ROBUST, LOCALLY OPTIMAL MODEL REFERENCE ADAPTIVE CONTROLLER.

Abstract

A new model reference adaptive controller is proposed. The innovation of this algorithm is that it adds a filtering of the plant inputs and outputs prior to their use in determining the adapted controller parameter vector update. This modification is based on engineering intuition that suggests that this filtering should provide an improvement in the robustness of the adaptive controller to unmodeled plant actions outside the bandwidth of this added filter. Global convergence of this algorithm is easily proven in ideal use, i. e. , when there are no unmodeled plant components. In nonideal use, this filtering is interpreted as appropriately influencing the cost function being locally minimized by this adaptive controller, if stable. This optimization interpretation provides a guide of how to choose this added filtering to improve algorithm robustness. These results are illustrated with simulations.