Transient Assessment of Power Systems Based on Graph Attention Networks

The increasing proportion of new energy equipment makes the spatio-temporal coupling of power system extremely complex, and the traditional transient stability assessment (TSA) means can no longer meet the requirements of rapidity and accuracy. To solve this key problem, a transient stability assessment method based on graph attention network (GAT) is proposed. We utilize the multi-head attention mechanism to aggregate the grid topology and node feature information, use the IFL function to reduce the influence of unbalanced sample data on the results, and finally build the offline training and online evaluation model of TSA based on GAT algorithm, and fully verify the effectiveness of the proposed model by comparing and analyzing it with other deep learning algorithms.