Rolling force model is the core of all the mathematical models of plate for rolling process, but the accuracy of traditional rolling force model is not high enough in application, so in this study the rolling force model of plate is researched and improved. The effects of different physical conditions on resistance of deformation are decoupled, and the formula acquired is practical. While the composition, Nb is used to calculate residual strain. At the same time, the self-learning method, which is based on the thickness layer is applied. The on-line application results show that the predictive error between force model calculated and measured can be controlled at less than 9% and 80% of the passes can be controlled within 5%.