Linear Quadratic Optimal Learning Control for Multivariable Systems

• Main Authors: CHING-YAO LIN, 林景堯
• Other Authors: Sheng-Hong Tsai
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/12933683394681751643

碩士 === 國立成功大學 === 電機工程學系碩博士班 === 92 === A linear quadratic optimal learning control solution to the problem of finding a finite-time optimal control history for multivariable systems is proposed in this thesis. Even though there is no detailed information of the system that is influenced by unknown but repetitive disturbances, it yields the learning achieve optimization. The newly added basis functions synthesize the optimal solution at a time, and it makes the outcome reaches optimality in a finite number of trials. These system-dependent basis functions have two characteristics: first, each newly added basis function will not alter the previously optimized ones; second, each basis function is selected by using the data from previous learning trials. Furthermore, some remarkable observations on the proposed approach for various system matrices and outputs are also presented. As a result, a desired connection with a pre-design feedback control before the learning process for the proposed learning control is newly pointed out in this thesis, which will significantly improve the effectiveness of the learning process. Numerical examples are used to illustrate the proposed methodology.