Photon-Counting-Based Characterization and Classification of Partial Discharge for HVDC Gas-Insulated Equipment

High-sensitivity detection of direct current (DC) partial discharge (PD) in HVDC gas-insulated equipment (GIE) remains challenging because conventional electrical measurements are susceptible to ambient interference and DC PD lacks a phase reference for phase-resolved analysis. Although photon counting techniques provide exceptional sensitivity and noise immunity, their diagnostic application has so far been confined to alternating current (AC) conditions. In this study, a photon-counting-based measurement platform was developed to investigate DC PD generated by three representative gas–solid insulation defects, namely conductor protrusion, surface-attached metal, and free metallic particle. Photon pulse sequences were acquired under both positive and negative voltage polarities. Successive inter-pulse time intervals were then mapped into two-dimensional kernel density estimation heatmaps to visualize defect-dependent temporal characteristics. A Random Forest classifier, integrated with SHapley Additive exPlanations (SHAP) for feature reduction, was employed for quantitative classification. The proposed method achieved classification accuracies of 97.50% and 99.17% for positive and negative polarities, respectively. Notably, the model adaptively prioritized angular-distribution features over radial-distribution features under space-charge-suppressed conditions. These results demonstrate the feasibility of photon-counting-based time-domain characterization and defect classification for DC PD, providing a quantitative, less experience-dependent framework for insulation defect identification in DC gas-insulated systems.