On Using the Parameter Covariance for Improving the Recursive Least-Squares Algorithm

Normally, the gain matrix P(t) of the recursive least-squares algorithm has to be adjusted when parameter variations are detected. In this work, an on-line estimation of the parameter covariance Q(t) is proposed for this adjustment on the basis of the prediction error ¿(t). Simulation results of a process with time-varying parameters show that on-line auto-adaptation of this covariance brings better results compared to the use of a forgetting factor or the blind choice of an additionnal term to the gain matrix.