Hydrological time series forecasting via signal decomposition and twin support vector machine using cooperation search algorithm for parameter identification

Abstract

Accurate hydrologic forecasting is of vital importance for water resource management and reservoir operation under the changing environments. However, it is difficult for a standalone method to track the complex hydrological time series with highly nonlinear and nonstationary features. To improve the forecasting accuracy, this study proposes a hybrid model using compete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and twin support vector machine (TSVM) optimized by metaheuristic algorithm. Firstly, the CEEMDAN tool is chosen to divide the original runoff data into several relatively stable subcomponents with various resolutions and frequencies. Then, the TSVM forecasting model is established to extract the deep features of each subseries, while the emerging cooperation search algorithm is used to search for satisfying parameter combination. The final forecasting value for the original hydrological time series is formulated by aggregating the outputs of all TSVM models. Long-term streamflow data from several hydrological stations in China’s Yangtze River is used for simulation. Experimental results reveal that the hybrid model significantly outperforms several conventional forecasting models in prediction accuracy. For instance, the proposed method improves the root mean squared error and mean absolute error value with about 36.5% and 41.6% reductions compared with the traditional support vector machine at station B. Thus, a novel and effective evolutionary artificial intelligence model is developed for hydrological forecasting.