Experimental Data for Model Validation

Abstract

This chapter is concerned with experimental modelling techniques and the use of experimental test data for the purposes of model evaluation. It includes an overview of system identification and parameter estimation methods and gives emphasis to the importance of issues associated with the optimisation of the model structure. In the context of model validation, the potential “generalisation” capabilities of an identified model are considered in terms of model predictions for experimental situations which are not exactly the same as those used in the identification process. Issues of over-fitting and under-fitting are also discussed and procedures for the estimation of parameters of physically-based nonlinear models are considered, including evolutionary computing approaches, such as Genetic Algorithms and Genetic Programming. Issues of identifiability are discussed in some detail, as are questions of experimental design, the selection of the most appropriate test-input signals for system identification and model validation, together with ways of assessing the accuracy of parameter estimates.