Dynamic State Estimation and Anomaly Detection in Smart Grid Using Point-Based Gaussian Approximation Filtering

We consider the problem of dynamic state estimation and anomaly detection in smart grid, which is a typical cyber-physical system that is described by a nonlinear model. The state estimation problem is solved by the point-based Gaussian approximation filter, which incorporates different quadrature rules to compute the posteriors. This filtering method is compared with its traditional counterpart - the extended Kalman filter - and much higher tracking accuracy is achieved as expected. The point-based Gaussian approximation filter is then combined with the widely-used anomaly processing method to detect the bad measurement and sudden load change in smart grid. The innovation vector, which is used in the update step of filtering, is first examined for the presence of anomalies, and then processed to perform skewness-test for anomaly discrimination.