Characteristics of VLF/LF electric field waveforms generated by first and subsequent lightning return strokes

The accurate identification of the first and subsequent return stroke (RS) waveforms is important in the protection of power systems. In this paper, a classification model based on CNN model is proposed to accurately classify the RS waveforms. The raw waveform, wavelet transform coefficients and frequency spectra were combined as multi-channel inputs to extracts waveform features in both time and frequency domain through a multi-channel convolution kernel. Based on 6600 RS waveforms collected from Hefei, China, this study analyzed parameters like peak values, rise times, fall times, and zero-crossing times. Although the overlaps in parameter distributions make classification challenging, the proposed model successfully classified these RS waveforms with an accuracy of 87%.