Zusammenfassung
The most effective approaches for evolutionary
identifying dynamical processes depend on iterative
trial-error searches in a hierarchical fashion: a new
structure is proposed first; then, its set of
parameters is numerically determined, and the process
is repeated until a model accurate enough is
found.
Canonical Genetic Programming has been used to automate
this search; but its output can be diffcult to
interpret. Because of this reason, the use of
hierarchical learning methods, that combine GP search
of structures with deterministic optimisation
algorithms, has been proposed. We will show in this
paper that the output of such methods can be further
improved with non hierarchical algorithms. In
particular, we will show that the use of GA-P improves
the interpretability of the models and does a better
model search than previous approaches.
Nutzer