Optimal Selected Forgetting Factor for RLS Estimation

Abstract A new robust recursive least squares (RLS) algorithm of which an optimally varied forgetting factor is derived for parameter identification in a noisy environment. The concerned forgetting factor can now be updated recursively. It is a function of system noise variances and the derivation is based on the concept of nonlinear filtering using Bayesian Approach. A three sigma rule together with Huber model are used to vary the forgetting factor upon an optimal selected routine. The resulting algorithm has an ability to forget for time-varying system, and yet it also has an ability to emphasize old data as compared with the new data considered to be an spurious data. It has been found from simulation results that this algorithm is less sensitive to noise while its tracking ability is well retained.