Short-Term Streamflow Forecasting Based on Gated Recurrent Unit and Attention Mechanism

High-precision streamflow forecasting has important implications for the water resources efficient planning and rational allocation. Recently, data-driven models represented by deep learning have become a research hotspot. However, existing study less on application of attention mechanism in prediction. This proposes novel model based gated recurrent units mechanism, GRU-AT. The automatically extracts relationship between long-sequence input features prediction targets through uses to extract time-series dependencies within runoff sequence, thereby establishing mapping historical sequences future streamflow. proposed is employed predict next-day flow at Zhutuo station. Six comparative five scoring metrics are validate superiority GRU-AT model. Experiments expresses that surpassed all metrics, verifying can provide accurate short-term results.