Parameter estimation in partitioned nonlinear stochastic models

The application of maximum likelihood estimation and prediction error methods on dynamical systems require that it is possible to compute the innovations of the systems model which more or less implies that it must be possible to invert the system model. For nonlinear stochastic models this can be very difficult. In this contribution it is shown how this can be done very efficiently for a very rich class of nonlinear models by way of exact linearization. The method is illustrated on two non-trivial examples.