Zhao et al. (2025) Water Flow Forecasting Model Based on Bidirectional Long- and Short-Term Memory and Attention Mechanism

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-07-16
• Authors: Xinfeng Zhao, Shengwen Dong, H. Raghav Rao, Wuyi Ming
• DOI: 10.3390/w17142118

Research Groups

[Not specified]

Short Summary

The study proposes the AT-BiLSTM model, which integrates a bidirectional LSTM layer and an attention mechanism, to improve the accuracy of river water flow forecasting.

Objective

• To develop a prediction model capable of capturing long-distance dependencies and focusing on key temporal features of water flow time series to enhance forecasting precision for flood control and resource allocation.

Study Configuration

• Spatial Scale: Shizuishan monitoring station, Yellow River, China.
• Temporal Scale: 72 h (3 days) forecast horizon.

Methodology and Data

• Models used: AT-BiLSTM (Bidirectional Long Short-Term Memory with Attention layer), RNN, and four other comparative models.
• Data sources: Actual monitoring dataset from the Shizuishan station.

Main Results

• The AT-BiLSTM model outperformed the RNN model, reducing Mean Absolute Error (MAE) by 27.16%, Mean Squared Error (MSE) by 42.01%, and Root Mean Square Error (RMSE) by 23.85%.
• The proposed model demonstrated the best predictive performance among the six models evaluated.
• For a 72 h forecast period, the model maintained an average prediction error of less than 6%.

Contributions

• Introduces a hybrid architecture that combines bidirectional information processing with an attention mechanism to better handle the non-uniform contributions of different time steps in water flow sequences compared to standard LSTM models.

Funding

[Not specified]

Citation

@article{Zhao2025Water,
  author = {Zhao, Xinfeng and Dong, Shengwen and Rao, H. Raghav and Ming, Wuyi},
  title = {Water Flow Forecasting Model Based on Bidirectional Long- and Short-Term Memory and Attention Mechanism},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17142118},
  url = {https://doi.org/10.3390/w17142118}
}