Graph Based GP Applied to Dynamical Systems Modeling

Model construction is usually guided by a trial-error process, where each iteration is divided into two steps: (i) collect or refine the set of equations that direct the system behaviour, normally in differential form, solving them using, most of the time, the S transform, and (ii) fit a set of properties (parameters) in the model obtained using observations taken from the real system. There have been many attempts to automate this process. We will extend an approach based on a search of a model of the system in a block diagram representation, where the trial-error process is solved with Genetic Programming. Some modifications over this approach are made to allow a more general family of models and to enhance its efficiency.