GIS Surface Discharge and Metal Particle Size Identification Based on Silicon Photomultiplier and Lightweight CNN

Metal particles in gas-insulated switchgear (GIS) may attach to the insulator surface and cause surface discharge (SD), which can lead to flashover failure and threaten the stable operation of power grid. To develop an efficient method for SD detection and risk assessment, an optical pulse measurement module based on silicon photomultiplier is designed, and SD experiments of metal particles with different sizes are carried out on the basin insulator of a real 126 kV GIS device. By analyzing the SD inception voltage, time-domain and phase characteristics of optical pulses, a novel method for SD and particle size identification is proposed by combining phase resolved optical pulse pattern and lightweight convolutional neural network. Test and comparison experiments demonstrate that the proposed method exhibits superior generalization and real-time performance. The identification accuracy of SD sample by this method is 100%, and the identification accuracy of particle size is 98.52%. The average runtime of the proposed lightweight identification model in the embedded device is 173.96 ms. Finally, various interpretability techniques are combined to analyze the inference mechanism of the identification model and the feature importance of the optical pulse phase pattern samples.