Study On Forecasting Model Of Wheat Yield Based On Long-term Sequence Meteorological Data

In recent years,extreme weather events have occurred frequently,especially meteorological disasters such as drought and freezing damage have extremely serious impacts on crop yields.Therefore,actively carrying out research on crop yield meteorological forecasts is of great significance to the management,decision-making,macro-control of crop production and the response to food security issues.In practical applications,different forecasting methods and theories have different results of crop yield prediction in different regions and varieties.It is particularly important to screen out the methods and models suitable for local crop yield prediction.In this study,the wheat yield in Yuncheng City was taken as the research object,and the wheat yield prediction modle was studied using the meteorological data from 2000 to 2019.The trend of wheat yield per unit area in Yuncheng was analyzed,the trend yield and meteorological yield were separated.The random forest algorithm was used to screen the key meteorological factors affecting the yield,and different feature subsets were established.The wheat yield prediction models based on linear regression(LR),decision tree(DT),random forest(RF),e Xtreme Gradient Boosting(XGBoost)and LightGradient Boosting Machine(Lightg BM)were established,and the prediction accuracy of different models was compared.The results were as follows:1.In the past 20 years,the wheat yield in Yuncheng area changed greatly from year to year,and there were more disaster years.The wheat yield fluctuated greatly within a certain range from 2000 to 2010,showed a significant upward trend from 2010,and showed a significant downward trend in 2019.HP filter method,exponential smoothing method and moving average method were used to separate the trend yield and meteorological yield of wheat per unit area.The results show that the meteorological output separated by the HP filtering method and the actual output have the characteristics of synergistic changes,and the meteorological output separated based on this method can accurately capture the output changes brought about by the climatic conditions of the typical rich and poor years.2.Using the actual yield as the target variable,the random forest algorithm was used to screen out the feature set with the feature number of 40.(RY-RF feature set),and the meteorological yield was the target variable,and the random forest algorithm was used to select the feature set with the feature number of30.(MY-RF feature set)and through Pearson correlation analysis,the random forest algorithm was used to screen out the feature set with the number of features of 50.(MY-Pearson feature set).3.Three feature subsets were used as input data sets,and five regression prediction algorithms(LR,SVR,RF,XGBoost and LightGBM)were used to establish the wheat yield prediction model.Through comparative analysis,when RY-RF feature set was used as model input,the prediction model based on each algorithm had the highest accuracy and was the optimal feature subset,among them,the prediction model based on LightTGBM algorithm had the highest accuracy,the R~2 of the training set was 0.93 and the RMSE was 364.58,the R~2 of the test set was 0.84,the RMSE was 572.19.The predicted result was closest to the real value.The model can find the characteristic factors closely related to the wheat yield in Yuncheng city,and provide reference for the accurate prediction of the wheat yield in this area,which has a certain practical value.