Related Experiment Videos
Multi-feature fusion monthly runoff prediction under different climate conditions using APO-optimized
Wen-Chuan Wang1, Yi-Fei Wang2, Wei-Can Tian2
1College of Water Resources, North China University of Water Resources and Electric Power, Zhengzhou, 450046, China. wangwen1621@163.com.
Scientific Reports
|June 29, 2026
Summary
This study introduces a hybrid deep learning model for accurate monthly runoff prediction, outperforming traditional methods by capturing complex hydrological dynamics. The novel approach enhances forecasting accuracy across different climate zones.
Area of Science:
- Hydrology
- Artificial Intelligence
- Deep Learning
Background:
- Monthly runoff prediction is challenging due to nonlinear and nonstationary hydrological data.
- Traditional single models struggle with long-term dependencies and abrupt changes in runoff sequences.
Purpose of the Study:
- To develop a hybrid deep learning model for improved monthly runoff prediction accuracy.
- To integrate Arctic Puffin Optimization (APO), Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Units (BiGRU), and Self-Attention mechanisms.
Main Methods:
- A hybrid model combining CNN for local feature extraction, Self-Attention for feature weighting, and BiGRU for dependency detection.
- Hyperparameter optimization using the Arctic Puffin Optimization (APO) algorithm.
- Validation using Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Nash-Sutcliffe Efficiency (NSE), correlation coefficient (R), and Kling-Gupta Efficiency (KGE) at Yingluoxia (YLX) and Manwan (MW) stations.
Main Results:
- The hybrid model significantly reduced Mean Absolute Percentage Error (MAPE) by 26.681% and improved Nash-Sutcliffe Efficiency (NSE) by 5.498% at the YLX station.
- At the MW station, the model reduced MAPE by 21.874% and enhanced NSE by 3.673%.
- The APO algorithm showed superior convergence compared to Whale Optimization Algorithm (WOA) and Grey Wolf Optimizer (GWO).
Conclusions:
- The proposed APO-CNN-BiGRU-Self Attention model offers robust and accurate monthly runoff prediction.
- The hybrid approach effectively addresses the nonlinear and nonstationary characteristics of hydrological data.
- This model demonstrates superior performance across diverse climatic zones compared to standalone models and other optimization algorithms.