Research On Export Moisture Control Of Cut-tobacco Drier Based On Machine Learning Techniques

With the rapid development of the Internet and information technology,big data has played an important role in all walks of life.The tobacco industry is actively introducing big data technology to meet the needs of the times and national development,and has a certain foundation of intelligent manufacturing applications,but the process control in the cigarette production process still requires a lot of manual intervention work,which triggers more intelligent needs.The water content of the tobacco in the process of filament production is an important control index in the production of cigarettes,and the control of the moisture at the exit of the dryer is one of the key links,the parameters to be regulated are redundant,and the characteristics of the data such as multi-disturbance,non-linearity and high coupling greatly increase the difficulty of data processing and modeling,and the results of the initial experiments are not satisfactory.However,with the parallel emergence and continuous iteration of machine learning and other algorithms,new ideas have been opened for industrial control,and intelligent control requires digital and process-oriented solutions.Therefore,this research will be based on machine learning related methods to give the optimal combination of key process parameters and equipment parameters for different export moisture requirements,in order to achieve the goal of fine control of the export moisture of the dryer.In this thesis,we will use R and Matlab software to complete the research on the exit moisture control of the yarn dryer based on nearly 100 production batches with about 150,000 data from August to October 2021 of a cigarette specification in the tunnel-type temperature and humidity increasing process and leaf yarn drying stage.After data pre-processing,a part of the process parameters with the greatest influence on the exit moisture content in the research problem was extracted,and a Bayesian network model between the control variables and the moisture content of leaf yarn drying was constructed through structure learning and parameter learning.After that,the model was optimized in four stages from the production reality,and the data were divided into six categories according to the export moisture setting value,and the complex relationship between equipment parameters,process parameters and the moisture content of leaf yarn drying was re-modeled based on the total data set and the six categories respectively to complete the Bayesian network inference task.Finally,seven exit moisture control strategies for the yarn dryer were developed in a targeted manner and the control effects were evaluated in the test set.The probability of the exit moisture content falling into the optimal interval during the drying phase of the leaf yarn increased by 0.2612 on average,and the results met the expectations.This study makes full use of the new production factor of data,and to a certain extent,lays a theoretical and application basis for correcting the current distribution of moisture content in leaf drying,and provides new ideas for the intelligent control of moisture content in the key processes of cigarette enterprises in the filament making stage.