本論文處理非線性Hammerstein和Wiener離散系統之鑑別，前者非線性靜態方塊在前，線性動態方塊在後；後者線性動態方塊在前，非線性靜態方塊在後。根據輸出入訊號量測來鑑別此類型模式的最大困難在於無法測量介於動態和靜態部分之間的內部變數，因此Voros (1999; 2003) 提出以疊代方式估測其值，也就是先假設一組內部變數初值，再利用線性回歸方程式反覆計算直至所有參數和內部變數的最小平方估測收斂為止。我們發現Voros法的收斂性不佳，對於一些實際操作狀況，例如量測雜訊存在或模式結構不吻合，該法無法獲得滿意的結果，甚至無法收斂。

　　我們認為Voros法收斂性不佳的主因是它忽略了不同取樣時間資訊彼此間的關聯性，因此本論文引入一時間加權離散濾波器，將一指定時間區間內取樣訊號轉換成以產生一較具關聯性之線性回歸方程式，然後再利用移動區間最小平方法以疊代方式來估測模型參數及內部變數。模擬結果顯示，本法對於Hammerstein和Wiener離散系統的鑑別十分有效，能夠在雜訊存在及模式結構不吻合的情況下，保證疊代式最小平方估測具有良好的收斂性及正確性。Voros法對於前述情況相當敏感，即使使用加入有益變數的最小平方估測亦無法有效消除雜訊的影響。

　　This thesis deals with the identification of nonlinear Hammerstein and Wiener discrete systems. The former consists of a nonlinear static block followed by a linear dynamic block, whereas the latter is the opposite. The major difficulty of identification of those systems based on input-output measurements is that the internal variable between the static and the dynamic part is inaccessible to measurement. Hence, Voros (1999; 2003) proposed to identify its value in an iterative manner. That is, first assume an initial function for the internal variable, and then utilize the linear regression equation to calculate all the parameters and the internal variable repeatedly until their least-squares estimates converge. We found that the convergence of Voros’s method is poor. It cannot obtain satisfactory results or even diverges under certain practical conditions, such as measurement noise or model structure mismatch.

　　We think that the convergence of Voros’s method is poor because it ignores the connection between data sampled at different instants. Therefore, this thesis introduces a time-weighted discrete filter to transform the sampled signals over a specified time interval and produce a better connected linear regression equation. We then employ the moving-horizon least-squares algorithm to estimate all model parameters and the internal variable iteratively. Extensive simulation reveals that this method is very effective for the identification of Hammerstein and Wiener discrete systems. It often ensures good convergence and accuracy of the iterative least-squares estimation algorithm under measurement noise and model structure mismatch. Voros’s method, on the other hand, is quite sensitive to these conditions. The effect of noise cannot be removed even by incorporating the instrumental variable in the least-squares algorithm.

[1] Abonyi, J., R. Babuska, M. Ayala Botto, F. Szeifert and L. Nagy, “Identification and control of nonlinear systems using fuzzy Hammerstein model,” Ind. Eng. Chem. Res., 39, 4302 (2000).
[2] Anbumani, k., L. W. Patnaik and I. G. Sarma, “Self-tuning minimum-variance control of nonlinear systems of the Hammerstein model,” IEEE Trans. Automatic Control, 74, 1807 (2001).
[3] Clark, D. W., “Generalized least-squares estimation of parameters of a dynamic model,” IFAC Symposium-Identification in Automatic Control Systems, Paper 3.17, Prague (1967).
[4] Giri, F., F. Z. Chaoui and Y. Rochdi, “Parameter identification of a class of Hammerstein plants,” Automatica, 37, 749 (2001).
[5] Haber, R. and L. Keviczky, Nonlinear System Identification - Input-Output Modeling Approach, Kluwer Academic Publiahers, Dorchrecht/Boston/London, (1999).
[6] Hsia, T. C., “On multistage least-squares approach to system identification,” Proceedings, IFAC Sixth World Congress, Paper 18.2, Boston (1975).
[7] Hsia, T. C., System Identification: Least-Square Methods, Lexington Books, Lexington, Mass. (1977).
[8] Huang, H. P., M. W. Lee and M. S. Chiu., “Linear parameters varying model approach for nonlinear process control,” Proceedings of International Symposium on Advanced Control of Industial Processes, Kumamoto, Japan, Paper 353 (2002).
[9] Hwang, S. H. and M. L. Lin., “A method for identification of discrete parametric models with unknown orders and delays,” J. Chin. Inst. Chem. Engrs., 33, 373 (2002).
[10] Johansson, R., System Modeling and Identification, Prentice-Hall: Englewood Cliffs (1993).
[11] Kung, M. C. and B. F. Womack, “Stability analysis of a discrete-time adaptive control algorithm having a polynomial input,” IEEE Trans. Automatic Control, 28, 1110 (1983).
[12] Ljung, L., and T. Söderström, Theory and Practice of Recursive Identification, MIT Press: Cambridge (1983).
[13] Ljung, L., “On the estimation of transfer functions,” Automatica, 21, 677 (1985).
[14] Ljung, L., System Identification-Theory for The User, Prentice-Hall: New Jersey (1987).
[15] Lovera, M., T. Gustafsson and M. Verhaegen, “Recursive subspace identification of linear and non-linear Wiener state-space models,” Automatica, 36, 1639 (2000).
[16] Murray-Smith, R. and T. A. Johansen, Multiple Model Approaches to Modeling and Control, Taylor and Francis, Landon, (1997).
[17] Söderström, T., “On the generalized least-squares method. Counter-examples to general convergence,” Automatica, 10, 687 (1974).
[18] Söderström, T. and P. Stoica, “System identification,” Prentice-Hall Interational: UK (1989).
[19] Voros, J., “Iterative algorithm for parameter identification of Hammerstein systems with two-segment nonlinearities,” IEEE Trans. Automatic Control, 44, 2145 (1999).
[20] Voros, J., “Parameter identification of Wiener systems with discontinuous nonlinearities,” System and Control Letters, 44, 363 (2001).
[21] Voros, J., “Modeling and identification of Wiener systems with two-segment nonlinearities,” IEEE Trans. On Control Systems Technology, 11, 253 (2003).
[22] Wong, K. Y. and E. Polak, “Identification of linear discrete time systems using the instrument variable approach,” IEEE Transactions on Automatic Control, AC-12, 707 (1967).
[23] Zhang, J. X. and S. J. Lang, “Adaptive weighted suboptimal control for linear dynamic systems having a polynomial input,” IEEE Trans. Automatic Control, 44, 2145 (1999).