Research And Prediction For Monthly Rainfall Based On Model SEPG-S

High precision prediction of medium and long term precipitation is necessary for evaluating the flooding risk correctly and constructing the reliable early warning system of flood and drought reasonably.As for rainfall time series,it has nonlinear and high fluctuate characteristics,so it is difficult to predict them with high precision using the traditional and singular methods.Thought these methods have advantages of easy and simple recognition,they can not achieve the ideal effect in medium and long term forecast for nonlinear rainfall sequence.Because of that,the precipitation estimate combinatorial optimization model SEPGS was build up based on seasonal index adjustment(SIA),ensemble empirical mode decomposition(EEMD),partial auto-correlation function(PACF),generalized regression neural network(GRNN)and seasonal auto-regressive integrated moving average(SARIMA)methods to solve the prediction problem on precipitation time series.First,the seasonal index(SI)was extracted from original rainfall sequence by SIA.Then the monthly rainfall time series with typical nonlinear and nonstationary features was made data processing by EEMD method,so that it was decomposed into some different scale features intrinsic mode functions(IMFs)and one residual sequence.Finally,these IMFs and residual sequence were predicted by GRNN and SARIMA respectively based on their own characteristics.In this paper a rainfall sequence in Jiuquan of Gansu province was as the samples for research,and its result was compared with other prediction methods.The contrast results indicated that: The method which was put forward has high forecast accuracy in this paper.Meanwhile,in order to verify that the proposed method has good predictive accuracy and robustness,it has been used in the precipitation forecast in Dunhuang and Yumen of Gansu province,which has obtained ideal results.In the research process,we also have found: It is convenient to research rainfall time series which was eliminated the effect of SI by SIA Method.The EEMD method can effectively decompose precipitation sequence with nonlinear and non-stationary characteristics,and retain various distribution rule in time and space scales within the original rainfall sequence.PACF has a well auxiliary effect on the determination of the number of neurons in input layer of GRNN,and this approach effectively circumvents the influence of human factors.GRNN and SARIMA played their respective advantages to predict the IMFs and residual sequences with different scale features separately.The method combined above methods not only effectively inherited their own advantages,but also took into account the features of original sequence by itself.Simultaneously,this model is very valuable for forecast problem of nonlinear and non-stationary sequence.Through the comparison with other forecast methods,it found that the combined model had high forecasting precision and a strong generalization.