Efficient parameterisation of nonlinear system models: a comment on Nöel and Schoukens (2018)

Overview & Synopsis

Nöel and Schoukens (2018) discuss a methodology for the discrete-time state-space identification of nonlinear systems and apply this to experimental data from the well known Silverbox nonlinear circuit, producing a model characterised by 13 parameters. This model explains the data very well but the parameter estimates are not well defined in the optimization results, with the very large confidence bounds suggesting that the model is over-parameterized. This comment shows that this is indeed the case and that the data can be explained equally well by an alternative continuous-time, State-Dependent Parameter (SDP) transfer function model with only 6 parameters, the estimates of which are well defined with very tight confidence bounds. The comment also raises questions about how the model form for nonlinear systems such as the Silverbox should be identified and suggests that the Data-Based Mechanistic (DBM) approach to modelling has some advantages in this regard.