Runoff Forecast Model Based on an EEMD-ANN and Meteorological Factors Using a Multicore Parallel Algorithm

• Published in: Water resources management Vol. 37; no. 4; pp. 1539 - 1555
• Main Authors: Liao, Shengli, Wang, Huan, Liu, Benxi, Ma, Xiangyu, Zhou, Binbin, Su, Huaying
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.03.2023; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial neural networks; Atmospheric Sciences; Civil Engineering; Correlation coefficient; Correlation coefficients; Coupled modes; Earth and Environmental Science; Earth Sciences; Environment; Forecasting; Geotechnical Engineering & Applied Earth Sciences; Hydroelectric power; Hydroelectric power stations; Hydrogeology; Hydrology/Water Resources; Inflow; Mathematical analysis; Mathematical models; Neural networks; Optimization; Performance prediction; prediction; Prediction models; Runoff; Runoff forecasting; water power; Water resources; ANN; Parallel algorithm; Ensemble empirical mode decomposition; Long-term runoff forecast; Filtered meteorological factors
• ISSN: 0920-4741; 1573-1650
• DOI: 10.1007/s11269-023-03442-y

Accurate long-term runoff forecasting is crucial for managing and allocating water resources. Due to the complexity and variability of natural runoff, the most difficult problems currently faced by long-term runoff forecasting are the difficulty of model construction, poor prediction accuracy, and time intensive forecasting processes. Therefore, this study proposes a hybrid long-term runoff forecasting framework that uses the antecedent inflow and specific meteorological factors as the inputs, is modeled by ensemble empirical mode decomposition (EEMD) coupled with an artificial neural network (ANN), and computed by a parallel algorithm. First, the framework can transform monthly inflow and meteorological series into stationary signals via EEMD to more comprehensively explore the relationships of the input factors through the ANN. Second, the selected meteorological factors that are closely related to inflow formation can be filtered out by the single correlation coefficient method, which contributes to reducing coupling between input factors, and increases the accuracy of the prediction models. Finally, a multicore parallel algorithm that is easily accessed everywhere and that fully utilizes multiple calculation resources while flexibly contending with various optimization requirements will improve forecasting efficiency. The Xiaowan Hydropower Station (XW) is selected as the study area, and the final results of the study show that (1) the addition of targeted meteorological factors does indeed greatly enhance the performance of the prediction models; (2) the five criteria for evaluating the prediction accuracy show that the EEMD-ANN model is far superior to the prediction performance from the ordinary ANN model when run under the same input conditions; and (3) the optimization time of the 32-core model can be reduced by as much as 25 times, which significantly saves time during the forecast process.