Developing a novel hybrid model based on GRU deep neural network and Whale optimization algorithm for precise...

Area of Science:

• Hydrology
• Artificial Intelligence
• Environmental Science

Background:

• Streamflow is a critical indicator for assessing human and climate impacts on the hydrological cycle.
• Accurate streamflow prediction is essential for water resource management and flood control.

Purpose of the Study:

• To develop an innovative deep neural network (DNN) structure for enhanced mean monthly streamflow prediction.
• To integrate a double Gated Recurrent Units (GRU) model with a whale optimization algorithm (WOA) for improved accuracy.

Main Methods:

• A hybrid 2GRU×-WOA model was developed, incorporating a multiplication layer and meta-heuristic optimization.
• Feature selection using Pearson's correlation coefficient (PCC) and Cosine Amplitude Sensitivity (CAS) identified precipitation (P_m) as the key input.
• The model was optimized with specific parameters: tanh-softsign activation, 0.5 dropout rate, and 70 hidden neurons.

Main Results:

• The hybrid 2GRU×-WOA model achieved superior performance with R²=0.79, NSE=0.76, MAE=0.21 (m³/s), MBE=-0.11 (m³/s), and RMSE=0.36 (m³/s).
• Compared to benchmark GRU and Bi-GRU models, the hybrid model showed a 6.8% increase in R² and a 20.4% reduction in RMSE.
• Individual GRU and Bi-GRU models yielded lower performance metrics (e.g., R² of 0.59 and 0.66, respectively).

Conclusions:

• The proposed hybrid 2GRU×-WOA model offers a significant advancement in streamflow prediction accuracy.
• This approach provides a robust tool for hydrological forecasting, aiding in better water resource management.
• The study highlights the effectiveness of integrating deep learning with meta-heuristic algorithms for complex environmental modeling.