An output error recursive algorithm for unbiased identification in closed loop (Q1360482)

By introducing a particular form of an output-error predictor for the closed-loop, which is used to update the plant parameter estimates, one derives a parameter estimation algorithm for the plant model which is globally asymptotically stable and asymptotically unbiased in the presence of noise. Sufficient conditions for stability in a deterministic environment and convergence in a stochastic environment are related to a positive real condition (which can be relaxed by data filtering or adding a proportional adaptation). Illustrations of the theoretical approach by simulation and real experiments are finally presented.