一种基于物理与数据辅助的抽水蓄能电站瞬态过程预测框架

Abstract Achieving accurate predictions of transient processes for pumped-storage hydropower stations (PSHSs) remains a key challenge due to uncertainties in on-site parameters, particularly the pump-turbine characteristic curves (PTCCs), and limitations of the physics-based models themselves. To address this issue, this study proposes a transient prediction framework for PSHSs, centered on on-site measurements and incorporating both the physics-based model calibration and the data-aided correction. A method for reconstructing PTCCs using point distribution models (PDMs) is proposed, where PDMs act as prior models and are innovatively developed by defining multiple feature points on PTCCs to accommodate potential non-rigid deformations. This approach allows the reconstruction of complete PTCCs using a surface reconstruction algorithm, requiring only limited measured data from steady-state and transient experiments. To further compensate for errors in the physics-based model, a data-aided correction using nonlinear autoregressive with exogenous inputs (NARX) is proposed. The NARX model is optimally tuned by selecting the most sensitive model inputs which have the highest correlations with the predicted error of the physics-based model. Compared with the conventional model, the proposed framework reduces the predicted tendency errors for discharge, pressure at the volute , pressure at the draft tube, and rotational speed by average values of 10.82 %, 13.88 %, 36.67 %, and 7.37 %, respectively, across all experimental cases. The proposed transient prediction framework enables highly accurate predictions for a diverse range of transient processes of PSHSs and serves as a pre-warning basis for real-time monitoring systems , facilitating the sustainable operation of PSHSs.