| Summary: | 博士 === 國立臺灣科技大學 === 電機工程系 === 97 === In this dissertation, we report our study on the relationship between the optimization based robust control mechanisms and the learning based control systems. In this study, a class of uncertain nonlinear control systems is considered and key points of integrating these two control systems are proposed and discussed. In the literature, it can be observed that the optimization based robust control mechanisms have been widely used in the learning control system and have nicely fits with the properties of the learning concept. Our first study is to understand the optimization based robust control notions generally. Dissipative control theory is studied and a dissipative controller is proposed. The obtained dissipative controller owns the H∞ tracking ability and has the energy convergence property of L2-gain. Therefore, the relationship between the input and output energies can be represented by a controllable attenuate parameter. The selection of the attenuate parameter becomes a means of coping with some undesirable phenomena of the learning control systems. It is also the original idea in robust learning control systems. Next, the dissipative controller is re-derived in terms of the normal L2-gain property. In this study, an L2-gain state feedback controller with an additional integral control term is designed and applied to the direct adaptive fuzzy control system. Besides, based on the property of L2-gain, the high initial gain problem of the L2-gain state feedback controller is resolved using a genetic adaptive scheme. With the use of the proposed adaptive scheme, not only the high initial gain problem is resolved effectively but also the initial tracking performance is not sacrificed. Finally, a novel adaptive law is proposed for the adaptive fuzzy control system to speed-up the learning process. From our simulations, it is evident that the learning speed of the adaptive fuzzy control system is significantly improved.
|
|---|
