Learning for the adaptive control of large flexible structures

An important problem in the adaptive control of large flexible structures is to select the adaptive controller parameters appropriately so that good performance is obtained. A method, based on machine learning, for solving this problem is introduced and discussed. It is shown that learning by observation and discovery can be effectively used in the adaptive control design, and in particular in optimizing the system performance. The search for the optimal performance is formulated as an unconstrained nonlinear optimization problem where the variables are the parameters in the adaptive controller and the cost function is the performance index which is defined as a weighted sum of the root-square-error, the maximum error, and the settling time. The learning system is built on top of the adaptive controller, and it employs a knowledge-based system which consists of a rulebase and a database. The results obtained are used to propose an intelligent adaptive control system where the parameters in the adaptive controller are to be tuned online without human supervision. Results of simulations are performed on the model of a large space antenna are given