Neural Network Model Predictive Control Design for Nonlinear Processes with Unmeasured Disturbances

• Main Authors: Chih-Wei Wang, 王志偉
• Other Authors: Jung-Hui Chen
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/91691560839672871346

碩士 === 中原大學 === 化學工程研究所 === 89 === Unmeasured disturbances usually plague the process and result in the defect products in chemical plants; hence, the identification and control of the process with the presence of disturbances is important. This paper completely develops the neural network model predictive control (NNMPC) from the model design to the controller design for nonlinear processes with unmeasured disturbances. In the model design, an input-driven output neural network ARX model (NNARX) combining with a disturbance AR model, called NNARX+AR, is proposed. NNARX and AR represent the input-output characteristics without the corrupted disturbances and with the disturbances respectively. The Levenberg-Marquardt algorithm for NNARX and the least square algorithm for AR are synchronously used to train the process model. In the control design, a constrained NNMPC based on NNARX+AR via the successive quadratic programming is developed to search the optimal control actions. To demonstrate the proposed identification and predictive control strategies, two multivariable cases with the presence of unmeasured disturbances, including a nonlinear mathematical equations and a pH neutralization system, are presented.