| Summary: | Widely confirmed and applied, data-driven models are an important method for watershed runoff predictions. Since decomposition methods such as time series decomposition cannot automatically handle the decomposition process of date changes and less consideration of influencing factors before decomposition, resulting in insufficient correlation analysis between influencing factors and forecast objects, we propose a method based on hydrological model decomposition to generate time series state variables (broadening the range of influencing factors to be considered). In this study, we constructed hydrological models wherein rainfall and other hydrological elements are decomposed into hydrological and hydrodynamic characteristic state variables to expand the range of the prediction factors. A data-driven model was then built to perform runoff predictions in the Han River Basin. The results showed that compared with the single prediction model, the prediction results based on the coupling model were superior, the performance evaluation grade of the coupling model was high, and the coupling model had a higher stability.
|
|---|