Runoff forecasting model based on variational mode decomposition and artificial neural networks.

Area of Science:

• Hydrology
• Water Resource Management
• Artificial Intelligence in Environmental Science

Background:

• Accurate runoff forecasting is crucial for effective water resource management.
• Decomposition-based models show promise but often neglect inter-component correlations.
• Existing models may not fully leverage the complex dynamics within runoff data.

Purpose of the Study:

• To propose a novel hybrid model, Variational Mode Decomposition-Convolutional Neural Network-Long Short-Term Memory (VMD-CNN-LSTM), for enhanced runoff forecasting.
• To investigate the effectiveness of incorporating time-delay and correlation information among decomposed sub-signals.
• To improve the accuracy and robustness of runoff forecasting compared to traditional methods.

Main Methods:

• Variational Mode Decomposition (VMD) to decompose runoff series into intrinsic mode functions (IMFs).
• Convolutional Neural Networks (CNN) to extract features from a 2D matrix representing time-delay and correlation information of IMFs.
• Long Short-Term Memory (LSTM) networks to forecast runoff using extracted features from CNN.

Main Results:

• The proposed VMD-CNN-LSTM model demonstrated superior performance compared to baseline models in runoff forecasting.
• The model effectively utilized the correlation information among decomposed sub-signals via the CNN component.
• Experiments on monthly runoff data from the Wei River showed the model's robustness and stability across different lead times.

Conclusions:

• The VMD-CNN-LSTM hybrid model offers a significant advancement in runoff forecasting accuracy and reliability.
• Integrating VMD with CNN and LSTM effectively captures complex temporal dependencies and inter-component correlations.
• This approach provides a valuable tool for water resource management and hydrological forecasting.