智慧型適應控制於飛彈防禦系統及分數階系統

Abstract

為解決飛彈防禦系統及分數階系統的控制問題，本論文提出的控制架構分成兩個部分。第一部分，一個新的適應性自組織映射與回饋式神經網路控制器被提出於飛彈防禦系統中，並做到任務分配以及路徑演化。而第一個架構包含了一個回饋式神經網路控制器以及一個監督控制器。回饋式神經網路被設計來迫使防禦導彈朝向來襲導彈，而監督控制器被設計來減少回饋式神經網路控制器和理想控制器之間的誤差。在達成任務分配之後，我們所提出新的自組織映射與回饋式神經網路控制器的權重因子能被更新調整，並調度防禦飛彈朝向它們相對應的目標。我們提出了數個例子顯示，所提出的自組織映射與回饋式神經網路控制器於飛彈防禦系統中可以實現良好的控制性能和減輕計算負擔。在第二個部分我們集中在分數階系統的應用，我們發展了一個新的創新方法來做系統鑑別，使得一個整數階模型系統近似分數階系統。給定一個分數階系統，一個新的系統鑑別矩陣能被推導出來以鑑別整數階轉移函數的係數。另外，我們提出了一個新的均值為基礎的適應性模糊神經網路滑動控制，以解決主-從分數階混沌系統的同步問題。在與傳統的泰勒方法相比，所提出的均值為基礎的方法可以估測其在鑑別模型上的一階微分項，這在某種程度上能減輕計算上負擔。藉由學習演算法，其適應律和控制律能被即時線上調整以同步主-從分數階系統。此外，閉迴路系統的穩定性不僅可以得到保證，而且同步的問題也可以得到實現。最後從模擬結果驗證其所提出方法的可行性與適用性。

In this dissertation, two part of control schemes are proposed to solve the control problems of missile defense system (MDS) and fractional order system. The first part is a new adaptive self-organizing map (SOM) with recurrent neural network (RNN) controller for task assignment and path evolution of MDS. The first scheme is composed of a RNN controller and a monitoring controller. The RNN controller is designed to force an agent (or defending missile) toward a target (or incoming missile), and a monitoring controller is also designed to reduce the error between the RNN controller and ideal one. After task assignment, the weighting factors of our new SOM with RNN controller are activated to dispatch the agents toward their corresponding targets. Several examples are presented to show that the proposed SOM with RNN for MDS can achieve favorable control performance and alleviate computational burden. The second part is focused on the application of fractional order systems. We develop a new innovative method for approximating fractional order system by an integer order model for system identification. A new SID (system identification) matrix can be derived to identify the coefficients of an integer order transfer function to approximate the given fractional order system. In addition, a new mean-based adaptive fuzzy neural network (FNN) sliding mode control is proposed in solving the synchronization problem among the master-slave fractional order chaotic systems. In comparison with traditional Taylor method, the proposed mean-based method can estimate the first order derivative term on the identifier model, which will somehow alleviate the computational burden. Based on the learning algorithms, the adaptive laws and control laws can be tuned on-line to synchronize the master-slave fractional order systems. Furthermore the stability of the closed-loop system can not only be assured but the synchronization problem can also be achieved. Finally, the simulation results demonstrate the feasibility and applicability of the proposed methods.