Research On State Prediction Method Of Tobacco Curing Process Based On Model Fusion

China’s tobacco plantation industry has a large scale and many employees.However,the current tobacco curing control technology not only has high labor intensity but also cannot be adjusted in time according to the conditions of different batches of tobacco leaves,which will reduce the quality of tobacco.Modeling the state prediction of the tobacco curing process,accurately predicting the state of the tobacco curing,and making timely adjustments to the curing process can improve the quality of the tobacco after curing and reduce labor intensity.The area,color,weight,and some chemical substances of tobacco leaves change significantly during the tobacco curing process,which can theoretically be used as the input feature of the state prediction model.However,due to the complexity of the intensive curing room environment,it is difficult to calculate the changes of the area and the changes of chemical substances in real time.Only features such as color and weight are easier to extract,which means that the model has fewer available features,and the accuracy of prediction using a single model is relatively low.Aiming at this problem,a state prediction fusion model(SPFM)combining Long-Short Term Memory(LSTM)and Extreme Gradient Boosting(XGBoost)is proposed.A modeling study was conducted on tobacco curing data.The structure of SPFM was designed and analyzed in detail.And SPFM was compared with the traditional single model and the base models of SPFM.At the same time,according to the characteristics of the data set,a method for processing the tobacco curing data set is proposed: the image is denoised,the characteristic values of the RGB(Red,Green,Blue),HSV(Hue,Saturation,Value)color space are extracted.Then data pre-processing such as data cleaning,standardization and label digitization are performed on the data.Based on the real data collected by a tobacco station in 2019,a comparative test was conducted.The results show that compared with support vector machines,artificial neural networks and the base models of SPFM: XGBoost and LSTM,SPFM has better performance.The accuracy of SPFM is 0.974,an increase of 4.8%-59.7%;the macro precision of SPFM is 0.952,an increase of 8.2%-49.9%;the macro recall of SPFM is 0.936,an increase of 8.2%-75.0%;the macro F1-score of SPFM is 0.943,an increase of 9.7%-108.6%.The performance of SPFM is excellent.In addition,an intelligent tobacco curing platform integrating data collection,online monitoring,data mining and status prediction is designed,and SPFM is embedded in the platform.