Integrated metaheuristic algorithms with extreme learning machine models for river streamflow prediction.

Area of Science:

• Hydrology
• Water Resource Management
• Computational Intelligence

Background:

• Accurate river streamflow prediction is crucial for water resource planning and flood management.
• Traditional forecasting models face challenges with nonlinearity, stochasticity, and convergence reliability.
• Developing robust and accurate streamflow prediction models is an ongoing scientific challenge.

Purpose of the Study:

• To introduce and evaluate novel hybrid models for river streamflow forecasting.
• To compare the performance of extreme learning machines (ELM) integrated with various metaheuristic optimization algorithms.
• To assess the predictive accuracy, convergence, and stability of these hybrid models.

Main Methods:

• Developed 20 hybrid models by combining ELM with eight metaheuristic optimization algorithms (PSS, INFO, RUN, etc.).
• Utilized streamflow data from the Aswan High Dam on the Nile River for model training and validation.
• Performed a comparative analysis of model performance using metrics like RMSE, R, NSE, MAPE, MAE, and KGE.

Main Results:

• Mathematically inspired metaheuristic models demonstrated superior predictive accuracy, convergence, and stability.
• The Pareto-like sequential sampling-ELM (PSS-ELM) model achieved high performance (RMSE: 2.0667, R: 0.9374, NSE: 0.8642).
• INFO-ELM and RUN-ELM models showed robust convergence and high Kling-Gupta efficiencies (0.9113, 0.9124).

Conclusions:

• The proposed hybrid ELM models significantly enhance river streamflow forecasting capabilities.
• These models offer improved solutions for water management strategies and risk reduction.
• Adoption of these advanced models can lead to more effective resource planning and flood control.