Prediction Of Precipitation Based On The Divide-Integration Method Of Three Times In The Source Region Of Yellow River

The source region of the Yellow River is located in the northeastern of the Tibetan Plateau with an area of 122 thousand square kilometers,which is an important part of water source of the Yellow River Basin.The precipitation is the main recharge of runoff and water source of the source region,therefore,the rainfall forecast in this region has become an important part of the hydrological and meteorological observation.The forecasts can not only provide data support for the decision-making of hydrological research,but also improve the ability to respond to the disasters.In consideration of the fact that the rainfall of the source region of the Yellow River is influenced by the climatic variation seriously,the monthly rainfall data from 1998 to 2008 of Jimai,Maqu,Tangnaihe are selected as the representative of the source region of the Yellow River to do the research.Firstly,the ensemble empirical mode decomposition(EEMD)is used as the denoising method for the original sequence.Secondly,the seasonal adjustment method is used to decompose the smooth series to several seasonal indexes and a trend series.Thirdly,the full decomposition of the trend series with complete ensemble empirical mode decomposition(CEEMD)and then several intrinsic mode functions(IMFs)would be obtained.As for the IMF1,the support vector regression(SVR)is chosen as the main model to do the prediction.As for the rest of IMFs,the extreme learning machine(ELM)is selected as the main model to do the forecast.Finally,after summing all IMF forecasts as the prediction of trend series,the final forecasts of the original sequence would be the aggregation of the seasonal indexes and the forecasts of the trend series.At last,the fitting effect of the final forecasts is good.As for the reason that the processing order of the proposed hybrid model is EEMD-SAM-CEEMD-SVR-ELM,and then it is named ESCSE.Considering that ESCSE involves many sub models,the forecasts of ESCSE along with its sub models are selected to compare with the true values.And the final comparison results have shown that the ESCSE model owns the higher fitting accuracy.In the meantime,the hybrid model fits the trend and fluctuation of the monthly rainfall series well,so it can provide decision support for the local rainfall forecast and the hydrological research.