| In order to meet the high demands of the cold rolling system, improving the flatness quality increasingly has become one of the most important questions that are urgent to be solved. Therefore, flatness control has become the key to the steel system. Because flatness control is many-variable, non-linear, high combined and time-delay, it is difficult to control flatness by use of conventional control methods. Recently the artificial intelligence theory was widely used in the field of flatness control system. The technology of control based on rough set theory has received researchers'attention more and more. The neural network and rough sets are unified can display respectively superiority effective and make up its insufficiency. Based on the theory of artificial intelligence a study has been made on approaches to flatness control.Above all, the method of neural network controller and disadvantage of complex network structures has been analyzed. The CMAC (Cerebellar Model Articulation Controller) neural network controller is set up based on the characteristics, such as simple of network structure, quick study, generalization, ability to study multi-dimensional nonlinear mapping.Secondly, the localization of controller was analyzed by traditional network which has fixed learning rate. In this paper, the dynamic learning rate is imported in the CMAC control model. By this method the precision and speed were proved, so the purpose of flatness control online is achieved.Thirdly, the number of hidden nodes of and the initial weights of CMAC network are randomly selected, so the structure of network is not the best. Use rough set to solve to determine the structure of network system.Finally, the flatness control model has been set up based on the CMAC neural network. The simulation results on 1220 cold rolling mill have proved its good correctness and feasibilities. |
