Application Of Rough Set And Flex In Mid-long Term Runoff Forecasting

Water resource is essential for human being and the society's sustainable development. The mid-long term runoff forecasting is important for the protection, exploitation, management of water resource, it uses mathematical models to forecast runoff, and the forecast period preponderats over the maximum concentration time of the watershed. Its result provids an important basis for the mid-long term operation of water resource, and is widely used in the fields of reservoir operation, hydropower scheduling, flood control and drought combat, allocation of water resource, shipping management. However, due to the long forecast period, and the complexity of the formation mechanism of runoff, coupled with the impact of human production activities, the accuracy of the mid-long term runoff forecasting is not high.This paper introduces the rough set theory to the mid-long term runoff forecasting, selects the most influential fators from the historical runoff data, excludes secondary factors, then uses support vector regression model to predict runoff based on the core factor set. By using Flex, it designs and develops a mid-long term runoff forecasting system, which has the strong mutual ability and the rich rendering ability, and the system can support the optimal scheduling of hydropower. The main works and research achievements are as follows:1)The rough set theory is proposed for the study of the mid-long term runoff forecasting. The paper introduces its modeling and application with the yearly runoff data of the Yamadu station on Yili river in Xinjiang. It uses an algorithm based on the class-information entropy for the discretization of condition attributes, and a genetic algorithm is used for attribute reduction, then the core factor set would be gained. As the rough set doesn't need any additional information outside the necessary data collection, its description and processing for the issue of mid-long term runoff forecasting are very objective. The rough set discretizes attributes, eliminates redundant information, it reduces training data for SVM significantly, and thses improve the system's speed.2)As the rough set can only deal with discrete data,and its anti-interference ability is poor, the SVM theory is proposed for runoff forecasting. The SVM as a post-system has good fault tolerance and anti-interference capability. It uses genetic algorithm and cross validation to optimize the relevant parameters.It gets rid of the blindness and subjectivity on parameters selection, and improves the prediction's accuracy. The forecasting results show that the rough set combining with the SVM has better generalization and anti-interference capability and gains a good prediction performance.3)A mid-long term runoff forecasting system has been developed with Flex. Flex has a powerful component library and can make custom components flexibly. The system is a web application, which has the strong mutual ability ,the rich rendering ability and a perfect user experience. The Web application, which is developed by Flex, implements the MVC framework, and separates the user control logic from the server code, so the system's structural levels are clear. Flex links data with the interface components closely, the client communicates with the server asynchronously. The user doesn't need to stop for data refresh when the data is transporting.4)The mid-long term runoff forecasting system calls the algorithm dynamic link libraries with JNI in the server-side. The libraries which are written by C++, are old modules. They make up for the shortage of Java's performance, improve the operation speed, and also protect the intellectual property of the prediction algorithms. On the basis of ensuring the system's functions, they shorten the development cycle, reduce the costs, and protect the historical investment.