Abstract: The existing partial discharge detection methods are susceptible to interference and have limited ability to capture weak signals, resulting in low detection accuracy. Propose a partial discharge online detection method based on gallium nitride transient to ground voltage detection. Firstly, based on the distribution law of electric field strength, gallium nitride sensors are deployed in the weak insulation area to capture transient ground voltage signals, and dynamically extract time-domain, frequency-domain, and time-frequency domain features, combined with improved cosine similarity algorithm and fuzzy membership function to achieve partial discharge pattern recognition. The results show that the quantization error of this method is only 1.2%~3.5% within the equivalent charge range of 20~900 pC partial discharge, and the average recognition confidence of corona, surface, and air gap discharge is 0.91, which is significantly better than the comparative method and has higher detection accuracy.