Medium- and Short-Term Forecasting Of Grain Consumption：One Dynamic Model Combining Modified ARIMA and Multiple Regression Model

Abstract

In order to improve predictive accuracy of medium- and short-term grain consumption, this paper have proposed a new data pre-processing method and combined prediction model. First, the original data are subjected to stabilizing process by optimal gliding barycenter grouping method to remove the abnormal points, which include the original grain consumption and several impact factors. Then, the relational degree between grain consumption and the impact factors is evaluated and the key impact factors are chosen as the input to a multiple regression model to forecast urban ration, urban feed grain rural ration and rural feed grain consumption year by year respectively, which are summed as total food grain consumption. Besides, one modified Box-Jenkins model is used to predict the trend of impact factors, in which the stabilizing preprocess and the residual test are increased compared with the general Box-Jenkins model, which can provide higher fitting accuracy. Thereby, the training data of grain consumption and the corresponding key impact factors can be updated with year. The simulation analyse have shown that the proposed model can capture the dynamics of the impact factors, and the prediction accuracy on the grain consumption is improved significantly, compared to the existing models.