Short-term streamflow modeling using data-intelligence evolutionary machine learning models

Area of Science:

• Hydrology and Water Resources
• Computational Intelligence
• Environmental Modeling

Background:

• Accurate streamflow prediction is critical for effective water resource management.
• Machine learning (ML) models offer advanced capabilities for hydrological forecasting.
• Hybridization of ML with optimization algorithms can enhance predictive accuracy.

Purpose of the Study:

• To comparatively evaluate hybrid ML models integrated with bioinspired optimization algorithms (BOA) for short-term, multistep streamflow forecasting.
• To identify the most effective BOA and ML model combination for streamflow prediction.
• To assess the performance of hybrid models against traditional methods.

Main Methods:

• Applied five ML models: XGBoost (XGB), MARS, ELM, EN, and SVR.
• Utilized three BOA: Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Differential Evolution (DE) for hyperparameter tuning.
• Compared model performances using statistical metrics, graphical analysis, and hypothesis testing.

Main Results:

• Hybridization of ML models with BOA significantly improves data-driven streamflow forecasting.
• Particle Swarm Optimization (PSO) outperformed Genetic Algorithm (GA) and Differential Evolution (DE) in optimizing ML hyperparameters.
• The XGBoost (XGB) model, particularly when integrated with PSO, demonstrated superior performance in multistep streamflow prediction compared to other tested models.

Conclusions:

• The proposed XGBoost-PSO hybrid model is a highly effective and superior alternative for daily streamflow forecasting.
• This data-driven approach provides crucial improvements for water resources planning and management.
• Hybrid ML-BOA models represent a promising direction for advancing hydrological prediction accuracy.