Research On Under-sample Space Of Cut Tobacco Dryer Process Parameters Based On Generative Adversarial Network

As an important machine in tobacco production,the selection and optimization of parameters of the cut tobacco dryer have a direct impact on the product quality and production efficiency.Because the under-sample space of drying process parameters does not contain all characteristic sample points,the prediction accuracy of the under-sample space area data is low,which affects the research of process parameters.However,there is no effective method to deal with the under-sample space of drying process parameters.In order to improve the prediction accuracy of sample data in under-sample space,the under-sample space of process parameters of cut tobacco dryer was taken as the research object in this paper,a processing method of process parameters under-sample space based on generative countermeasure network was proposed,and a data generation model was constructed.The specific research contents and main conclusions were as follows:(1)The working principle of the cut tobacco dryer and the drying process of the leaf silk were analyzed,and the characteristics of the process parameters were studied according to the characteristics of the data acquisition mode.The area of the under-sample space and its influence on the accuracy of data prediction were obtained,and the necessity of processing the under-sample space was put forward.(2)Aiming at the limitation of the conventional under-sample space processing method in processing the data of the under-sample space of the cut tobacco dryer parameters,a processing method of under-sample space based on the generative countermeasure network was proposed.According to Pearson correlation analysis method,the interpolation objects of the parameters were obtained,and the rules of the under-sample space interpolation of the cut tobacco dryer parameters were formulated,and the sample number of the under-sample space is determined,the interpolation threshold and the number of interpolation data groups were determined.(3)Aiming at the problem of generating under-sample space interpolation,a data generation model of under-sample space was constructed.The network structure and parameters of discriminator and generator were designed,the data generation model was optimized from loss function and model connection mode,and a model with higher accuracy and learning efficiency was obtained.(4)The reliability of the sample generation model was verified by the significance test method.The data prediction model based on BP neural network and the multiple regression model were compared to verify the accuracy of data prediction and the prediction accuracy of under-sample data was improved by using the method and the model.