In order to increase the precision of flatness control, considering the principle and the measured data of rolling process essence, the theory-intelligent dynamic matrix model of flatness control is established by using theory and intelligent methods synthetically. The network model for rapidly calculating the theory effective matrix is established by the BP network optimized by the particle swarm algorithm. The network model for rapidly calculating the measurement effective matrix is established by the RBF network optimized by the cluster algorithm. The flatness control model can track the practical situation of rolling process by on-line self-learning. The scheme for flatness control quantity calculation is established by combining the theory control matrix and the measurement control matrix. The simulation result indicates that the establishment of theory-intelligent dynamic matrix model of flatness control with stable control process and high precision supplies a new way and method for studying flatness on-line control model.