selection components are utilized to construct voltage-current trajectories, which act as inputs for ResNet18 during model training or fault identification. In closing, we validate the proposed algorithm through simulation models implemented in Matlab/Simulink, illustrating its superior identification accuracy and substantially reduced detection time when compared to existing methods.