A Comparative Study Of Grain Yield Prediction Method Based On The Land Use Planning

The grain yield prediction has an important strategic position in China. It is the basisof socio-economic development and population control strategy, as well as the basis for ourcountry to establish land use strategy. It is also a need to realize national food security.Taking Shanxi province as an instance, we conduct in-depth research on prediction methodof grain yield and attemp to find applicable conditions of prediction method of grain yield.The purpose of our research is to improve the prediction accuracy for grain yield, providevalid reference for the provincial grain planning strategy and systematic predictions of thegrain yield.This dissertation has the following contents. Firstly, the essay explains researchbackground and purposes, elaborates the main research methods and content, and putsforward the structure and route for our research. Secondly, the author summarizes thecurrent researches of grain yield prediction method and then selects linear regressionanalysis, exponential smoothing method and BP neural network model to analyze theirmodeling theory.Then we provide general information about Shanxi province the studyarea its status of food production and its features of the grain yield. Lastly, usingmultivariate linear regression analysis model, exponential smoothing model and BP neuralnetwork model, we make prediction about the grain yield in Shanxi province and takecomparative study on their prediction results, model theory, data requirement and modelingprocess.Based on previous researches, this paper adopts multivariate linear regressionanalysis model, exponential smoothing model and BP neural network model to predict thegrain yield in Shanxi province. We obtain the following findings:(1)if the grainyield has an evident linear relationship with time and other yield-related factors andhistorical data in the study area is abundant, multiple linear regression model ispreferentially adopted to predict the grain yield; if the grain yield and the yield-relatedfactors has little historical data or their historical data are insufficient, exponentialsmoothing model is preferred to predict the grain yield; if the grain yield has a complexnon-linear relationship with yield-related factors and historical data is more fully, andhistorical data of the related factors can be collected, BP neural network regressionanalysis model should be considered to predict the grain yield; with absence of historicaldata of the yield-related factors, BP neural network time series model can be used topredict the grain yield;(2) the grain yield prediction model is established based on the results of repeated testing and various comparisons;(3) the accuracy of the grain yieldprediction can be improved by establishing multiple-coupled prediction model.