A directional forgetting algorithm based on the decomposition of the information matrix (Q5925928)

This paper develops a new directional forgetting algorithm based on an (orthogonal) decomposition method of the information matrix. In the proposed approach, only a specified part of the information matrix is forgotten at each update, and the undesirable phenomenon of estimator windup does not occur because nothing is forgotten in the part of the information matrix which is orthogonal to the excited space in some sense. Compared with various modifications to the standard exponential forgetting algorithm, the new directional forgetting algorithm is simpler, being based only on two adjustable parameters that can easily be prespecified. These properties make the novel proposal an attractive one for on-line identification techniques.