Abstract: Noise can cause distortion in the waveform of partial discharge signals, resulting in changes in signal characteristic parameters and affecting the accuracy of identifying partial discharge anomalies. Therefore, a method for automatic identification of partial discharge anomalies in the winding of the converter valve side based on interference signal FIR low-pass filtering algorithm is proposed. The FIR low-pass filtering algorithm is used to process the collected partial discharge signal, and the signal is converted to the frequency domain through Fourier transform. The frequency domain filtering method is used, combined with the filtering coefficient and sampling interval time, to achieve low-pass filtering. Obtain IMF components from the denoised discharge signal through EMD decomposition, and remove IMF with smaller energy values. Calculate the correlation coefficient and sensitivity factor of IMF, and use the energy threshold method to determine the authenticity, in order to extract the characteristics of partial discharge signals. Using convolutional neural networks for anomaly recognition of processed signals, automatic recognition of partial discharge anomalies is achieved through convolutional kernel operations, pooling dimensionality reduction, and fully connected classification. The experimental results show that the proposed method can identify different types of partial discharge signals. The standard deviation of the recognition result is 1.1 signals per cycle, which can accurately identify the partial discharge phenomenon that occurs within each hour. In 600 samples, the proposed method accurately identified the type of partial discharge, with high recognition accuracy and an accuracy rate of 100%.
? Keywords: Interference signals; FIR low-pass filter; Valve side windings of converter transformers; Partial discharge; Identification method;