On-line identification using hybrid method of regularized exponential forgetting

Abstract

The paper compares the ability of forgetting methods to track time varying parameters using various method of regularized exponential forgetting (REF). Three forgetting methods are analyzed in this paper: regularized exponential forgetting, second is regularized exponential forgetting with alternative covariance matrix (REFACM) and third is the novel hybrid regularized exponential forgetting method with a choice to disable or enable alternative covariance matrix (HREFACM). The comparison is performed in the terms of the quality of the observed output parameters. The observed parameters for all benchmarks are the integral sum of the Euclidean norm of the deviation of the parameter estimates from their true values (IS) and a selected band prediction error count (PE). As a supplementary information the eigenvalues of the covariance matrix are observed. All algorithms are validated by simulations in the Matlab Simulink software environment.