Study On Forecasting Methods Contrast Of Hydrologic Time Series

Water resource is not noly basic natural resources of human survival and development, but also strategic natural resources of society development. Especially in the dry regions, the shortage of water resources and deterioration of ecological evironment have become the most serious problem. In this region, dry climate, rare rain and strong evaporation make hydrological process show very complex change process, hydrological time series present highly nonlinear and uncertainty. Time series analysis plays an important part in hydrological regularity analysis and hydrological forecasting. Researching hydrological time series in dry region, which can fully show change regularity of hydrological systems, and provide scientific basis for sustainable utilization of water resources in this region.In order to excavate the forecasting methods are appropriate for hydrological time series of dry region, this paper chose three typical methods of stepwise regression, support vetor machine and grey self-memory to research from traditional and modern times. According to hydrological time series of dry region change greatly, and complex influence factors, this paper will research these methods in aspect of multivariate modeling, through comparative analyses, the main achievements were as follows:(1) The stepwise regression model is chosen from traditional forecasting methods, and used in Minqin region, to examine regional apaptability of the model. Compared with other methods, this paper analyses value of traditional forecasting methods in Minqin region.(2) KPCASVM model was built according to kernel principal component analysis theory and support vector machine. This paper gave the detailed process of the modeling, in application of Minqin region, the results have shown that SVM model after feature extraction has a batter forecast accuracy.(3) The multivariate grey self-memory model was built according to the grey theory and self-memory theory. In application of Minqin region, the multivariate grey self-memory had better forecast performance compared with univariate grey self-memory model, forecast accuracy improved efficiently, this study tries to provide references for continuous research of self-momory in the future.