Machine Learning-Based Islanding Detection in Grid-Connected Photovoltaic Systems

Doreen Edwina Owoo

Keywords: Islanding detection, Random Forest, adaptive thresholding, grid-connected PV, non-detection zone, machine learning

TL;DR: This paper investigates a novel method using machine learning algorithms to detect when a photo-voltaic system connected to a grid has isolated itself.

Abstract: The growing integration of grid-connected photovoltaic (PV) systems into distribution networks has made reliable islanding detection a critical safety requirement. Islanding occurs when a PV-connected segment remains energised after utility disconnection, endangering maintenance personnel and risking equipment damage. Traditional passive and active detection methods are constrained by Non-Detection Zones (NDZ) and power quality degradation. This study develops and evaluates a suite of machine learning classifiers, including Random Forest, Support Vector Machine (SVM), Multi-layer Perceptron Neural Network (M), and XGBoost, trained on features extracted from MATLAB/PSIM simulations of a grid-connected PV system. Features were extracted from three-phase voltage signals at the Point of Common Coupling(PCC). Random Forest with an adaptive decision threshold of 0.45 achieved the best overall performance, attaining 98.35% accuracy, 97.23% precision, 94.75% recall, and an F1-score of 95.97%, with a near-zero NDZ and no active signal injection, satisfying the IEEE 1547 two-second detection requirement.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 14