Corrector Least Squares (CLS) Identification Algorithm for Adaptive Control

Of all the identification algorithms used in real-time applications, the recursive least-squares (RLS) algorithm and its extensions are the most popular. With most RLS implementations, severe restrictions on the system are required in order to insure the validity of the estimation, e.g., the plant parameters under investigation often are constrained to be time-invariant. The CLS method of this paper capable of identifying systems with changing parameters. The approach implements the notion of adaptation of the covariance matrix (or the gain vector) in a recursive fashion so that both fast and slow changes in dynamics of the system may be tracked with good accuracy. CLS also contains a novel data richness test, called the ratio test, to mitigate the problems of covariance blowup and parameter drift.