Prediction Model Of Cucumber Downy Mildew And Powdery Mildew In Solar Greenhouse Based On Bayesian Network

【Object】Solar greenhouse is one of the main facilities for protected vegetable production in China,which provides a good environmental control basis for vegetable cultivation,so as to realize the periodic,off-season,all-weather large-scale production and planting.However,the proper temperature and humidity inside the greenhouse lead to the occurrence of downy mildew,powdery mildew and other leaf diseases,which often result in serious yield reduction,or even total crop failure.In order to accurately predict the occurrence of greenhouse cucumber disease in autumn,Bayesian network was used to establish cucumber disease prediction model in solar greenhouse,which could provide reference for cucumber disease control in actual production.【Methods】Fruit cucumbers were used as the experimental material in this study to conduct experiments from the Sept.to Nov.in 2020.Four greenhouses were chosen as the experimental greenhouse,which were Xiaotangshan National precision agriculture experimental base(No.5)in Changping District,Hongke farm(No.6)in Fangshan District,Shounong Manor(No.9)in Haidian District and Yunong company(No.7)in Daxing District,Beijing.According to the chessboard pattern,the temperature and humidity light sensors were deployed on 9 sampling points and investigation was conducted on cucumber downy mildew and powdery mildew in greenhouses.The environmental data were collected every 1 h.After transplanting,the investigation was carried out every day until the early symptoms of the disease appeared,and the date of the first disease was recorded.After that,the survey was conducted at fixed point,and 12 plants were selected at each point.The cycle was changed to 3～4 days for 1 time,and the incidence rate was statistically analyzed.The severity of disease was classified and recorded according to GB/T 17980.26--2000.Combining the threshold of environmental parameters determined by literatures with the multi-point survey data,the Bayesian network analysis model was established to obtain the probability of disease occurrence and the conditional probabilities of various factors.The Bayesian network model was established to predict whether cucumber downy mildew and powdery mildew by forming a probability structure chart,and to compare with the observation of the disease in the greenhouse.【Results】(1)Bayesian network was used to predict the occurrence of cucumber downy mildew in four greenhouse bases.The results show that the accuracy of the model weas respectively 0.92,0.91,0.94 and0.84.The mean square error(MSE)were 0.08,0.09,0.09 and 0.16,and the root mean square error(RSME)were 0.28,0.30,0.24 and 0.40,respectively.The prediction was consistent with the actual occurrence,and when the prediction probability was greater than 0.82,the occurrence of disease could be judged,which indicated that the model had certain universality.This model could provide decision-making reference for guiding the control and management of cucumber downy mildew.(2)The results showed that the ACC(accuracy)of the model to predict the cucumber powdery mildew in four greenhouse bases were 0.95,0.92,0.91,0.87.respectively.The prediction was consistent with the actual occurrence The Youden index J were 0.90,0.86,0.84,0.70 and 0.74,which could provide reference for the prediction of cucumber powdery mildew in actual production.(3)A cucumber disease prediction program based on Bayesian network algorithm was developed,which can run independently.【Conclusion】The accuracy of the Bayesian network model in four greenhouses was 0.92,0.91,0.94,0.84 for downy mildew and 0.95,0.92,0.91,0.87 for powdery mildew,which indicated that the model had good prediction effect and could provide reference for cucumber disease control in actual production.