本論文目的在於設計一基於適應性控制及自動學習演算法之智慧型控制系統,以解決複雜非線性的控制系統問題,本論文所提智慧型控制系統包含了類神經網路控制系統及小腦模型控制器系統,當中的控制器包含了主動控制器及補償控制器,主動控制器採用了類神經網路控制器或小腦模型控制器,作為最佳控制器的近似器,而輔助控制器則是用來減少剩餘的近似誤差,以滿足穩定性或 追蹤特性,並可提昇控制系統性能。 自動學習演算法可以藉由追蹤誤差來調整智慧型控制器的參數設定而不需要依靠過去的設計經驗,因此,為了讓系統有最佳化的參數,採用最陡坡降法及李亞普諾夫穩定理論來推導參數學習法則,並確保系統之穩定性;最後以系統模擬及實務操作之應用來驗證所提出設計方法的效能,包含兩軸手臂系統、混沌電路、彈簧-阻尼器系統、資料融合系統及無刷直流馬達,模擬及實作的驗證成果可以說明所設計之具有自動學習演算法之智慧型控制系統具有良好的控制性能。
This dissertation focused on the design of intelligent control systems based on the adaptive control and auto-learning algorithm for uncertain nonlinear systems. The proposed intelligent control systems include a neural network (NN) control system, and a cerebellar model articulation controller (CMAC). The developed control scheme is comprised of a main controller and an auxiliary compensation controller. The main controller, including an NN controller or a CMAC, is utilized to approximate an ideal controller, and an auxiliary compensation controller is utilized to attenuate the residual of approximation error with guaranteed stability or specified tracking performance. The auto-learning algorithms can adjust the parameters of intelligent control system by using the tracking error and without the need for preliminary knowledge. The on-line parameter auto-learning methodologies using both of the gradient descent method and the Lyapunov stability theorem are developed to increase the system learning capability and to guarantee the stability of the system. The developed design methods are then applied to some control systems, such as two-link manipulator systems, unified chaotic circuit, a mass-spring-damper mechanical systems, data fusion systems and brushless DC (BLDC) motors to demonstrate the effectiveness of the proposed design methods.