Abstract: In order to get rid of time lag and promote control precision in strip casing, a compound control system was established based on ARX feedforward model and self-adaptive neuron PID, wherein the flow rate of casing was predicted by the ARX model. On the basis of the predicted flow rate and feedback error, the frequency of casing pump motor was determined and used to control casing volume; the control coefficient was automatically corrected by the self-adaptive neuron PID to adapt to the variation of working conditions. The performance of the system was tested on the tobacco blend of cigarette brand "Diamond (Hehua)" in Shijiazhuang Cigarette Factory. The results showed that this method significantly reduced the over regulation of head tobacco by 11.2 percentage points and effectively promoted the casing precision in a whole batch, the response time of the system reduced by 3 s, and the accumulated casing precision increased by 0.25 percentage points. The proposed method provides technical supports for improving casing control system in tobacco primary processing.