Related Experiment Video
Updated: Jul 11, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
SMGformer: integrating STL and multi-head self-attention in deep learning model for multi-step runoff forecasting
Wen-Chuan Wang1, Miao Gu2, Yang-Hao Hong2
1College of Water Resources, North China University of Water Resources and Electric Power, Zhengzhou, 450046, China. wangwen1621@163.com.
Scientific Reports
|October 9, 2024
Summary
This study introduces the SMGformer model for accurate runoff forecasting, significantly improving upon existing methods. The model enhances water resource management and disaster reduction by providing more reliable predictions.
Area of Science:
- Hydrology and Water Resources Engineering
- Artificial Intelligence in Environmental Science
- Time Series Forecasting
Background:
- Accurate runoff forecasting is crucial for water resource allocation, flood control, and disaster reduction.
- Runoff sequences exhibit inherent randomness, posing significant challenges for traditional forecasting models.
- Nonlinear and non-stationary characteristics of runoff data complicate prediction accuracy.
Purpose of the Study:
- To propose a novel SMGformer runoff forecast model integrating advanced deep learning techniques.
- To address the challenges posed by the randomness and complexity of runoff sequences.
- To enhance the accuracy and extend the effective forecast period for monthly runoff prediction.
Main Methods:
- Seasonal and Trend decomposition using Loess (STL) for data preprocessing.
- Informer's Encoder layer to capture key input features.
- Bidirectional Gated Recurrent Unit (BiGRU) for temporal information learning.
- Multi-head Self-Attention (MHSA) mechanism to emphasize important features.
- Integration of these components within the SMGformer architecture.
Main Results:
- The SMGformer model demonstrated superior performance compared to the Informer model.
- Significant reductions in Mean Absolute Error (MAE) by 42.2% and 36.6% at the 1st step.
- Substantial decreases in Root Mean Square Error (RMSE) by 37.9% and 43.6% at the 1st step.
- Marked improvements in Nash-Sutcliffe Efficiency (NSE), increasing from 0.936 to 0.975 and 0.487 to 0.837.
- High Kling-Gupta Efficiency (KGE) values (≥0.8) maintained at the 3rd step.
Conclusions:
- The SMGformer model effectively captures essential information within monthly runoff sequences.
- The proposed model offers a significant advancement in runoff forecasting accuracy and reliability.
- SMGformer extends the effective prediction horizon for hydrological forecasting, aiding water management and disaster preparedness.
