State Estimation in Smart Grids Using Temporal Graph Convolution Networks

State estimation (SE) is one of the key functions of smart grids. The availability of large volumes of measurement data introduces new opportunities for improving and complementing the conventional model-based SE in power systems. In this work, a data-driven approach based on Graph Convolution Neural Networks (G-CNNs) is presented for SE in smart grids. The G-CNN can learn the features in the non-Euclidean domain of graphs, which can capture the structures and interactions among the components of power grids. By integrating the temporal dependencies in the time-series data, a temporal G-CNN (T-GCN) is adopted for the SE problem. Specifically, a message-passing G-CNN is used to capture the topological structure of the smart grid and the gated recurrent units are used to capture the dynamic variation of state information for temporal dependencies. The performance evaluation of the presented method for two cases of full measurement availability and availability of a subset of measurements in comparison with some of the existing SE techniques shows promising performance.