Partial Discharge Fault Diagnosis of GIS Based on the Stacking Ensemble Model

To address the limitations of existing methods for Partial Discharge(PD) fault type diagnosis, which rely on single classification models and struggle to learn representations of multiple feature spaces, an improved PD fault diagnosis method based on a Stacking ensemble of multiple models is proposed. The base learners and meta-learner of the model are constructed from 1D CNN-ResNet based on deep learning models, and XGBoost, Random Forest, and SVM based on machine learning models. Experimental results demonstrate that the Stacking ensemble model, which integrates the advantages of multiple differentiated single models, achieves a fault recognition accuracy of 97.77% on an open-source PD dataset. The results surpass those of single classification models, indicating that this method has high application value in PD fault type diagnosis.