Research On Wheat Yield Prediction Method Of Henan Province Based On Machine Learning And Intelligent Optimization Algorithm

Agriculture is the basic industry of Henan Province,wheat production is an important part of the agricultural industry,plays an important role in maintaining food supply and security,and relates to the economic lifeline of Henan province.Accurate prediction of wheat yield can provide important information for agricultural economic regulation and policy making.In order to improve the accuracy of wheat yield prediction,the related factors affecting wheat yield were comprehensively considered.This paper collected the wheat yield data of Henan Province for 44 years from 1978 to 2021,and determined the data of 11 influencing factors affecting wheat in Henan province from three perspectives of grain production status,human activities and disaster conditions.Firstly,the correlation between influencing factors and yield was analyzed based on Pearson correlation coefficient and Spearman correlation coefficient,and then three machine learning prediction models were constructed to predict wheat yield in Henan province.Finally,two intelligent optimization algorithms were combined to optimize the three models,and finally the wheat yield prediction model was obtained.Based on BP neural network,Elman neural network and Support Vector Regression(SVR)model,the wheat yield in Henan province was predicted and analyzed.The minimum Mean Absolute Error(MAE)of BP neural network model is35.93 when the number of hidden layer nodes is 6.When the number of hidden layer nodes of Elman neural network model was 14,the minimum Mean Absolute Error(MAE)of the model was 73.83.The 5-fold cross validation method was used to find the best penalty factor(64)and kernel parameter(0.0156),so as to train the SVR model with radial basis kernel function to minimize the prediction error.Among the three models,BP neural network is better,Elman neural network is second,SVR model is worse.Particle Swarm Optimization(PSO)algorithm and Genetic optimization(GA)algorithm are used to optimize the initial weights and initial thresholds of BP and Elman neural networks respectively.The penalty factor c and kernel parameter g of SVR model were optimized,and finally six optimized models were obtained and compared with the original machine learning prediction model.Taking the error MAPE as an example,the MAPE of the BP model after particle swarm optimization is 46.4% lower than that of the BP neural network model.The MAPE of the Elman model after particle swarm optimization is 82.5% lower than that of the Elman neural network model.The MAPE of PSO-SVR model is 36.6% lower than that of SVR model.The BP model optimized by genetic algorithm is 12.4% lower than that of BP neural network model MAPE.The MAPE of Elman model optimized by genetic algorithm is 23.5% lower than that of Elman neural network model.The MAPE of GA-SVR model is 66.4% lower than that of SVR model.It can be seen that the prediction error of the optimized model is lower than that of the original machine learning prediction model.Among them,the optimization effect of particle swarm optimization algorithm on Elman model is stronger than that of BP and SVR model,and the optimization effect of genetic optimization algorithm on SVR model is stronger than that of BP and Elman model.Finally,PSO-Elman,PSO-BP,GA-SVR and GA-BP of the six combination optimization models were confirmed to be better in predicting wheat yield in Henan province,and the MAPE was 0.35%,0.52%,0.79% and 0.85%,respectively.The prediction effect of PSO-Elman model is better than other models.