Estimating system inertia in sparsely measured dynamic power systems using graph neural networks

The transition to renewable energy sources poses significant challenges for grid stability due to reduced system inertia traditionally provided by synchronous generators. Accurate inertia estimation is critical for ensuring the stability and reli ability of modern power grids. This study explores the potential of Topology Adaptive Graph Convolutional Networks (TAGCN) for inertia and damping estimation in sparsely measured electrical grids. By leveraging graph-structured data, TAGCN models exhibit strong adaptability to the dynamic nature of power systems. Using simulations based on the IEEE 39-bus system, the TAGCN model demonstrated superior performance compared to other conventional neuronal network architectures, particularly in scenarios with limited measurement density. However, key challenges remain, including the need for more diverse datasets, real-world grid testing, and exploration of alternative architectures, hyperparameter tuning and operational conditions.