Study On The Price Prediction Of Agro-prod Ucts

Agriculture as the primary industry of our country. In recent years, in order to quicken rural economy by promoting agricultural informatization and then keep the operation of china’s agriculture stable, china government introduces related polices constantly. However, the price fluctuation of argo-products leading to unbalanced demand and supply of markets, thus restricting sustainable development of agriculture. Therefore, on the basis of making timely predictions on the price of argo-products, in one aspect, it helps the macro-control of China’s argo-products markets, in the other aspect, it contributes to maintaining the development of China’s argo-products markets.Due to the original data of the price of argo-products has some features, like continuity and time sequence, it’s hard to make an accuracy prediction. On the purpose of improving the prediction accuracy of the price of argo-products, this paper conducts researches from several ways stated below:(1) Research on prediction methods of argo-products price. Commonly used methods on one-dimensional and multi-dimensional time series prediction models were compared. This work introduced multi-dimensional time series prediction models GS-RSR-SVR (SVR based on Geo-statistics and Reasonable Sample Rejection), which fused time series analysis and regression analysis by nonlinear way in our previous study. Order determination, variable selection and training sample selection are the important and difficult problems in time series analysis; however GS-RSR-SVR solved these problems satisfactorily and with high and robust prediction accuracy. GS-RSR-SVR was just suitable for multi-dimensional time series prediction in early stage; we optimized the model and generalized it to one-dimensional time series prediction on the basis of previous study, which was more universal.(2) One-dimensional time series prediction of argo-products price. One-dimensional time series prediction analyzed the rule of each period observation by historical observations to predict the tendency in future. GS-RSR-SVR was used to predict one-dimensional time series data of four kinds of agricultural product monthly price, including cotton, pig, pork and corn. The prediction accuracy of GS-RSR-SYR was much better than commonly used one-dimensional time series models such as quadric exponential smoothing and trend extrapolation, and all the mean square error were less than8%in four kinds of agricultural product price. The results showed that, with advantages of high prediction accuracy, high efficiency, rationality and stability, GS-RSR-SVR held extensive application prospect in the field of time series forecasting.(3) Multi-dimensional time series prediction of argo-products price. In actual research, not only internal factors but also external factors should be taken into consideration when forecasting agricultural product price. Commonly used multi-dimensional time series models were on the premise of influence factors in the same period were known, which was not real forecasting. The forecasting of variables that not happened was real forecasting. Therefore, this paper preprocessed the original data by moving up the dependent variables, namely utilized the independent variables of moment t to forecast the dependent variables of t+1. Taken the price and influence factors of cotton and pig as examples, GS-RSR-SVR was used to forecast the two multi-dimensional time series datas, the results showed that the prediction accuracies of GS-RSR-SVR were much better than BPNN and SVR, which further indicated that GS-RSR-SVR held excellent application prospect in the price prediction of argo-products, and provided beneficial reference on price prediction in other economic fields.