| 摘要: | 已知現有之適應性類神經網路控制設計可保證系統追跡的半全域均勻穩定性，其先決條件是系統軌跡必須始終在類神經網路的有效區內，否則，追跡效果會明顯變差甚至引起不穩定現象產生。遺憾的是此條件的事前檢驗並不容易，特別是在嚴格回授系統的應用上。如同前一計畫所提，目前常用的一種方法是一旦系統脫離類神經網路近似有效區時，啟動另一替代之高增益強健控制器，將系統快速拉回，再換回原有之適應性類神經網路控制器，這種切換式控制設計可達到全域均勻穩定的目的。可是截至目前為止，此種方式僅限於可性化系統且不確定性為匹配情況下。有鑑於此，本計畫即在推廣其至嚴格回授系統且不確定性為不匹配的應用上。基本架構仍為一作用於類神經網路有效區內的適應性類神經網路控制；另一為在有效區外之高增益強健控制器；最後則為負責兩者調配之切換機制。有別於現有之設計，此種切換機制必須足夠平滑，以應付回步設計對虛擬控制器不可避免之微分所需。在理論設計告一段落後，本計畫擬將此設計應用於輪型機器人之防滑控制上，以驗證其有效性及廣泛應用姓。
Mostly, the adaptive neural tracking controllers ensure the semiglobal uniformly ultimately bounded (SGUUB) stability on the condition that the neural approximation remains valid for all time. However, such a condition is difficult to verify beforehand, which may lead to deterioration of tracking performance or even instability in reality. A direct remedy for this is to activate an extra high-gain controller outside the neural approximation region to pull back the transient. Such an approach, however, is restricted for linearizable systems with matched uncertainty so far. The objective of this project is to extend it to strict-feedback systems with bounded mismatched uncertainties. The control design will basically contain an adaptive neural controller, a robust controller, and a switching algorithm monitoring the exchange of the two aforementioned controllers. In particular, the switching algorithm shall be sufficiently smooth and hence can be incorporated with the backstepping tool. After the theoretical design for strick-feedback systems, we intend to extend it to the slippage avoidance of mobile robots to demonstrate its usefulness and wide applicability. |
