|
THE RBF-ARX MODEL BASED MODELING AND PREDICTIVE CONTROL FOR A CLASS OF NONLINEAR PROCESSES
H. Peng*, T. Ozaki**, V. Haggan-Ozaki†, Y. Toyoda††
* College of Information Engineering, Central South University, Changsha, 410083,
China. Currently a visiting researcher at the Institute of Statistical Mathematics,
4-6-7 Minami Azabu, Minato-ku, Tokyo 106-8569, Japan. E-mail: peng@ism.ac.jp
** The Institute of Statistical Mathematics, 4-6-7 Minami Azabu, Minato-ku, Tokyo
106-8569, Japan. E-mail: ozaki@ism.ac.jp
† Sophia University, 4, Yonbancho, Chiyodaku, Tokyo 102-0081, Japan. E-mail:
v_ozaki@sophia.ac.jp
†† Bailey Japan Co. Ltd., 511 Baraki, Nirayama-cho, Tagata-gun, Shizuoka 410-2193,
Japan. E-mail: toyoda@bailey.co.jp
This paper considers modeling and control problems of the non-stationary nonlinear processes whose dynamics depends on the working point. A hybrid RBF-ARX model-based predictive control (MPC) strategy without resorting to on-line parameter estimation for this kind of processes is presented. The RBF-ARX model is composed of the RBF networks and a rather general form of ARX model, which is identified off-line, and whose local linearization may be easily obtained. A quickly-convergent estimation method is applied to optimize the RBF-ARX model parameters. The modeling validity and the MPC performance is illustrated by an application to Nitrogen Oxide (NOx) decomposition process in thermal power plants.
Keywords: Nonlinear systems, non-stationary process; modeling, model based predictive control, radial basis function networks, ARX model
Session slot T-Th-M21: Posters of Design Methods and Optimal Control/Area code 2d : Optimal Control
 |
