Switching error detection of rheological tap-changer for UHVDC transmission based on voice print feature

• Propose a non-intrusive detection method for identifying switching errors in UHVDC converter transformer tap-changers based on voiceprint features • Optimize acoustic signal processing techniques for enhanced tap-changer operation state identification • Construct a misoperation feature vector library to characterize abnormal tap-changer behaviors • Achieve stable switching error detection to enhance UHVDC transmission system safety Traditional error detection methods primarily rely on the measurement and analysis of current, voltage, and other electrical signals; however, these approaches exhibit limitations in addressing fault detection in complex power systems. To overcome this challenge, a UHVDC on-load tap-changer fault detection method based on voiceprint features is designed. The electrical stress characteristics of the on-load tap-changer in the UHVDC system are analyzed, and a misoperation feature vector library is constructed. The db3 wavelet basis function is employed for discrete wavelet transformation, and the original time-domain acoustic signal is segmented into frames, windowed, and processed using short-time discrete Fourier transform to obtain a time-frequency spectrogram. The Mel-frequency cepstral coefficient (MFCC) is utilized to characterize the voiceprint features of the UHVDC converter transformer, and a Transformer neural network is applied to detect tap-changer faults in the UHVDC converter transformer. Experimental results demonstrate that the stability index of the voiceprint features obtained by the proposed method exceeded 0.6, approaching 0.9, with low error. The training set loss rate decreases steadily, and the detection time is shorter, consistently below 3 s.