Initializing Parameter Estimation Algorithms Under Scarce Measurements

Abstract In this paper, the problem of initializing identification algorithms with non-regular sampling is addressed. When a recursive identification algorithm is used to estimate the parameters, the convergence of the parameters is affected by the existence of wrong attractors. The initialization of the algorithm is studied in different situations. First, the algorithm starts without past information about the model parameters. An interpolation method is used to estimate the missing data. If a change of the control action updating rate is planned, the new model parameters are initialized by estimations obtained either by interpolation (if the periods are multiple) or by approximate δ modelling using the measurements taken under current operating conditions. Some examples illustrate the attractors avoidance and some conclusions are drafted.