A Topology-Aware Power Grid State Estimation Approach via Graph Attention Networks-Based Representation Learning

State estimation plays a crucial role in ensuring the safety, stability and operational efficiency of the current power grid. However, with the increasing size and complexity of power grids, traditional power grid state estimation methods face significant challenges in reducing estimation errors and improving model accuracy due to that they neglect the correlation and varying degrees of influence among nodes in the power grid. To address this issue, this paper proposes a topology-aware power grid state estimation method (TPGSE), which combines a multihead self-attention mechanism with message passing graph neural networks. The topological structure of the power grid is naturally considered as a graph model. An anomaly detection mechanism is utilized to identify and handle abnormal data, followed by the imputation of missing node and edge features. Then, a message passing graph neural network with a multi-head self-attention mechanism is employed to aggregate the features of nodes and edges, capturing the latent relationships between them. Finally, a shared-parameter fully connected layer is used to independently handle the prediction tasks for each node. Experimental results demonstrate that the proposed method achieves accurate state estimation and prediction, exhibiting strong generalization and robustness across complex power grid topologies.