Improving Prediction Accuracy of Hydrologic Time Series by Least-Squares Support Vector Machine Using Decomposition Reconstruction and Swarm Intelligence

Accurate hydrologic forecasting plays a significant role in water resource planning and management. To improve the prediction accuracy, this study develops a hybrid hydrological forecasting method based on signal decomposition reconstruction and swarm intelligence. Firstly, the ensemble empirical mode decomposition is utilized to divide the nonlinear runoff data series into several simple subsignals. Secondly, the least-squares support vector machine using the gravitational search algorithm is used to recognize the relationship between previous inputs and the target output in each subsignal. Next, the forecasting result is obtained by summarizing the total outputs of all the models. Four famous indexes are used to evaluate the performances of various forecasting models in monthly runoff of two hydrological stations in China. The applications in different scenarios show that the hybrid method obtains better results than several control models. For the runoff at Cuntan Station, the hybrid method makes 58.9% and 52.4% improvements in the root-mean squared error value compared with the artificial neural network and support vector machine at the training phase. Thus, a practical data-driven tool is developed to predict hydrological time series.

Automated summary

Accurate hydrologic forecasting is crucial for water resource planning and management. A hybrid hydrological forecasting method incorporating signal decomposition reconstruction and swarm intelligence has been developed, demonstrating improved prediction accuracy compared to traditional models. The method utilizes ensemble empirical mode decomposition and least-squares support vector machine to predict monthly runoff data from two hydrological stations in China.