Comparison And Study Of The Different Prediction Methods About Rainfall In Xining City

Xining is the capital city of one of the least rainfall, drought has plagued the development of agriculture in Xining. Therefore, the study of rainfall may play a role in drought mitigation of Xining. Flood disasters are more frequent, in the event of floods, it will make losses to people's lives and huge property ,to social stability it also will have beend affected. Therefore, the rainfall prediction study will make great significance on the reduction of Drought and flood.In this paper we based on rainfall and the meteorological data in 1955 -2006 in Xining, respectively, using multiple linear regression model, autoregressive model, artificial neural network model and the gray Theoretical model to make the prediction of rainfall, and to test the feasibility of rainfall forecast, the results of this paper describes the application of these methods are feasible to forecast rainfall. To single forecast model, the artificial neural network model maked the highest reliability. The BP network has strong nonlinear mapping ability, to avoid the forecast factors and forecast to determine the amount of function. While the black box because the network effect, not the predictors and prediction of nonlinear relationship between restrictions on the polynomial and other simple functions, and enhance the reliability of prediction.This paper also attempts using the linear combination forecast model and the nonlinear combination forecast model for the prediction of rainfall, because the combined forecasting model comprehensive utilized each individual prediction model, to avoid unstable of the single forecasting model, predictions were more stable, and the predicted results also shows a certain degree of accuracy of the combination forecasting model higher than the single prediction model. Especially the non-linear combination forecast model, both fit the average relative error (3.76%), or test, the average relative error (10.60%) are less than any single forecasting model and linear combination model, that forecastig nonlinear portfolio model to predict the annual rainfall is feasible and worth further study and promotion.However, when using these methods used in the forecast above is not feasible in forecasting the monthly rainfall, this has been proved with examples in this paper. Including multiple linear regression model and BP neural network model predicted a few months of fitting results are satisfactory, but the ideal test results only is August. Illustrate these methods in the prediction of monthly rainfall can not be universal.