Enhancing System Inertia Estimation: Multi-Head Graph Attention Networks Leveraging PMU Measurements

Accurate estimation of power system inertia is crucial for ensuring stability and security in electrical grids. This paper introduces a novel approach utilizing a multi-head Graph Attention Network (GAT) for inertia constant estimation. The proposed model's performance outperform that of established models like GCN, CNN, and DNN, exhibiting significant accuracy even under Signal-to-Noise Ratio (SNR) conditions of 45 dB. Our proposed model provides valuable insights into critical graph components influencing inertia estimations, empowering informed decision-making in power system analysis. The integration of superior performance, and adaptability to varying signal conditions positions our proposed GAT model as a promising tool for precise and insightful inertia estimation within power systems, promising significant advancements in this critical domain. Our model achieved 15.63% and 31.7% in MSE compared with GCN with/without considering noise.