查看文章

Accurate and reliable streamflow prediction is crucial for water resources and flood control, and various models including physically based models and data-driven models have been applied for rainfall-runoff simulation and forecasting. Among them, the Long Short-Term Memory (LSTM) network achieved state-of-the-art results in rainfall-runoff modeling powered by the recurrent architecture. However, due to this sequential structure, the LSTM model cannot do parallel computing efficiently and is hard to learn the long-term dependencies between the inputs and output. Recent studies have shown that the attention mechanisms can achieve better performance in time series predictions than the recurrent structures, which gives it great potential for hydrological modeling applications. In this study, we developed a new data-driven model for rainfall–runoff modeling, which is based on the attention mechanisms. The …