Residual Current Waveform Identification Based on Gaussian Filtering and Time-Frequency Analysis

Abstract With the continuous evolution of smart-grid technologies and electrical equipment, the waveform of the residual current has become increasingly diverse under complex operating conditions. Conventional residual current protection strategies generally rely on the root-mean-square (RMS) value for fault diagnosis, which only identifies the occurrence of residual current while lacking the capability to effectively distinguish among different types of residual current waveform. Herein, this paper proposes a method which uses Gaussian filtering and condition-triggered hierarchical time-frequency analysis to identify the residual current waveform. The sampled signals are first preprocessed using an adaptive Gaussian filter (AGF) to suppress noise, followed by temporal feature extraction and decision. If time-domain evaluation results are ambiguous, frequency-domain information is further incorporated as supplementary discriminative evidence, unlike conventional fusion approaches that apply both time- and frequency-domain analyses to all signals. Experimental results reveal that the adaptive Gaussian filter method provides excellent denoising performance across various types of residual current signals, achieving a signal-to-noise ratio (SNR) improvement of approximately 8-10 dB. Compared with the schemes based either on time- or frequency-domain analysis, the proposed time-frequency fusion approach achieves the best overall identification performance after filtering, with accuracies for double-pulse DC waveform reaching 97.3% and others approaching 100%. All of these findings confirm that the proposed method is capable of significantly boosting the identification accuracy of complex residual current waveforms while maintaining low computational complexity, which is thus highly promising to offer an efficient, robust and practically applicable solution for the leakage protection and even predictive maintenance in modern power system.