GridPACK Application Concept (State Prediction)
State Prediction
Title: A Statistical State Prediction Methodology to Improve Reliability and Efficiency of Power System Operation
Team: N Zhou, DJ Haglin, FK Tuffner, Y Chen, TA Ferryman, G Lin, J Yin, M Vlachopoulou
This project is motivated by the challenges of increasing uncertainty and variation brought in by the high penetration of renewable generation to power grid operation. A state estimator is an essential tool for power grid operation. Due to the delays from communications and computations, current state estimators can only provide power grid status in the past. This delay has typically been on the order of 2 to 5 minutes, and recent developments may reduce this delay to 30 to 45 seconds. Even with the relatively fast update on the state estimates, the power grid has to be operated based on its past states. Scheduling interchanges and dispatching generation is currently handled through forecasting of the system demand. This kind of practice is acceptable when a power grid does not change very much or does not deviate from the forecasting model significantly. However, with high level penetration of renewable generation (e.g., wind and solar), the North American power grid is going to experience significant levels of variation and uncertainty in power flow. With quick changes and large uncertainty brought in by renewable generation, operations based on the past deterministic states can lower the reliability and efficiency of power grid operation.
The study will result in a power system state predictor, which cannot only provide prediction of power system states, but also quantify prediction errors (or uncertainty) on those estimates. The prediction method can generate power system state estimates for the current and future time. Combined with past states from a traditional state estimator, the state predictor can provide a whole picture of the power system, in the past, current and future with uncertainty quantification. It is expected that the whole picture can improve operator’s situational awareness of the power grid and risks, enable the proactive operation of the power grid, and thus help improve operational efficiency and reliability.
The technologies that have been proposed in the same power grid initiative can be leveraged to bring out more benefits from the proposed state prediction. Power system modeling efforts (proposed by Shuai Lu) can increase the modeling accuracy, which, in turn, helps increase accuracy of the simulation and prediction in this project. To further shorten the computational time, the state prediction can be implemented on a high performance computer platform. The state estimation procedure can be solved effectively by leveraging the study of “linear algebra solver” for the high performance computation (proposed by Barry Lee). With only partial data needed through smart sampling study, the data can be processed locally and only condensed information needs to be passed. Thus the distribution computation technology (proposed by Jenny Liu) can be leveraged to further reduce the communication and computation time. The scalable middleware study (proposed by Jian Yin) can further reduce the communication delay by providing a unified scalable data management service. To guarantee that the computation be finished within the prediction lead time, the state prediction can be implemented on the real time operating system for high performance computers (proposed by Peter Hui).
