A Machine Learning-Based Fault Identification and Classification Method for UPFC-Compensated Transmission Lines Integrated with Solar PV Systems

Received: 28 February 2026 | Revised: 7 April 2026 and 29 April 2026 | Accepted: 30 April 2026 | Online: 6 June 2026

Corresponding author: Aniruddha Ray

Abstract

Unified Power Flow Controllers (UPFCs) are deployed to enhance transmission network performance, but can disrupt traditional distance protection, especially when integrated with solar power. This paper investigates the impact of a UPFC and a solar Photovoltaic (PV) plant on distance relay operation in a modified 39-bus New England test system, focusing on four fault types: single line-to-ground, line-to-line, double line-to-ground, and three-phase faults. The study finds that Zone-1 protection starts to malfunction under certain fault scenarios, including single line-to-ground and line-to-line faults with 40 Ω resistance, double line-to-ground with 10 Ω, and three-phase faults with 60 Ω. To address this, a Machine Learning (ML) approach is proposed for fault identification and classification. A dataset of 400 fault cases was generated by varying fault resistance, type, and location. The performance of four ML classifiers, Fine k-Nearest Neighbors (kNN), Linear Support Vector Machine (SVM), Bagged Trees, and Fine Decision Tree acting as a digital relay was analyzed using the dataset within the MATLAB Simulink environment. Linear SVM classifier achieved the highest performance, attaining 99.98% accuracy and 100% precision, 99.87% recall, and an F1-score of 0.9993 for the ABC fault; 99.78% precision, 100% recall, and an F1-score of 0.9989 for the C–G fault; and 100% precision, recall, and F1-score for the remaining fault scenarios while exhibiting a superior Receiver Operating Characteristic (ROC) curve. The trained SVM classifier identified the fault types correctly even with a 10 dB random noise.