Abstract: To achieve single data source and strong coupling of variables in the control of tobacco drying and reduce unstable moisture content in cut tobacco output from a cylinder dryer, intelligent sensors were installed for smart sensing of ambient temperature, relative humidity and steam dryness. By combining with the multi-parameter collaborative correlation analysis, a dynamic prediction model based on random forest incremental learning was established. The multi-level control strategy for moisture control in cut tobacco output from cylinder drying was developed to realize the multi-parameter characteristic collaborative optimization control of moisture content. Tests were conducted with blended tobacco material for"Furongwang(Hard)"brand cigarette in production line A, and the results showed that: 1)Ambient temperature, relative humidity and steam dryness significantly affected the control process of cut tobacco drying. 2)Compared with production line B using the traditional PID control mode, the average process control capability index for moisture content in the cut tobacco increased by 45%, the average standard deviation decreased by 52%, and the average unsteady time decreased by 15% in production line A. In addition, the control data from several batches in line A was validated with higher consistency. This method effectively improves the moisture control capability in cut tobacco output from the cylinder dryer and ensures the consistency of the cut tobacco quality.