Forecasting Venezuelan Caroni River Flow through Support Vector Machines
Cesar O. Seijas1, Sergio Villazana1, Jorge Guevara2, Edilberto Guevara2
1Electrical Engineering school, University of Carabobo, Valencia, Carabobo, Venezuela; 2Civil Engineering school, University of Carabobo, Valencia, Carabobo, Venezuela

The aim of this research is to model the time series formed by the monthly flows of Caroni River in Venezuela, from 1950 to 2003 using regression based on Support Vector Machines (SVR). The mentioned time series was preprocessed by extracting the trend and seasonal components trough Independent Component Analysis (ICA) and the resulting stochastic series was modeled as a Nonlinear Autoregressive Moving Averaging process (NARMA), of order defined by Singular Value Decomposition (SVD), using SVR. The model was validated by obtaining a prediction error in the flow, lower than traditional statistical models. The results of this study demonstrate the strength of nonlinear computational models to predict river flows.

Keywords: Regression with support vector machines; Flow modeling; Independent component analysis; Singular value decomposition

Biography: Venezuelan, Electrical Engineer, Full Professor at Electronic and Communication Department of Electrical Engineering School, Faculty of Engineering, University of Carabobo. Magister in Electrical Engineering, Dr. in Engineering. Areas of investigation: Emerging Computing, Artificial Intelligence Applications in the area of Support Vector Machines applied to Bioengineering, Time Series Estimations applied to Environment and Finances; Process and Electric Controls.