A way to Stabilize LS-Algorithms Used for the Identification of Non-Persistently Excited CTM-Processes

Abstract

The identification of continuous time dynamic processes (CTM-processes) requires the presence of persistantly exciting test signals that enforce dynamic process responses. However, in many cases processes are restricted to operate under regular conditions without any special test signals like random signals e.g. Then an applied least squares parameter estimation algorithm does not get enough information to show sufficient parameter convergence. This paper proposes a method that enables least squares parameter estimation algorithms used for continuous time systems to give good parameter estimates in spite of a non-persistant process excitation. Simulation studies show a convergence comparison between the modified least squares algorithm and the classical approach.