Monthly streamflow forecasting based on improved support vector machine model

To improve the performance of the support vector machine (SVM) model in predicting monthly streamflow, an improved SVM model with adaptive insensitive factor is proposed in this paper. Meanwhile, considering the influence of noise and the disadvantages of traditional noise eliminating technologies, here the wavelet denoise method is applied to reduce or eliminate the noise in runoff time series. Furthermore, in order to avoid the subjective arbitrariness of artificial judgment, the phase-space reconstruction theory is introduced to determine the structure of the streamflow prediction model. The feasibility of the proposed model is demonstrated through a case study, and the results are compared with the results of artificial neural network (ANN) model and conventional SVM model. The results verify that the improved SVM model can process a complex hydrological data series better, and is of better generalization ability and higher prediction accuracy.