本論文將探討非線性系統之智慧型控制器設計上的一些課題，並提出解決方案。一般而言，設計適應性控制器所遇到的重大難題，是如何決定其控制規則及適應性法則，以確保能迅速且強健地控制非線性系統。 首先，吾人將非線性系统的穩定性分析降階為線性矩陣不對稱（LMI）求解問題，提出具有 性能之適應性控制器設計，由遺傳演算法決定控制規則和後件部（consequent）參數初始值，並根據Lyapunov穩定準則使系統狀態變數達到所要求的 追蹤性能，進而穩定非線性系統。 接下來，為處理多變數非線性系統之穩定性控制問題，吾人採用奇異激勵（singular perturbation）概念，將非對稱多變數系统解耦（decouple）成數個較低階數且具不同時間常數的獨立子系統後，後件部決策參數的初始值由遺傳演算法決定，並將邊界層函數引入所推導之改良式適應性法則，確保系統狀態誤差收斂在一定值區間。 最後，為處理類神經網路（neural networks）控制複雜非線性系统常遇到的強健性問題，吾人先以放射狀基底函數網路（radial basis function networks）近似一個非線性控制裝置，由遺傳演算法決定其輸出鍵結權值（weights）之初始值，再根據Lyapunov穩定性準則推導之改良式適應性法則來確保對非線性系統之穩定控制。 綜上所述，本論文提出下列控制設計方法：(1)具 追蹤性能之遺傳演算適應性模糊滑動模式控制器設計、(2)遺傳演算適應性模糊滑動模式控制器設計及(3)遺傳演算適應性類神經控制器設計。 吾人所提出的智慧型控制系统的設計程序與改良方法可藉由電腦模擬來說明，可顯示出所提出控制方法及改良式控制器的強健性和效能，皆達到預期的控制目標。 Abstract In this thesis, we investigate and discuss some intelligent controller designs for nonlinear systems, proposing several strategies. Generally, the biggest difficulty encountered when designing an adaptive controller which is actually capable of rapidly and robustly controlling nonlinear systems is the selection of the control rules and of the most appropriate initial values for the parameter vector. The first step is to reduce the stability analysis of the nonlinear system into linear matrix inequality (LMI) problem solutions, for which we then propose an adaptive control strategy incorporated into an tracking control scheme. The control rules and the consequent parameters are decided via the use of genetic algorithms (GA). Then, based on the Lyapunov’s stability criterion, we utilize the rules and parameters that guarantee the best tracking performance throughout the entire system states. After this we use the most singular perturbation scheme to decouple a non-square multi-variable system into several reduced-order isolated square multi-variable subsystems for multi-variable nonlinear system stability analysis. The initial values of the consequent parameter vector are decided via genetic algorithms. The boundary-layer function is introduced into these modified updating laws to cover parameter errors and modeling errors, and to guarantee that the state errors converge into a specified error bound. Finally, we look at the type of robust problem often met with in designing a neural network controller for complex and nonlinear systems, wherein we use radial basis function networks to approximate the control plant for nonlinear systems. The initial values of the consequent weight vector are decided via GA after which a modified adaptive law is derived based on Lyapunov’s stability criterion to simultaneously stabilize and control the nonlinear systems. Focusing on the fore-mentioned general topics, we mention the following control strategies: (1) the GA-Based Adaptive Fuzzy Sliding Mode Controller (GA-Based AFSMC); (2) the GA-Based Adaptive Fuzzy Sliding Mode Controller (GA-Based AFSMC); (3) the GA-Based Adaptive Neural Network Controller (GA-Based ANNC). The design procedure for the proposed intelligent control systems (with some modified methods) is illustrated by computer simulations the results of which are utilized to demonstrate the control methodology, as well as the robustness and efficiency of the constructed controllers.
