A neural network based explicit model reference adaptive controller

A stability proof is presented for closed-loop systems using controllers based on approximate models of the plant. The resulting controller structure is inherently MIMO (multiple-input multiple-output) and allows for online learning in response to changes in the plant dynamics. The price for this is fairly high: the controller is iterative, and some strong assumptions are made on the existence on a controller minimizing υ and on the ability of the network to learn the plant dynamics. This controller structure has been successfully applied to single degree of freedom linear systems with step changes in their mass properties, and to the traditional cart/inverted pendulum problem