Performance analysis of intelligent classifiers for high impedance fault detection in a PV-integrated IEEE-13 bus system

High impedance faults (HIFs) present significant challenges in power systems, particularly when an electrical wire contacts a high-resistance material, leading to low currents that are difficult for traditional relays to detect. With the increasing integration of photovoltaic (PV) systems, these challenges are exacerbated due to the complex behavior of PV-generated signals. This study aims to enhance the detection of HIFs in PV-integrated systems using advanced machine learning techniques. The approach employs various classifiers including artificial neural networks, support vector machines, decision trees, and Random Forest to improve fault identification accuracy. A MATLAB/SIMULINK simulation was conducted on an IEEE 13-bus system with a 300 kW solar PV plant. The Discrete Wavelet Transform (DWT) with the db4 wavelet was used for feature extraction, focusing on phase energy values. The classifiers were evaluated under different scenarios such as normal operation, load switching, capacitor switching, HIF, and line-to-ground (LG) faults. The Random Forest classifier outperformed others, achieving a fault detection accuracy of 99.4083%, demonstrating its robustness in adapting to various fault conditions. The Naive Bayes, Multilayer Perceptron, and Logistic Regression classifiers achieved lower accuracies of 78.6982%, 76.9231%, and 80.4734% respectively. These results indicate a significant improvement in fault detection capability, enhancing the stability, reliability, and resilience of electrical grids integrated with PV systems. The findings suggest that the Random Forest classifier is highly effective for HIF detection, which is crucial for the protection and efficient operation of modern power grids with high renewable energy penetration.

Text 118346.pdf - Published Version Restricted to Repository staff only Download (774kB) Official URL or Download Paper: https://ieeexplore.ieee.org/document/10970376/

Actions (login required)

View Item