Reservoir water level prediction using combined CEEMDAN-FE and RUN-SVM-RBFNN machine learning algorithms

Accurate prediction of water level changes in reservoirs is crucial for optimizing the operation of reservoir projects and ensuring their safety. This study proposed a method for reservoir water level prediction based on CEEMDAN-FE and RUN-SVM-RBFNN algorithms. By integrating the adaptive complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method and fuzzy entropy (FE) with the new and highly efficient Runge–Kuta optimizer (RUN), adaptive parameter optimization for the support vector machine (SVM) and radial basis function neural network (RBFNN) algorithms was achieved. Regression prediction was conducted on the two reconstructed sequences using SVM and RBFNN according to their respective features. This approach improved the accuracy and stability of predictions. In terms of accuracy, the combined model outperformed single models, with the determination coefficient, root mean square error, and mean absolute error values of 0.997 5, 0.241 8 m, and 0.161 6 m, respectively. In terms of stability, the model predicted more consistently in training and testing periods, with stable overall prediction accuracy and a better adaptive ability to complex datasets. The case study demonstrated that the combined prediction model effectively addressed the environmental factors affecting reservoir water levels, leveraged the strength of each predictive method, compensated for their limitations, and clarified the impacts of environmental factors on reservoir water levels.