Real-time online prediction of hydraulic states in pumped hydropower systems based on Kalman filter

With the growing penetration of intermittent renewable energy in power grids, pumped storage plants (PSPs), as the most widely adopted energy storage technology, are facing increasingly demanding load regulation tasks. Frequent switches between different operations have made the hydraulic states more unpredictable. Early warning of extreme hydraulic states (e.g., pressure) has become critical issue for ensuring the safe operation of PSP systems. However, existing approaches passively compare field measurements with offline simulation results, lacking the capability for predictive control. To address this gap, this paper proposes a real-time online hydraulic state prediction method based on Kalman filter theory. First, a physics-based PSP model is established using graph theory. This is then followed by the development of a real-time data assimilation and prediction framework with a sliding window. This framework enables real-time model updates by incorporating newly available data from the most recent period, as well as the prediction of future states of the event before it occurs. Finally, the proposed method is validated using field measurements. The results in the validation case indicate that the new method improves the accuracy from -3.47% in the offline simulation to 0.17% in predicting the peak spiral case pressure (SCP) with a neglectable error in its occurring time. The confidence intervals of the predicted values effectively encompass the fluctuation range of the measured SCP, ensuring reliable prediction and thus early warning of extreme conditions before they occur. The proposed method has great potential for implementing predictive control strategies before extreme conditions occur. • A fourth-order equivalent electrical circuit model for pumped storage plants is developed using graph theory. • A data assimilation framework that integrates Kalman filter theory with real-time measurements is proposed. • A sliding-window prediction strategy with confidence intervals is introduced. • The proposed method is validated using field data.