Partial Discharge Identification of Distribution Cable Based on Multi-Feature Machine Learning

Distribution cable is one of the most important equipment in the process of power transmission. In the manufacturing process of distribution cable, it is inevitable that there will be defects in production and manufacturing, which will further lead to the generation of partial discharge (PD) of distribution cable. It is of great significance to monitor and identify the types of partial discharges to ensure the running status of distribution cables and improve the service life of cables. In this paper, a fault identification algorithm of PD UHF signals based on multi-information fusion is proposed. Firstly, four common types of PD defect models are simulated. The time-domain fault signals and PRPD pattern of PD are extracted by UHF sensors, and the typical features of PD fault signals are further obtained. The traditional machine learning algorithms support vector machine (SVM) and Gradient Boosted Decision Tree (GBDT) are used to statistically learn the signal features, and the deep residual network is used to identify the PRPD pattern image. A multi-model weighted fusion algorithm is used to identify PD defect types. The proposed method has a certain generalization ability and makes full use of the information contained in the discharge pulse and PRPD image to realize the task of insulation fault diagnosis.