非對稱歸屬函數之第二型模糊類神經網路分析及其應用

Type-2 Fuzzy Neural Network with Asymmetric Membership Functions Analysis and Its Applications

潘泓毅(Hung-Yi Pan)

指導教授 ： 李慶鴻

元智大學/工程學院/電機工程學系/碩士(2006年)

https://doi.org/10.6838/YZU.2006.00007

摘要

本文提出一個具非對稱歸屬函數(或可稱為Footprint Of Uncetainty, FOU)之第二型模糊類神經網路，並應用至非線性系統鑑別與控制。根據過去文獻提出的網路架構，本文以非對稱FOU取代常用的高斯對稱FOU。有別於高斯對稱FOU，非對稱FOU由四個高斯函數建構，可描述歸屬函數中心不明確(uncertain mean)與寬度不明確(uncertain variance)之特性，藉由非對稱之參數更新法則，可增加參數調整的效率與網路的近似能力。由系統鑑別的模擬結果可知，給定同樣的誤差需求，具非對稱FOU網路所使用的規則數與調整參數個數，皆少於對稱FOU網路。在非線性系統控制的應用上，對於可控典型式之受控系統，採用補償控制架構，配合Lyapunov穩定定理設計參數更新法則，模擬結果再次驗證使用非對稱FOU之優點；而對於非線性串接系統(nonlinear cascade system)，以backstepping之概念將系統視為兩個子系統，依序設計虛擬控制器(virtual controller)與實際控制器使原系統穩定，稱為解耦合控制架構，由非線性TORA系統與球桿系統模擬結果可證明其可行性與性能。

關鍵字

第二型模糊類神經網路 ； 非對稱歸屬函數 ； 補償控制 ； 解耦合控制 ； 非線性串接系統

並列摘要

This thesis proposes a type-2 fuzzy neural network (type-2 FNN) with asymmetric membership functions (also called Footprint Of Uncertainty, FOU), and applies it to identification and control of nonlinear systems. Based on the previous type-2 FNN structure, the mostly used Gaussian symmetric FOUs are replaced by asymmetric ones. An asymmetric FOU consists of four Gaussian functions. It describes the properties of both uncertain mean (center) and uncertain variance (width) in membership functions. Based on the Lyapunov approach, the corresponding adaption laws are derived to increase the efficiency of tuning parameters, and the approximation ability. The type-2 FNN with asymmetric FOU has small network structure (fewer rules and tuning parameters than the one with symmetric FOU) from the simulation results of system identification. For nonlinear system control, a type-2 FNN controller and compensated control scheme are used to treat the control problem of system with controllable canonical form. Based on the Lyapunov approach, the adaption laws and stability are also guaranteed. By the concept of backstepping, a nonlinear cascade system is decoupled into two sub-systems, then virtual controller and actual controller are designed sequentially to stabilize the system. Simulation results demonstrate the performance of our approach.

並列關鍵字

type-2 fuzzy neural network ； asymmetric membership functions (FOU) ； compensated control ； decoupled control ； nonlinear cascade system

參考文獻

[1] T. G. B. Amaral, M. M. Crisostomo, and V. F. Pires, “Adaptive Neuro-fuzzy Inference System for Modelling and Control,” 2002 First International IEEE Symposium on Intelligent Systems, Vol.1, pp. 67-72, 2002.

[2] V. Barra and J. Y. Boire, “Automatic Segmentation of Subcortical Brain Structure in MR Images Using Information Fusion,” IEEE Trans. on Medical Imaging, Vol. 20, pp. 549-558, 2001.

[4] R. T. Bupp, D. S. Bernstein, and V. T. Coppola, “Experimental Implementation of Integrator Backstepping Control and Passive Nonlinear Controllers on the RTAC Tested,” Proc. of the 1996 IEEE Int. Conf. on Control Applications, pp. 279-284, 1996.

[6] O. Castillo and P. Melin, “Adaptive Noise Cancellation Using Type-2 Fuzzy Logic and Neural Networks,” Proc. of 2004 IEEE International Conf. on Fuzzy Systems, Vol. 2, pp. 1093-1098, 2004.

[7] Y. C. Chen and C. C. Teng, “A Model Reference Control Structure Using a Fuzzy Neural Network,” Fuzzy Sets and Systems, Vol. 73, pp. 291-312, 1995.