A heterogeneous multi-model stacking framework for predicting pump complete characteristic curves

Pumped hydro energy storage (PHES) is the most mature large-scale energy storage technology and plays a vital role in addressing the intermittent and variable nature of renewable energy. This study aims to address the high cost and low efficiency of traditional methods for obtaining pump complete characteristic curves, which are crucial for the safe and stable operation of PHES. A stacking ensemble model (IWOA-Stacking-Ridge) based on an improved whale optimization algorithm is proposed to predict the pump complete characteristic curves. This method uses specific speed and third-quadrant pump characteristic curve data as input to predict dimensionless head (WH) and torque (WB) for all operating conditions. This study pioneered the use of stacking ensemble learning in the field of pump complete characteristic curves prediction, constructing three heterogeneous base model groups. The improved whale optimization algorithm (IWOA) was introduced to intelligently select the optimal base model combination from 12 machine learning algorithms and collaboratively optimize hyperparameters. Verification showed that the IWOA-Stacking-Ridge model achieved prediction accuracies exceeding 0.949 for all three pump types. Compared to the optimal single model, the integrated model achieved a 4.2 % improvement in R2, a 22.5 % reduction in RMSE, and a 27.8 % improvement in MAE. The proposed method provides efficient and reliable technical support for the preliminary design of PHES, stability analysis of S-shaped characteristic regions, and the development of load shedding and mode transition control strategies. This approach significantly benefits the deployment and operation of pumped storage systems.