Uncertainty Forecasting for Streamflow based on Support Vector Regression Method with Fuzzy Information Granulation

Abstract

Accurate and comprehensive forecasting of streamflow plays an important role in the uncertainly analysis of the hydrologic system. It is widely accepted that prediction interval (PI) can provide more precise and detailed information than deterministic forecasting when the uncertainty level of streamflow increases. Support vector regression (SVR) is a supervised learning model for classification and regression analysis based on associated learning algorithms. In this paper, fuzzy information granulation (FIG) is combined with SVR model (FIG-SVR) for uncertainty forecasting of streamflow. On behalf of evaluating the performance of the forecasting results, the evaluation metrics of point forecasting and interval prediction results are introduced. The real streamflow data from the Three Gorges in the Yangtze River are used to validate the proposed method based on the proposed method. The results show that the proposed method provides the high-quality point predictand and PIs, and the uncertainly of streamflow can be well handled.