This work is concerned with the identification of global models for non-linear
dynamical systems using multiobjective evolutionary algorithms. The 
identification of non-linear systems involves the processes of structure 
selection, parameter estimation, model performance and model validation and 
involves a complex solution space. Evolutionary Algorithms (EAs) are search 
and optimisation tools founded on the principles of natural evolution and
genetics, which are suitable for a wide range of application areas. Due to the
versatility of these tools, two evolutionary methods, Genetic Algorithms (GA)
and Genetic Programming (GP), are investigated for use in non-linear system 
identification.

Initially, these evolution-based approaches are used to evaluated the Akaike
Information Criterion (AIC) in order to identify the best non-linear system
model. However, only model size (a model structure attribute) and short time 
prediction error (a model performance attribute) are considered within this 
information criterion.

GP was found to provide superior performance for this single-objective function
and as a result, its use was extented to encompass multiobjective functions.
MultiObjective Genetic Programming (MOGP) is applied to multiple, conflicting
objectives and yields a set of candidate parsimonious and valid models which 
reproduce the original system behaviour at different working conditions (global
models).

The MOGP identification approach is applied, first, to simulated data and 
then to sets of observations from a real system, a Rolls-Royce gas turbine
engine.

The benchmark problem of the 6-Boolean Multiplexer is also introduced to 
demonstrate the applicability of MOGP in a domain other than system 
identification.

Finally, MOGP is demonstrated as being a highly applicable non-linear system
identification technique by addressing the identification of systems with
chaotic dynamics and its ability to deal with rational system representations.