A Novel State Estimation Method Based on Cross-Track Graph Neural Network

This paper proposes a novel graph neural network GCrossNet (Graph-based Cross Network) for fast and reliable power system state estimation considering topology perturbations, load fluctuations, and measurement errors. In order to capture the underlying load flow relations, a cross-connection graph neural network architecture is proposed allowing loose decoupling between less correlated states while maintaining information exchange between the branch and bus tracks. It is expected to yield greater returns than traditional networks due to its superior ability in extracting features among different system states and topologies in graph representations. The end-to-end estimation framework has been tested on the IEEE 39-bus system and the ACTIVSg2000 system, reducing state estimation errors by over 96% compared to other algorithms while also improving robustness and generalization.