A fitness case strategy in genetic programming to improve system identification

Abstract

This article discusses the use of genetic programming for system identification. To this end, several experiments have been realized using observations obtained from a power transformer. The proposed strategy is to maximize the likelihood of convergence when searching for the model of a particular system. A traditional strategy for system identification in Genetic Programming is to take all the observations and evaluate the process of evolution to find a system model instance. Contrary to this, the proposed methodology is based on a partial subset of the observations, and then this subset is incremented until reaching the total set of observations. Furthermore, for comparison purposes we have used Eureqa, an open genetic programming based software tool for system identification.