Study On Prediction Models Of Three Main Diseases Of Cucumber In Solar Greenhouses

As one of the necessary facilities in the facility agriculture industry in northern China,solar greenhouses occupy a high economic position in vegetable production.However,greenhouses’unique climate and environment can easily lead to severe diseases and pests.The current commonly used chemical control methods may lead to drug residues,endanger food quality and safety,etc.Therefore,it is essential to accurately and effectively predict diseases and provide management and control decisions for farmers to take measures.The cucumber’Lyujingling No.2’was used as the experiment material in this study.From March to December 2021,the wireless network environment monitoring system was placed in the No.16greenhouse of Beijing Xiaotangshan National Experiment Station for Precision Agriculture in Changping District,Beijing,the C9 greenhouse of Shounong Manor in Haidian District,and the D6 greenhouse of Hongke Farm in Fangshan District.According to"The standard for investigation on forecast of Cucumber downy mildew in protected cultivation"(DB 11/T 286-2005),"Pesticide-Guidelines for the field efficacy trials(I)—Fungicides against cucumber powdery mildew"(GB/T 17980.30-2000),"Technical rules of diagnosis and control for grey mold in cucumber"(DB 22/T 2218-2014),a fixed-point and plant-based investigation was carried out on the three diseases by using the diagonal five-point sampling method.Based on the environmental data in the greenhouse acquired by sensors and the disease occurrence data adopted by manual investigation,through data preprocessing and variable selection,eight prediction models(Binary Logistic Regression,SVM,Decision Tree,KNN,ANN,RNN,GRU,LSTM)were developed for downy mildew,powdery mildew,and botrytis respectively.The generalization and universality of these models were evaluated.The results showed that the LSTM model had the best results in developing and verifying the downy mildew occurrence prediction model and developing the powdery mildew occurrence prediction model and the botrytis occurrence prediction model.Among them,the accuracy of the prediction model for downy mildew occurrence was 90%,the AUC value was 0.9015,and the Kappa coefficient was 0.80;the model was verified by the test data in the second half of 2021,and the accuracy was 92%,the AUC value was0.9225,and the Kappa coefficient was 0.85;the accuracy of the prediction model for the occurrence of powdery mildew was 96%,the AUC value was 0.9327,and the Kappa coefficient was 0.88;the accuracy of the prediction model for the occurrence of botrytis was 97%,the AUC value was 0.9532,and the Kappa coefficient was 0.92.It showed that the LSTM model had better all-around performance and higher classification prediction accuracy It can be used to develop a general prediction model for three main cucumber diseases in solar greenhouses.The prediction results of the four environmental factors in the greenhouse were better by the LSTM model.Among them,the prediction results of soil temperature variables were the best,which were R~2=0.9982,RMSE=0.08℃,and MAE=0.05℃,which can be used to predict future environmental factors in the greenhouse and provide a scientific basis for future disease prediction research.In summary,the results of the LSTM model in the prediction of the three main diseases of cucumbers in the solar greenhouse are higher than those of the other seven models and have better classification and prediction accuracy,which can provide technical support for the prediction of the main diseases of cucumbers in the solar greenhouse in actual production,and provide technical support for the prediction of the main diseases of cucumbers in the solar greenhouse.To further build a prediction model for the occurrence of future diseases to lay a foundation,to provide technical support and decision support for farmers to manage and prevent diseases before they occur,and minimize the losses of diseases on cucumber production and economy.