A hybrid approach for partial discharge classification: combining traditional machine learning and deep neural network

Abstract

Partial discharge (PD) is a critical issue in high-voltage equipment, and the accurate detection and classification of PDs are essential for preventing equipment failure. In recent years, various approaches have been proposed for PD classification, including traditional machine learning methods and deep learning techniques. Traditional machine learning algorithms, such as decision trees, support vector machines (SVM), and k-nearest neighbors (KNN), have been widely used for PD classification. However, these methods rely on manual feature extraction, which can be time-consuming and may not capture the complete range of PD characteristics. In contrast, deep learning techniques, including CNN and RNN, have shown promising results in PD classification by enabling the automatic extraction of relevant features from PD data. However, it requires a large amount of training data. This study proposes a novel approach for PD classification, combining traditional machine learning algorithms with deep neural networks to perform transfer learning. Firstly, manual feature extraction is conducted to extract PD features. Traditional machine learning clustering algorithms, such as K-means and affinity propagation clustering will be applied to these features to separate noises from PDs. Subsequently, the Partial Discharge Pattern Recognition and Diagnosis (PRPD) is plotted and fed into a CNN to classify each cluster. In order to apply it in real-life applications, minimizing the missing detection rate is considered the priority of the tunning process. The proposed method can effectively detect and classify PD which can aid in the development of effective PD diagnosis systems and contribute to the safe and reliable operation of high-voltage equipment.