Identification of nonlinear systems using empirical data and prior knowledge -- An optimization approach (Q1914996)

A new formulation via an optimization problem of the joint modelling and identification problem is proposed. The standard prediction error criterion is augmented with penalty terms and constraints that incorporate some useful types of prior knowledge, including a regularization penalty term that enforces smoothness of the model. The solution to a simplified modelling problem using function space methods is explicitly found. Numerical procedures to solve the general problem are presented and discussed in detail. A fairly complete semi-realistic simulation example illustrates the power of the approach and the effect of different kinds of prior knowledge on the model.