A fading memory data-driven algorithm for controller switching

Abstract

A data-driven controller switching algorithm used for adaptive control is investigated. A new cost-detectable cost function based on fading memory data is constructed so as to reduce the influence of older data. A new controller switching algorithm is designed to guarantee that switching stops and that the closed-loop system is stable. Theoretical analyses and simulations are presented to show that, when the plant changes slowly or infrequently, the new algorithm can detect instability and switch to a stabilizing controller sooner and more smoothly, once the currently active controller becomes destabilizing for the new plant dynamics. It is also shown that this algorithm can be used to attenuate the Dehghani-Anderson-Lanzon phenomenon.