SVM-Based Visual Modeling System for Enhancing the Flexibility of Interactive Runoff Forecasting

In order to address the problems involved in low modeling efficiencies and complex programming design involved in implementing runoff forecasting using conventional modeling technologies, this paper presents a novel visual modeling system that integrates the visual modeling technology with the support vector machine (SVM) based model with the purpose of enhancing the flexibility of interactive and fast modeling of the complex, non-linear,and dynamic runoff process. The workflow for visual modeling consists of prediction schema creation, data collection and pre-procession, SVM model design,training and actual runoff forecasting. The system enables users to build up a model schema for a particular study through combination of appropriate model components. Users´ modeling activity can be strongly supported by the visual tool where they can interact with, manipulate and analyze data, with a very small training effort. The system has been demonstrated on the daily river runoff forecasting of the Qingjiang river basin. The results show that the system can greatly enhance easy-to-use capabilities of visual modeling and offers a positive prospect of improving the efficiency and effectiveness of current practice in hydrological modeling.