Use of Discrete Laguerre and Kautz Expansions for Identification of Nonlinear Processes

• Main Authors: Chu -YuHsieh, 謝楚御
• Other Authors: Shyh-Hong Hwang
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/97957488578321145022

碩士 === 國立成功大學 === 化學工程學系碩博士班 === 98 === To increase production and reduce cost, well-designed control systems play an important role in chemical industry, subjected to stricter environmental constraints. On the other hand, a good identification method is crucial to controller design. However, complex nonlinear dynamics in chemical processes can cause considerable difficulties in identification, and such a nonlinear behavior cannot be ignored because of its prevalence. This thesis applies the Laguerre and Kautz descrete expansions to identify good Hammerstein and Wiener models, suited to controller design, for a wide variety of nonlinear processes. The identification method based on the Hammerstein model describes the nonlinear static part by a polynomial, uses the Laguerre ARX (AutoRegressive with eXogenous input) model to represent the linear dynamic part, and successfully update the estimate of the internal variable by an iterative algorithm. Unlike conventional linear identification methods, the proposed method can not only deal with the entire nonlinear process characteristics, but also possesses excellent convergence and accuracy. Simulation results on several practical processes demonstrate that the proposed method performs well for a wide range of nonlinear dynamics and test conditions. The resultant Hammerstein model can be easily adapted to linear control techniques. This thesis also proposes a non-iterative identification method based on a Wiener model. The method describes the linear dynamics by Laguerre FIR (Finit Impulse Response) or Kautz FIR models and approximates the static nonlinearity by an inverse polynomial function. As a result, the unknown internal variable does not appear in the regression equation, thereby allowing non-iterative parameter estimation. This resolves the convergence problem often encountered in available iterative methods for identifying Wiener models. Moreover, the influence of test experimental designs and measurement noise can be reduced by adjusting the reference point of the inverse polynomial. The identification method is effective for a wide range of process nonlinearities and test conditions, and is rather useful for nonlinear controller design.