On the Initialisation of the Recursive Prediction Error Method for Nonlinear System Identification

The Recursive Prediction Error Method is widely used in nonlinear system identification and has been explored in considerable detail. However, the successful execution of the method relies on a suitable initialisation scheme. The aim of this research is to contribute to some fundamental issues concerning the initialisation of the method and therefore its ability to identify parameters of a nonlinear system. For a specific system, we identify regions in phase space, where nonlinear system identification is successful. More importantly, we also determine regions where identification fails and discuss these surprising results.