The three-phase inverter plays a pivotal role in various fields such as modern industry, rail transit, and aerospace. However, early parametric fault diagnosis of inverter switching devices faces challenges due to redundant feature data and the subtle differences in fault features across different degradation levels. To overcome these issues, we propose a novel method called HTISFF-VSPWN for parametric fault diagnosis. Our approach involves extracting key characteristics from the three-phase voltage and current data of the inverter using the Hilbert transform. Following this, we train the model after dimension reduction. Experimental results conducted on SIC MOSFETs parametric fault data reveal that HTISFF-VSPWN outperforms other methods. Compared to 1DCNN, 1DCNN-LSTM, DRSN, IWOA-1DCNN-LSTM and MTF-SPCNN, our method achieves a diagnostic accuracy improvement of 3.69%, 2.81%, 1.6%, 0.67% and 0.92% respectively. The diagnostic time was reduced by 64s, 106s, 120s, 182s and 99s compared with the comparison method.