Deep Learning for Partial Discharge Classification: A CNN-Based Approach

Classification and detection of Partial discharge (PD) play a crucial role in high-voltage equipment condition monitoring. This study presents an approach using Convolutional Neural Networks (CNNs) for automated PD classification. The proposed method involves preprocessing raw PD signals, designing an optimized CNN architecture, and training the model on a comprehensive dataset comprising corona, surface, and internal PD signals. The CNN model achieved 97.2% accuracy, outperforming traditional methods of machine learning. Detailed performance metrics, such as precision, F1-scores, and recall, demonstrate the effectiveness of this model across multiple PD types. Analysis of misclassified signals revealed challenges in distinguishing between surface and corona discharges, providing insights for future improvements. The study discusses the implications of these findings for real-world applications in condition monitoring and outlines potential areas for future research, including data augmentation techniques, hybrid models, and real-time implementation. This work contributes to the advancement of AI-driven techniques in high-voltage engineering and enhances the efficiency and reliability of electrical power systems.