Robust Forecasting-Aided State Estimation of Active Distribution Network With Multiple Distributed Generators

This paper develops a robust variational Bayesian unscented Kalman filter (RVBUKF) to track the state of active distribution networks with distributed energy resources. First, a spherical simplex-based unscented transform (UT) strategy is used to select Sigma points to reduce the computational burden. Subsequently, based on unscented Kalman filter and variational Bayesian theory, a variational update form of the covariance matrix is derived by using the inverse Wishart probability density function. To mitigate the impact of outliers, the Gaussian process regression model trained by historical data is used to synchronously predict observation values to replace abnormal measurements. Finally, simulation experiments are conducted on the distribution system containing distributed generations (DGs). The numerical results demonstrate that the developed method can accurately track state changes in uncertain scenarios while reducing the computational burden of UT by almost half. Note to Practitioners—In practical power systems, due to the randomness and fluctuation of distributed generation output, operating states of the active distribution network may frequently change, which places new demands on the dynamic state estimation of active distribution networks. Therefore, in this paper, a new forecasting-aided state estimation method that combines robust Kalman filtering with machine learning techniques based on Gaussian process regression is proposed. The effectiveness of the proposed forecasting-aided state estimation was verified in the constructed active distribution network. Preliminary simulation experiments have shown that the number of Sigma points selected by SSUT is almost 50% of that of traditional UT, and the proposed method is capable of accurately tracking the dynamic changes of active distribution network states in various uncertain scenarios. In future research, we will address the problem of active distribution network forecasting-aided state estimation based on non-parametric modeling and attempt to use actual power grid data for testing.