Abstract: To overcome the limitations of traditional single-detection methods in gas-insulated switchgear (GIS) partial discharge (PD) detection and severity assessment, this paper proposes an innovative method based on optical-electrical-field multiphysical fusion. A multisensor synchronous detection system integrating optical, ultra-high-frequency (UHF) electromagnetic, and mechanical vibration signals was designed, and holographic data for four typical defects were acquired experimentally. A two-level fusion assessment model combining dempster-shafer (D-S) evidence theory and artificial neural networks was constructed: First, D-S evidence theory was employed to address uncertainty and conflict among multisource information, after which the high-credibility fused features were input into an artificial neural network for defect severity classification. Laboratory validation demonstrated that the overall accuracy of the proposed model reaches 93.5%, significantly outperforming single-feature models.