Information Forgetting Using the Augmented UD Identification Algorithm

Fundamental analysis of information forgetting in recursive identification shows that there are only two basic approaches: relative and absolute forgetting. Using the augimented UD identification (AUDI) algorithm developed by the authors, it is found that all the information pertinent to identification and information forgetting is contained in a single matrix called the information accumulation matrix (IAM). Analysis of the UD factored form of this IAM shows that the effect of the new information contained in the regressor, relative to the information (e.g., parameter values) already contained in the current IAM can be controlled by relative and/or absolute modification of the diagonal D matrix produced by the UDU^T factorization. This simplifies the interpretation of new and existing information forgetting approaches and provides a basis for developing effective design guidelines. Also, since AUDI is a least-squares (LS) type algorithm, the basic principles and interpretation carry over to other LS algorithms.