Bias-eliminated parameter estimation for sandwich nonlinear systems

In this paper, a bias-eliminated parameter identification method is proposed for application to sandwich nonlinear systems under stochastic measurement noise. By linearizing such a system expression based on the key term separation technique, a multi-innovation recursive least-squares algorithm is developed to simultaneously estimate the parameters of a class of sandwich systems with one nonlinear static subsystem embedded between two linear dynamic subsystems. Moreover, an auxiliary model strategy is introduced to solve the problem of consistent estimation on inner unavailable variables of the sandwich nonlinear systems. Besides, an adaptive forgetting factor is used to guarantee asymptotic convergence of parameter estimation. Finally, the upper bound of estimation error and unbiased estimation performance are clarified under a persistent excitation condition. An illustrative example is shown to validate the superiority of the proposed method.