World Journal of Engineering Research and Technology

Isaiah U. Imoh*, Nseobong I. Okpura, Kufre M. Udofia

ABSTRACT

The increasing penetration of distributed generation and the dynamic, nonlinear nature of modern power systems have rendered traditional model-based voltage stability assessment methods inadequate, particularly because they cannot capture both feature-level nonlinearities and topological dependencies. This paper proposes a novel hybrid framework that synergistically combines Extreme Gradient Boosting (XGBoost) and Graph Neural Networks (GNNs) for accurate and interpretable voltage stability prediction. While XGBoost excels at modeling complex relationships among scalar system features such as voltage stability index (VSI), reactive power margin, and frequency deviations, GNNs capture the spatial and topological interdependencies of the power network by representing buses as nodes and transmission lines as edges. The hybrid model fuses the complementary representations via a weighted concatenation mechanism, followed by a logistic classifier to predict binary stability states (stable/unstable). Extensive experiments on the IEEE 14-bus system, using both balanced and unbalanced datasets, demonstrate that the proposed hybrid model significantly outperforms standalone XGBoost and GNN models. Under unbalanced conditions, the hybrid model achieves 0.87 in accuracy, 0.72 in recall for instability detection, and 0.72 in AUROC. Physically, it reduces VSI error to 0.25 and Q-margin deviation to 4.90, indicating superior alignment with system dynamics. These results establish the hybrid XGBoost–GNN architecture as a robust, topology-aware solution for real-time voltage stability monitoring.

[Full Text Article] [Download Certificate] https://doi.org/10.5281/zenodo.21718557