Graph Neural Network Based Power Grid Topology Identification and State Estimation for the Nigerian High-Voltage Transmission Network

This research develops a graph neural network framework for joint power grid topology identification and state estimation in the Nigerian high-voltage transmission network, constituting original contributions to the application of geometric deep learning to power systems problems in data-limited and topologically uncertain network environments. Accurate knowledge of real-time network topology is a prerequisite for reliable state estimation, but the Nigerian transmission network experiences frequent topology changes from both planned switching operations and unplanned outages that are not always reflected immediately in the TCN Energy Management System topology processor due to communication failures and manual update delays. The theoretical contribution introduces a novel GNN architecture that simultaneously performs topology identification from available voltage and current measurements and produces state estimates consistent with the identified topology, exploiting the inherent graph structure of the power network through message-passing neural network layers with physically motivated aggregation functions derived from the power flow equations. The architecture incorporates an uncertainty quantification mechanism based on deep ensembles that provides calibrated confidence intervals on both topology and state estimates, enabling operators to identify unreliable estimates requiring manual verification. Training data were generated using a validated simulation model of the TCN 330kV network perturbed with realistic topology uncertainty scenarios derived from analysis of TCN SCADA event logs. The model is validated on held-out TCN SCADA records from 2022 to 2023. Topology identification accuracy reaches 97.3 percent across tested scenarios, and state estimation RMSE is reduced by 31 percent compared to the current WLS estimator under topology uncertainty conditions. Keywords: graph neural network, topology identification, state estimation, transmission network, Nigeria.