A Unified Digital Twin Architecture for Integrated Power Grid and Infrastructure Management

INTRODUCTION: The increasing complexity of modern power grids, driven by the integration of distributed energy resources and dynamic operating conditions, presents significant challenges for stability assessment. Traditional stability analysis methods often fail to capture topological dependencies and nonlinear interactions among grid components, resulting in unreliable predictions. Furthermore, existing approaches such as static models and graph convolution networks lack effective node-level importance weighting, limiting their ability to distinguish between stable and unstable states. OBJECTIVES: This study aims to develop an advanced framework for power grid stability classification by integrating digital twin technology with Graph Attention Networks (GAT). The objective is to improve the modeling of inter-node relationships and enhance classification accuracy under complex grid conditions. METHODS: A digital twin-inspired graph model of the power grid is constructed, where nodes represent grid components and edges represent their interactions. A Graph Attention Network is employed to learn weighted inter-node dependencies using attention mechanisms, enabling effective differentiation between stable and unstable operating modes. The proposed framework is evaluated in an offline, simulation-based environment using the Smart Grid Stability dataset. RESULTS: Experimental results demonstrate the effectiveness of the proposed approach, achieving an accuracy of 0.9640, precision of 0.9411, recall of 0.9607, F1-score of 0.9508, and ROC-AUC of 0.9958. Comparative analysis indicates that the proposed model outperforms conventional methods, including Artificial Neural Networks (ANN), Deep Neural Networks (DNN), and Random Forest, in overall classification performance. CONCLUSION: The proposed digital twin-inspired GAT framework provides accurate and reliable offline stability classification, significantly improving upon existing methods. However, challenges related to scalability for larger grid systems and real-time cyber–physical synchronization remain, highlighting important directions for future research.