Dynamic topology and line parameter identification of power systems based on graph embedding learning

The grid integration of high-penetration distributed generation (DG) and the widespread deployment of smart meters have increased the complexity of power system operation and maintenance. Traditional topology and parameter identification algorithms suffer from limitations such as insufficient accuracy and poor robustness. A parameter-dynamic topology joint identification method based on graph embedding is developed. The method is implemented in three sequential stages, namely encoding, decoding, and optimization. The adjacency matrix of the power grid is reconstructed by analyzing structural characteristics and comprehensively considering first-and second-order similarity information. Node features are further refined through sampling and aggregation. Numerical simulations under different data anomaly scenarios are conducted on the IEEE 118-bus system. The results indicate that the proposed method exhibits favorable robustness against noise, data loss, and DG uncertainty. Compared with traditional approaches, the method can effectively capture deep structural correlations within the power grid, thereby significantly improving the accuracy of line parameter identification and reducing the root mean square error (RMSE) by approximately 20%~30%.