Research On Grain Yield Prediction Model For Small Sample Data

Grain Security is a worldwide strategic problem, related to the harmonious development of all countries in the world and the overall progress of human society. Grain production forecasting, not only for the local government to provide food production technical support and theoretical basis, and plays a very important role in ensuring the national food security and maintaining the social stability.This paper relies on the grain production data published by the national bureau of statistics, and analyses the characteristics of small sample based prediction model. Then the thesis introduces several typical prediction models for small samples which are now widely used. One improved algorithms based on Grey System Theory and SVM regression is proposed and brings new ideas to the prediction of grain production. The main works of this paper include the following five parts:Firstly, analyzing the small sample nature of grain production forecasting, and several prediction method was analyzed:Time series analysis method, support vector machine forecasting method, gray sequence forecast method and combination forecast method.Secondly, after analyzing the data of grain production and the applicability of the GM (1,1) model, conclusion is drawn that this model is inappropriate for the grain production forecast. Thus this paper tries to use grey Verhulst model to forecast the grain production for the first time. Based on the statistic data of China grain production from2002to2009, this study uses GM (1,1) model and the grey Verhulst model to forecast the grain production respectively.In ?-support vector machine we need to determine the ? parameter of the loss function in advance, but what values to take is difficult to determine. Thus this paper try to use v-support vector machine regression method as the foundation of the combination model.Based on the theory of the HP filter analysis a new combined model of gray Verhulst model and SVM regression is proposed. The core concept of it was using HP filter method to extract the trend sequence of grain production, and then using grey Verhulst model to forecast the trend sequence. The prediction result of the grey Verhulst model as the SVM input, residual data as the output of SVM. Adding the predicted residual data with original data and finally get the prediction result. A simulation experiment result of this method and the single grey forecasting models were compared and shows that the model is effective.