Noncontact Fault Classification in Three-Core Cables Using Random Forest With Magnetic Field Analysis

Distribution networks are vulnerable to various short-circuit (SC) faults caused by lightning strikes, storms, vegetation growth, animal interference, insulation breakdown, and other environmental factors. Prompt diagnosis of these faults is crucial to prevent power outages. This article presents a novel method for identifying SC fault types in three-core cables using noncontact magnetic field (MF) measurements. The proposed method employs highly sensitive magnetoelectric (ME) sensors with strong anti-interference capabilities to detect MF variations on the surface of three-core cables. Utilizing the Concordia transform, we generate the alpha- and beta -mode MFs. By observing MF waveform signals and Magnetic Concordia patterns, fault features are extracted based on both signal power and cosine similarity (CS) to identify different types of SC faults. Considering the combination of parameters such as fault distance, fault resistance, and fault inception angle, all SC fault types are assumed to have the same number of samples. SC fault classification is performed using the random forest (RF) algorithm. Simulation and experimental results validate the method's applicability. Furthermore, classification outcomes are compared with those obtained using support vector machines (SVMs) and the C4.5 decision tree (DT) algorithm, highlighting the efficacy of the proposed approach.