Study On Environmental Factor And Yields Prediction Of Maize Based On Machine Learning

China is a large agricultural country.Maize is one of the main crops in China.How to increase crop yield in agricultural production is an extremely important research topic and its growth environmental factors are related closely to its yields.Therefore,this dissertation aims to study on change rules and prediction of the growth environmental factors of maize,as well as its yield prediction based on data set obtained in this study.Firstly,this dissertation preprocesses the environmental factors data of the maize,and analyzes the correlation between the maize growth environmental factors and its phenotypic parameters,in order to extract the growth environmental factors,which have a greater correlation with the phenotypic parameters.Then,using the fitting method,we find the change rules of various environmental factors of maize.At the same time,for the maize growth environmental factors?air humidity,CO₂,soil salinity?with high frequency components,this study proposes a combined prediction model,named as WPD-CEEMD-GA-Elman where the Wavelet Packet Decomposition and the Complete Ensemble Empirical Mode Decomposition algorithm are used to decompose the environmental factor signals,and then the Elman neural network combined GA?named as GA-Elman?are used to predict them;for the maize growth environment factors?air temperature,soil temperature,soil humidity?with low frequency components,this study proposes another combination model,named as WPD-GA-Elman where we use Wavelet Packet Decomposition to decompose these factor signals,and then use GA-Elman to predict them.Experimental results show that the combination model has obtained the higher prediction effect than the single model on the growth environment factor data set of the maize obtained in this study.Secondly,in order to predict maize yield,this study uses the data of the past 40years from the statistical yearbook of Inner Mongolia.In the experiment,considering the change rules of maize growth environmental factors,based on existing researches,we take the monthly average temperature,the highest temperature,the lowest temperature,the average relative humidity of the air,the total rainfall,the average soil temperature and the average soil relative humidity as the input of the prediction models,and the yield per unit area of maize as the output of them.Experimental results show that the combination model of GA-Elman has higher prediction accuracy than the single Elman model,it improved by 59.38%compared with Elman neural network.Finally,the correlation analysis is carried out between the maize phenotypic parameters over data set obtained in this study.Experimental results show that there is a high correlation between the fresh fruit weight of maize and its fresh fruit perimeter,fresh fruit length,plant height in turn.Then the stepwise regression is utilized to obtain the regression equation of the fresh fruit and its other phenotypic parameters in the experiment,which lays a foundation for real-time monitoring the yield of maize.