Series Arc Fault Identification Based on a Novel Newton-Optimized Sparse Kernel Support Vector Machine

Existing machine learning-based methods for series arc fault (SAF) identification still suffer from slow training speed when dealing with large-scale SAF datasets. For this reason, we propose an SAF identification model based on a Newton-optimized sparse kernel support vector machine (NSKSVM). First, an SAF experimental platform is built, on which a large amount of single-circle SAF current signals are collected. The frequency-domain features of SAF current signals are analyzed and used as the feature set. Second, NSKSVM is constructed to identify SAF quickly for large-scale SAF datasets and improve the accuracy by the kernel function while ensuring the fast training speed of the model. Finally, a series of experiments is conducted to verify the advantages of NSKSVM in SAF identification, which can meet the needs of real-time detection.