Using Deep Transfer Learning Technique to Protect Electrical Distribution Systems Against High-Impedance Faults

The dependence of high-impedance faults (HIFs) detection methods on a large amount of training data has always been a fundamental problem in electrical distribution systems. This article proposes a novel protection system based on the transfer learning technique and GoogleNet architecture to reduce this dependence. The proposed protection system uses a small amount of data to extend the knowledge of pretrained GoogleNet architecture to the HIF detection problem. In this system, a small amount of third harmonic angle data of the current at the measurement point are obtained from the understudy electrical distribution system. Then, the preprocessing phase is performed, and the extracted data are converted to image data using the Wigner-Ville distribution. Afterward, these converted images are fed to the GoogleNet architecture as an input dataset to update the GoogleNet pretrained knowledge. Finally, the process of fault detection and classification is accomplished only by transferring the GoogleNet pretrained knowledge. The simulation results of the modified IEEE 13-bus and 34-bus distribution systems in EMTP-RV and MATLAB indicate the high accuracy of the proposed protection system despite the use of a small amount of input training data.