Pulse Extraction from Signals Sensed at Neutral Point of Large Hydro Generators

The difficulty of online monitoring of fault discharge of large hydro generators lies in the processing of monitoring signals intertwined by noise, interference and various fault discharge pulses. Pulsed high-fidelity extraction is a critical step. The pulse extraction method based on sliding window energy in the time domain is fast and has a good effect on the oscillating pulse, but the extraction effect will deteriorate rapidly when the signal-to-noise ratio is low. In addition, the background noise intensity, as a key parameter, relies on manual selection of appropriate data segments for estimation. The number of sliding window points, as a key parameter, relies on manual acquisition of the pulse oscillation period. In this paper, an iterative estimation method of background noise intensity is used to automatically determine the background noise intensity. In this paper, the signal is transformed into wavelets, and the wavelet domain sliding window is used to extract the decomposition coefficient of the pulse, and the number of sliding window points does not need to manually obtain the pulse oscillation period. The pulse extraction results of both simulated and measured signals show that the proposed method has a better pulse extraction effect than the wavelet threshold denoising, and has a better and more stable pulse extraction effect than the pulse extraction method based on time-domain sliding window energy.