Performance of least squares support vector machine for monthly reservoir inflow prediction
MetadataShow full item record
Over the past decade, artificial intelligence techniques have been widely used in the hydrological modeling studies. As favorite artificial intelligence algorithms have disadvantages relating to the local minima problems and slow convergence in some cases, a new alternative Kernel based-artificial intelligence technique called a Support Vector Machine (SVM) has gained popularity in modeling of hydrological variables recently. In this study, least squares support vector machine (LS-SVM), which is a simplified version of SVM, was used in prediction of reservoir inflows of Demirkopru Dam/Turkey. The modeling study applied with input data consisting of the concurrent monthly precipitation, temperature and one-month-ahead precipitation values and the corresponding inflows as output. The maximum determination coefficients (R-2) and the minimum mean square errors (MSE) of LS-SVM obtained with the related model parameters gamma = 16 and sigma(2) = 0.97, were 81.10 % and 932.15 (10(6) m(6)) and 73.58 % and 1228.24 (10(6) m(6)) for the training and testing periods respectively. The LS-SVM results were also compared with another kernel function based approach called generalized regression neural networks (GRNN) and a traditional method multiple linear regression (MLR). When the performances of the training and testing periods are compared, it is observed that LS-SVM approach has better performance for R2 values in the training and testing periods; on the other hand, in terms of MSE values, GRNN proves itself to be successful in the testing period. Furthermore, it was proved with this study that LS-SVM is a successful artificial intelligence technique and can be applied to other hydrological variables with a nonlinear nature to carry out a better performance than would be obtained from traditional techniques and kernel function based neural networks, such as the MLR and GRNN.