Flood Forecasting Based On Clustering Analysis

As a discipline of studying the law of nature ,the hydrology which involved numerous data greatly rely on the observation samples.It is a basic path to quest new discipline depend on the numerous observation samples to resolve the hydrologic problems. Flood forcast is the direct result of the development of Hydrological phenomena. People urgent need to know the quantity of flow of peak flood and when to arrive to reduce losses of life and property in practice. In order to improve the accuracy of flood forecast, feature selection and clustering analysis in data mining were introduced.Classify calibration of parameters is an important mean to improve the precision of forecast, and parameters of flood forecasting model are determined by flood samples. Considering that rainfall is the immediate cause of the flood., it is necessary to extract representative features of each rainfall and analysis by clustering, in order to improve the precision of classify calibration. Inevitably, there is some redundancy among the selected features, which would lead to more errors in clustering results. In this respect, flood clustering analysis based on unsupervised feature selection was proposed. Mitra algorithm was employed to se(?)ect features of flood samples, and Kmeans and FCM were employed as the methods of flood clustering. Results indicated that Mitra algorithm is effective in reducing redundant features. The results of flood clustering by Kmeans and FCM respectively have good agreement and the features of each flood category is obvious.In shaxikou flood forecast system,as a example, xin 'an river hydrological model and the flow algorithm Muskingum were adoptd. We give parameters belond to each kind of floods and provide the forcasting results using relevant parameters which precision is little imiproved compared with the original results.