| Peanut pests have a very serious impact on peanut growth,which will bring great economic losses to China ’ s peanut industry every year.There are many factors affecting the occurrence of pests.Peanut growers can not accurately determine the severity of pests,and can not correctly predict the occurrence trend of pests and thus can not make effective countermeasures.In this context,in order to solve the helpless dilemma of current growers in the face of pests,the peanut experimental field of Baixin Peanut Planting Cooperative in Luan County,Tangshan City,Hebei Province was taken as the object to study the prediction method of peanut pest occurrence grade.Firstly,pest trapping devices and meteorological sensors were placed in the field,and the main pest data and field meteorological data were collected and counted with the assistance of professional plant protection personnel.After data preprocessing,the pest data were clustered using K-means clustering algorithm and the current pest severity was marked.Seven types of meteorological data,such as air temperature,air relative humidity and precipitation,were collected by the meteorological sensor.After preprocessing,the principal component analysis was used to find the principal components that can describe the original meteorological data with a smaller number,and the pest data and meteorological data were combined to construct the data set.Secondly,in Python environment,four prediction models based on machine learning algorithm are established by calling the correlation function in Scikit-Learn library,namely,random forest,decision tree,RBF-SVM and Linear-SVM.In the optimization process of model parameters,grid search method and fivefold cross validation method are used to search the optimal parameter combination of each model.Different models are tested on the test set,and the performance of the model is compared from the accuracy,precision,recall and F1 value of the model.The experiment shows that the optimal parameter combination of the four models by grid search method can greatly improve the prediction accuracy and F1 score of the model on the test set.Principal component analysis can improve the performance of RBF-SVM model and Linear-SVM model,but reduce the performance of decision tree and random forest model.Through comprehensive analysis,it is determined that the optimal model is a random forest model without PCA training of original data.Its prediction accuracy on the test set reaches 96.4 %,and the F1 score is 0.95,which has practical application potential.Finally,based on the random forest model,a We Chat small program,namely peanut pest prediction system,was designed and developed,which realized the farmers desire to obtain the prediction information of pest occurrence level conveniently and quickly through mobile phones.It meets the needs of cooperative managers for peanut pest information management.The peanut pest prediction model based on machine learning provides a scientific and efficient method for pest prediction.The development of small programs meets the needs of growers and managers of cooperatives,and improves the peanut production efficiency of cooperatives and the information management efficiency of pest data. |
