Parameter-constrained adaptive control

Under certain conditions, parameter constraints that impose a priori information about the open-loop system can dramatically improve the performance of explicit adaptive controllers, Under other conditions, the constraints dan actually decrease performance. First, this paper presents a novel parameter-constrained identifier on the basis of an efficient, quadratic program solver applied semirecursively, making it ideal for real-time adaptive control. Second, several useful linear constraints for second-order ARMAX models are provided, along with a few examples of their development. Third, the algorithm and the constraints are applied to a benchmark model to explore several conditions, summarized as six guidelines, under which parameter constraints improve or worsen adaptive control. In this last part, it is shown that; con?mon orthogonal projection can produce poor results. It is also shown that a priori information is increasingly valuable as excitation decreases and that it is especially useful for adaptive control when combined with re-identification techniques. These results are then applied to the pharmacological control of a time-varying second-order ARMAX model of blood pressure.