Artificial intelligence based partial discharge detection using CNN and KNN to increase the quality of electrical insulation

Partial discharge (PD) detection is crucial for ensuring the reliability and longevity of high-voltage insulation systems, as undetected PD can lead to catastrophic failures. This paper presents a novel hybrid detection system integrating Convolutional Neural Networks (CNN) using the ResNet architecture for feature extraction and k-Nearest Neighbors (KNN) for classification. The proposed model achieves an impressive 98.79% accuracy, 99.41% precision, and 98.74% F1-score, significantly outperforming traditional methods such as SVM and CNN-SVM hybrids. The system demonstrates robust noise resistance, maintaining a detection rate above 98%, even in adverse operational conditions. Experimental validation on both synthetic and real-world PD datasets confirms the model’s scalability, computational efficiency, and ability to classify PD signals with minimal false positives and negatives. Additionally, it quantifies critical PD characteristics, achieving a mean true charge of 19.73 pC and a mean apparent charge of 3.15 pC, enabling improved insulation diagnostics. This work establishes a new benchmark in PD detection by addressing key challenges such as noise interference, real-time monitoring, and fault classification accuracy, paving the way for industrial deployment in high-voltage equipment monitoring systems. Future research will be directed toward extending the applicability of the model to other types of electrical insulation faults, enhancing its real-time performance, and integrating it into industrial monitoring systems for comprehensive fault diagnostics. These advancements position the proposed hybrid model as a pivotal tool for improving the safety and reliability of high-voltage insulation systems.