A Data-driven Approach to Power System Dynamic State Estimation
MetadataShow full item record
State estimation is a key function in the supervisory control and planning of an electric power grid. Typically, the independent system operator (ISO) runs least-squares based static state estimation once every few minutes. Inherently, however, a power system is mostly in a transient state owing to load fluctuations, outages and network switching. In such a scenario, dynamic state estimation facilitates real-time monitoring and control of the system. Dynamic state estimation is implemented using Kalman filtering techniques. Popular estimators for nonlinear systems include the extended Kalman filter (EKF) and unscented Kalman filter (UKF). Practical implementation, however, is inhibited by the lack of an accurate system model and the high computational complexity of Kalman filtering methods. I address the former issue of model unavailability and rely instead on measurement data from phasor measurement units for dynamic state estimation (DSE). I build an estimator for DSE which uses only measurement and input information, and operates without knowledge of the underlying system model. The algorithm considered uses a Gaussian process (GP) approximation of the state transition and observation functions in the implementation of a UKF-based state estimation. I analyze the performance of the estimator for different scenarios using root mean squared (RMS) error as the metric. The estimator, when evaluated on the IEEE 14-bus test case, gives a minimum accuracy rate of over 94% over all considered scenarios.
Kumari, Deepika (2017). A Data-driven Approach to Power System Dynamic State Estimation. Master's thesis, Texas A & M University. Available electronically from