Short-term Prediction Of Grape Downy Mildew Based On Grey Correlation Analysis And Optimized SVM

Grape downy mildew is one of the most serious diseases that endanger grapes,which can easily cause the ears to fail to grow normally,lead to the reduction of production and cause serious economic losses.Rapid and accurate prediction is an important means to effectively prevent and control the occurrence and development of grape downy mildew,which is of great significance to ensure high yield,safety and high quality production of grape.At present,there are few studies combining the prediction of the date of the first infection of grape downy mildew and the prediction of the disease degree from the initial infection.The strong generalization ability and the advantage of nonlinear fitting of SVM play a positive role in the prediction of grape downy mildew,but the selection of training learning parameters greatly affects the prediction performance of SVM model,and the existing research on the selection of prediction factors of grape downy mildew is still very scarce.According to the current research situation,this article takes C hardonnay from Bolongbao Winery in Fangshan,Beijing as the research object.Firstly,the date of the first infection of Grape Downy Mildew in this area is determined,and uses grey correlation analysis(GRA,Grey Relational Analysis)to select the meteorological factors that are most similar to the change of grape downy mildew as SVM model feature input vector.The short-term prediction model of grape downy mildew was constructed by searching for the optimal parameters of SVM through parameter optimization algorithm,and the prediction system of grape downy mildew was designed on this basis to achieve a more accurate and effective prediction of the incidence of grape downy mildew.The main researches in this paper are as follows:Firstly,to predict the date of the first infestation of grape downy mildew,choose to perform localized application operations on the typical agricultural meteorological model DMC AST,and adjust its predictive factors and initial values of the parameters on the basis of retaining its core algorithm to form the DMCAST-BJ model of the studied area.The localized DMC AST-BJ model was used to predict the initial infection date of grape downy mildew in Bolongbao winery in Fangshan,Beijing in 2012.The initial infection date predicted by the model is June 21,which is 3 days earlier than the actual infection date.Secondly,aiming at the problem of selecting forecasting factors in the prediction stage of grape downy mildew incidence,the grey correlation degree of 8 meteorological factors was calculated by grey correlation analysis method,and the average relative humidity,the lowest relative humidity,the accumulated rainfall and the accumulated rainfall day were finally determined as the main meteorological factors,so as to build GRA-SVM model to realize the prediction of the disease grade of Grape Downy Mildew in the next day.In order to verify the validity of the prediction factors determined by the grey correlation analysis,the SVM prediction model of grape downy mildew based on all factors was constructed.The prediction accuracy and running time of SVM model based on different prediction factors were compared and analyzed.The experimental results showed that the SVM model based on the gray correlation analysis has better prediction effect.Thirdly,in view of the selection of kernel function and the influence of other machine learning methods on the prediction performance of SVM model,this study uses four kernel functions and BP network methods to establish the SVM prediction model of grape downy mildew based on grey correlation analysis and the BP prediction model of grape downy mildew based on grey correlation analysis respectively.The SVM model with radial basis function as kernel function has better prediction effect after comparison.Under the same conditions,the prediction stability of SVM model is higher.The experimental results show that GRA-SVM model with radial basis function as kernel function can better predict the incidence of grape downy mildew.Fourth,aiming at the problem that traditional methods are easy to fall into local optimum while finding SVM parameters,this study uses particle swarm optimization(PSO,Particle Swarm Optimization),genetic algorithm(GA,Genetic Algorithm)and improved grid search method(GS,Grid Search)to optimize two parameters of SVM model and a prediction model of grape downy mildew based on grey correlation analysis and optimized SVM is constructed.And the confusion matrix method is used to analyze the three models from three angles of accuracy,precision and recall.The prediction results show that particle swarm optimization and genetic algorithm are better,the prediction accuracy is 95.24%,and the SVM model optimized by the two algorithms is biased in the accuracy rate of level 2 and recall rate of level 1,but the prediction accuracy rate and recall rate of other levels are all 100%.Finally,based on the incidence prediction model based on GRA-PSO-SVM,a prediction system of grape downy mildew based on Java Web was designed.It can complete the management of the prediction data and can realize the prediction and early warning of the incidence of grape downy mildew.The relevant personnel can timely and effectively control the grape downy mildew according to the warning information.This paper mainly studies the prediction of the initial infect ion date of grape downy mildew,the construction of the short-term prediction model of the disease degree of grape downy mildew and the design of the prediction system.It discusses the selection of the prediction factors of grape downy mildew and the optimization of the parameters of support vector machine.The established short-term prediction model based on grey correlation analysis and optimized SVM provides a new research idea for the prediction of grape downy mildew.