Application of empirical mode decomposition, particle swarm optimization, and support vector machine methods to...

Area of Science:

• Hydrology
• Water Resource Management
• Computational Intelligence

Background:

• Accurate streamflow modeling is essential for water resource planning, flood control, and drought management.
• Hybrid models offer potential for improved hydrological predictions.

Purpose of the Study:

• To evaluate the performance of hybrid models combining Least Square Support Vector Machines (LSSVM), Empirical Mode Decomposition (EMD), and Particle Swarm Optimization (PSO) for monthly streamflow modeling.
• To identify the optimal model configuration and input variables for streamflow prediction in semi-arid regions.

Main Methods:

• Utilized 42 years of monthly average streamflow data from two stations in the Konya Closed Basin (1964-2005).
• Employed LSSVM, EMD, and PSO techniques to construct hybrid models.
• Selected lagged streamflow values based on partial autocorrelation as model inputs.
• Assessed model performance using metrics like MSE, RMSE, and correlation coefficients, visualized with Taylor and Violin diagrams.

Main Results:

• The hybrid EMD-PSO-LSSVM model outperformed individual LSSVM, PSO-LSSVM, and EMD-LSSVM models.
• EMD improved the performance of LSSVM and PSO-LSSVM models by 1-5% (based on correlation coefficient R).
• Optimal input combination for semi-arid regions included 1-, 9-, 10-, 11-, and 12-month lagged streamflow values.

Conclusions:

• The EMD-PSO-LSSVM hybrid model is highly effective for monthly streamflow modeling in semi-arid climates.
• Empirical Mode Decomposition significantly enhances the predictive accuracy of LSSVM-based models.
• The study provides valuable insights for optimizing hydrological forecasting systems.