Identification of nonstationary stochastic systems using parallel estimation schemes

The multiple-model technique is proposed for the identification of nonstationary stochastic systems. First, the form of the optimal-local Bayesian predictor is derived under the assumption that system coefficients vary according to the random walk model and that the Kalman-filter-based algorithms are used for identification purposes. A rational extension of this strategy, which can be applied to identification algorithms of any form, is discussed. Specific suggestions are made concerning the possible choice of adaptation gains of the competitive adaptive filters. It is shown that the proposed scheme can significantly decrease the sensitivity of the identification algorithm to the rate of nonstationary of the analyzed system. Computer simulation results confirm the good estimation robustness properties of the parallel identification schemes.