Toward dynamic stability assessment of power grid topologies using graph neural networks

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Increasing renewable energy integration challenges power grid dynamic stability due to decentralization, reduced inertia, and production volatility.
• Traditional dynamic stability simulations for large power grids are computationally intractable and prohibitively expensive.
• Graph neural networks (GNNs) offer a promising approach to reduce computational costs in power grid analysis.

Purpose of the Study:

• To develop and evaluate GNN models for predicting power grid dynamic stability.
• To create and release large, open-source datasets of synthetic power grid dynamic stability for GNN research.
• To assess the effectiveness of GNNs in identifying critical vulnerable nodes ('troublemakers') within power grids.

Main Methods:

• Generation of large, synthetic power grid datasets for dynamic stability analysis.
• Application of graph neural networks (GNNs) utilizing only topological information for stability prediction.
• Training GNN models on smaller grids and testing their generalization on a large synthetic Texan power grid model.

Main Results:

• GNNs demonstrate surprising effectiveness in predicting highly non-linear dynamic stability targets from network topology alone.
• Achieved practical-use-case performance in dynamic stability prediction for the first time.
• Successfully identified vulnerable nodes ('troublemakers') in power grids with high accuracy.
• GNNs trained on small grids exhibit accurate prediction capabilities on large-scale synthetic power grids, indicating strong generalization.

Conclusions:

• GNNs provide a computationally efficient and accurate method for assessing power grid dynamic stability.
• The developed GNN models and datasets facilitate research and development in grid stability analysis.
• The findings highlight the potential of GNNs for real-world applications in power system management and stability assessment.