Combining CNNs and One-Class SVMs for Effective Anomaly Detection in Switchgear Faults via Phase-Resolved Partial Discharge Analysis

Early detection of faults in switchgear is crucial for preventing costly disruptions and to maintain the safety and reliability of electrical power grids. The patterns of Phase-resolved partial discharge (PRPD) provide a valuable tool for diagnosing the health of electrical systems, but their complex nature poses challenges for accurate fault detection.This study proposes a novel method that leverages the power of pre-trained deep learning models including VGG19, GoogLeNet, AlexNet, ShuffleNet, ResNet50, Xception, InceptionResNetV2, DenseNet201, SqueezeNet, InceptionV3, alongside a one-class support vector machine (SVM) to effectively identify anomalies in switchgear components.By combining the strengths of these advanced techniques, this approach overcomes the limitations inherent in traditional methods. The results demonstrate impressive performance, achieving a high accuracy of 94.74% in validation and 88.72% in testing on real-world PRPD data. the validation prediction times ranged from 9 msec to 204 msec, and the testing prediction times varied from 9 msec to 206 msec, indicating the efficiency of these models for practical applications. This innovative solution provides a valuable tool for proactive maintenance, ensuring the reliability and safety of critical electrical infrastructure in industrial applications.