Optimization Design Of RBF-ARX And Application Study To Flatness Control System

The mathematical model of controlled object and process is the very important basis in analysis and design of control system. Firstly, in order to control the object or process, their working mechanism and characteristics must be understood. Then, precise mathematical model should be established for quantitative analysis. Finally, desired effect will be got by controlling. The mathematical model of simple object or process is easy to establish, but for some complex systems or processes, their mathematical models are difficult to establish because of various reasons and their control schemes will not be able to design. In recent years, the rapid development of artificial intelligence theory and modeling method by data-driven have caused the attention of many scholars, they are applied into modeling of complex nonlinear system. In this paper, it takes optimal design of RBF-ARX(Radial Basis Function-AutoRegressive eXogenous) model and application in flatness control system as the research subject. On the basis of intelligent control theory, RBF-ARX modeling scheme optimized by Genetic Algorithm(GA) is proposed. Comparative study with the RBF-ARX model optimized by traditional structure nonlinear parameters optimal method(SNPOM), scheme proposed can implement the flatness defect pattern recognition and control very well.At first, the internal structure of RBF-ARX and SNPOM are got further study. Recursive least square method is introduced into SNPOM, in order to improve the shortage of SNPOM in process of parameters optimization, like complicated arithmetic and large amount of data. Meanwhile, in order to promote the application of RBF-ARX in the field of engineering, the thought of GA replaces SNPOM is generated. The scheme of system modeling and optimization by GA-RBF-ARX is formed, it greatly simplifies the optimization process for model parameters.Secondly, flatness defect pattern recognition model based on GA-RBF-ARX is established for a 900 HC reversible cold rolling mill. Simulation shows that recognition result of GA-RBF-ARX flatness defect pattern recognition model is better than that optimized by SNPOM. In addition, for verifying the modeling effect of RBF-ARX, discrete Hopfield network, which has the function of associative memory, is used to comparative study.Finally, predictive control strategy is introduced into GA-RBF-ARX model, and a complete flatness intelligent control system is established, also including flatness defect recognition model and flatness defect predictive model. Simulation shows that flatness defect predictive output of GA-RBF-ARX model can track the real output of mill flatness. The control accuracy of flatness defect can meet the requirements of flatness production, and this system is an effective method of modeling and control.