Prediction Of Winter Wheat Yield In Henan Province Based On Deep Learning

Food production is vital to the national economy and people’s livelihood,and ensuring food security is an important national task.Grain yield prediction is one of the important means to ensure food security.Winter wheat is the main grain crop in north China.The output of winter wheat in Henan province accounts for one fourth of the total wheat production in China.And Henan province is the largest wheat production province in China.It is very important to forecast winter wheat yield in Henan Province.In the past,people used empirical formulas or simple statistical models to predict food production.However,in the face of complex natural environment and uncertain market,the forecast results can be further optimized.At present,deep learning technology develops rapidly and its application in data prediction is constantly expanding.However,there are still some problems to be solved,which need further study and improvement.For example,in the current mainstream winter wheat yield prediction methods,there are few methods to explore the relationship between characteristics and the intensity of influence of different characteristics on winter wheat prediction.In the actual prediction of winter wheat yield,the inconsistency between training and prediction occurs due to the absence of some data.The factors of preferential agricultural policies were not included in the data set,so the data set was deficient.In view of the above problems,this paper carries out the following research:(1)A winter wheat yield prediction model(MAKD-xDeepFM)based on x Deep FM network architecture with multi-head attention mechanism and knowledge distillation optimization was proposed for winter wheat yield prediction in Henan Province.The model uses x Deep FM algorithm to implement feature crossing and model generalization.The different effects of different characteristics on winter wheat yield were learned by increasing the multihead attention mechanism.Finally,by adding knowledge distillation technology,the trainingprediction inconsistency caused by feature missing was further solved.(2)The multi-feature fusion data set of wheat yield prediction in Henan province was constructed.The data of four provinces with similar production environment factors as Henan province were included in the data set,which solved the problem of small data volume of a single province’s data set.The provincial characteristics and the sparse characteristics of multidimensional preferential agricultural policies were included in the data set.The problem that data sets of traditional agricultural yield prediction method have numerical features only was solved.A fuzzy comprehensive evaluation method for preferential agricultural policies was proposed;the weights according to the intensity of the influence of the preferential agricultural policies on the yield were set;the consumer price index correction factor was introduced;the fuzzy data on the intensity of preferential agricultural policies was generated.Finally,a multifeature fusion data set containing multiple feature types and influencing factors such as planting area,meteorological data,provincial characteristics and preferential agricultural policies was constructed.(3)Conduct experimental studies to explore the validity and predictive accuracy of the proposed winter wheat yield prediction model(MAKD-x Deep FM)and the constructed data set.Firstly,the prediction results of the proposed MAKD-x Deep FM model were compared with the usual winter wheat yield prediction models.Secondly,through the comparative experiment,to explore whether the increase of preferential agricultural policies can significantly improve the prediction accuracy.Finally,by comparing the single province data set and the multi-province multi-feature fusion data set in Henan province,the validity of the multi-province multi-feature data fusion data set is analyzed.The results showed that using the multi-province and multi-feature fusion data set and winter wheat yield prediction model(MAKD-x Deep FM)could effectively predict the winter wheat yield in Henan province with high accuracy.The introduction of preferential agricultural policies and the sparse feature processing of multi-dimensional preferential agricultural policies have obvious advantages and strong practicability of data sets.Algorithm design and data set construction can provide reference for the research of grain yield prediction technology based on deep learning.It also has reference value for other data prediction.