Identification of model parameter and measurement error in large-scale power grid base on graph attention network

The analysis of deviation between measured value and estimated value in power system state estimation mainly relies on manual experience, and there is a lack of an intelligent auxiliary tools for model parameter and measurement error identification. This article proposes an intelligent method to diagnose the problem of the model parameter and measurement near the deviation point of state estimation (the point where the measured value deviates greatly from the estimated value) base on graph attention network (GAT). The subgraph is built according to the local network topology near the deviation point. We divide graph nodes into four categories: normal nodes, topology error nodes, measurement error nodes, and parameter error nodes, and establish a node classification network for diagnosis of model parameter and measurement error. The GAT network is trained by the method of supervised learning based on the power system model maintenance experience and automatic error generation. With proposed automatic sample generation technology, we can generate millions of training samples. We test the GAT network on China state grid model with more than 60000 buses and 70000 branches and the result has sufficient accuracy.