Research On Grain Output Prediction Based On Combination Method

Agriculture is the foundation of the national economy.Grain production is an important part of food security,how to predict the yield accurately is the key problem researched in the world.It is very important to use reasonable methods and models to predict the trend of food production.According to the characteristics of China's grain output data,the grain yield forecast is studied from two aspects.On the one hand,considering from the impact factor,the mapping relationship between grain yield and impact factor is established.Two combined forecasting models are proposed.In model one,the impact factors are selected by computing the gray correlation,then according to the Nonlinear characteristics of impact factors and the grain yield data,the BP neural network model is established,and its weight is further optimized by particle swarm optimization algorithm.Lastly,the grain yield is predicted by using this model and the residual error is got and analyzed.In the second combination forecasting model,the correlation coefficient between the impact factors are constructed by using principal component analysis.Lastly,the extreme learning machine model of the main component is established and the grain yield is predicted.On the other hand,the grain yield forecasting model is established based on the single time series of grain production,considering that the grain yield data has the characteristics of complexity,randomness and non-stationary,the forecasting model which combining the GM(1,1)model and ARIMA model based on wavelet transform in this paper.Aiming at the different characteristics of each component sequence after wavelet decomposition,the gray GM(1,1)model is used to predict the approximate component,then the ARIMA prediction model is used to predict the detail component.Finally,the prediction value of the yield is obtained by wavelet reconstruction.By comparing and analyzing the actual value of grain yield and the predicted value of the model,the prediction accuracy of each prediction model is calculated.The results show that the average error of BP network prediction model based on particle swarm optimization is1.7%,The average error of the prediction model based on principal component analysis andextreme learning machine is 1.3%,and the average error of the GM(1,1)-ARIMA prediction model based on wavelet variation is 0.8%.Compared with other predictive models,the combined method can effectively improve the prediction accuracy.The combined model proposed in this paper can predicts the grain yield effectively and provides a new method for grain yield prediction in short time.