Pulse-type interference at high-voltage (HV) and ultra-high voltage (UHV) transformer sites severely limits the effectiveness of partial discharge (PD) detection, as conventional pulse current measurement below 1 MHz leaves 12-pulse ripple, corona, and electronic switching pulses overlapping with genuine PD signals in both time and frequency domains. This work develops an improved wavelet threshold method tightly integrated with a wideband detection platform featuring a 150 kHz to 60 MHz signal conditioning bandwidth coupled with a 3–50 MHz high-frequency current transformer (HFCT) sensor. A continuously differentiable threshold function with adjustable factor α is constructed to bridge hard and soft thresholding, a layer-dependent adaptive estimator replaces the universal threshold to match noise variance at each decomposition level, and a time-frequency cumulative energy function performs pulse pre-localization before discrimination. Validation on 500 simulation samples, 320 laboratory records from a 220 kV physical platform, and 180 field signals from three substations shows an average SNR gain of 9.3 dB, combined fitness F of 2.38 at 5 dB input SNR (F is a figure of merit balancing noise suppression and waveform preservation), and laboratory identification rates of 95.2%, 93.6%, and 94.3% for the three interference categories. Pooled field detection accuracy reaches 93.7%, feature-parameter recovery errors stay within 8.1%, and single-record processing settles at 52 ms. Pilot deployments reduced the detection blind zone from 6%–8% down to 2.5%–3%. These findings demonstrate the practical value of the proposed method for improving the reliability of online insulation condition assessment in harsh substation environments.
