Fault diagnosis of surge arresters via electro-thermal coupled simulation and gradient boosting tree.

The Metal-Oxide Arrester (MOA) is a crucial device for ensuring the safety of power grids, but its fault diagnosis faces the challenge of scarce actual fault samples. In this study, a 3D electro-thermal coupled simulation model is developed to reconstruct four operating conditions: internal moisture, aging of the insulating sleeve, wet contamination, and normal operation. Specifically, fault mechanisms are physically modeled by altering the conductivity of the insulating sleeve for aging, mapping conductive layers with varying coverage on the porcelain housing for wet contamination, and attaching geometric water bands to the valve discs for internal moisture. Based on this model, current and temperature signals are obtained to analyze the electro-thermal characteristics under different faults. Subsequently, a CatBoost-based fault diagnosis framework is introduced to construct a fault diagnosis model. By preprocessing data and integrating features such as temperature statistics and geometric quantities, the model's recognition performance is significantly improved. Comparative experiments against XGBoost, fitcnet, and Support Vector Machine (SVM) demonstrate the superiority of the proposed method. The final model achieves an accuracy of 99.3%, outperforming XGBoost (98.6%), fitcnet (96.5%), and SVM (90.0%). This study provides an effective solution for MOA fault diagnosis under sample-scarce conditions, enhancing the intelligent assessment and maintenance of power grids.