Prediction On Rainfall Based On BP Neural Networks

In agricultural economic production activities, the most important fact that influences agriculture production decision is the change of the amount of rainfall. How to predict the amount of rainfall accurately has a great significance in agricultural production. However, the amount of rainfall is a nonlinear and unstable time series and influenced by lots of factors, the prediction of it is very difficult. Because of the prediction accuracy on the amount of rainfall is based on the build of data processing methods and prediction models to a great extent, this essay carried out researches to improve prediction accuracy of the amount of rainfall:(1)Study on prediction methods of the amount of rainfall. We proposed a new prediction model EMD-GS-IV-REMCC-BPNN based on empirical mode decomposition, geostatistics and IV-REMCC-BPNN model that merged early-stage laboratories’variable selection and network optimization into it. The specific steps were stated below: This paper began with the decomposing rainfall sequence into intrinsic mode function by using EMD. Then, implemented extension order based on GS for every obtained components and carried out independent prediction based on IV-REMCC-BPNN. Finally, added each prediction result of component together to get the prediction value of the amount of rainfall. The rainfall experiment data that obtained in literature was used to predict based on EMD-GS-IV-REMCC-BPNN model, and the results showed that compared to the traditional models, EMD-GS-IV-REMCC-BPNN model has preferable Prediction accuracy and generalization ability.(2)One-dimensional time series prediction of the amount of rainfall. It analyzed the rule of each period observation by the historical observations of the amount of rainfall to predict the tendency in future. This essay applied EMD-GS-IV-REMCC-BPNN model predict the one-dimensional time series of rainfall in Changsha,Yueyang and Dingcheng district of Changde. The results showed that the prediction accuracy of EMD-GS-IV-REMCC-BPNN model is much better than GS-BPNN、GS-REMCC-BPNN、 GS-IV-REMCC-BPNN and EMD-GS-REMCC-BPNN. The EMD-GS-IV-REMCC-BPNN has some advantages of stable high prediction precision and effective and reasonable, so it has extensive application prospect for regional rainfall forecast.(3)The influence of sunspot numbers that works for rainfall-forecast. In practical research, not only internal factors but also external factors should be taken into consideration when forecasting rainfall. On the basis of the meteorological factors applied in traditional weather forecast analysis model is too numerous and jumbled, It’s so hard to analyze. The climate change of earth is affected by solar activity and the numbers of sunspot is an important factor for solar activity. Therefore, this paper selected the monthly historical data of sunspot as an important factor for rainfall-forecast, and applied it to independent predication for monthly rainfall forecast of Yueyang and Dingcheng district of Changde., The results indicates that after introducing sunspot as a influence factor, The EMD-GS-IV-REMCC-BPNN model constructed in this paper had a better prediction over BPNN model、 SVR model that opens up a new way for selection of influence factors of rainfall-forecast, it provided an effective reference for academic to predict the rainfall-forecast and sunspot data.