State Estimation for Reconfigurable Distribution Systems Using Graph Attention Network

Obtaining the real-time state of the distribution system is the basis for intelligent operation of the power system. With the surge in new energy generation and the volatility in load demands, traditional state estimation methods face significant challenges. The distribution system requires frequent topology reconfigurations to maintain stable operation. However, current data-driven methods typically cater only to specific topologies. To address this issue, a complete distribution system state estimation (DSSE) framework is proposed to adapt to frequent topology reconfigurations. To ensure data quality, a measurement device configuration algorithm using node importance was designed. Then, a convolutional neural network-based topology identification model is utilized to provide real-time topology data to the DSSE, which uses measured data pre-processed by the Gramian corner field. Finally, we use graph attention network to model the DSSE as a node-level regression prediction problem on a graph abstracted from the distribution system. Simulation results on IEEE 33-bus and IEEE 118-bus distribution systems illustrate the feasibility and efficiency of the proposed framework. Further experiments show that the proposed framework has good robustness.