GARA: A genetic algorithm with resolution adaptation for solving system identification problems

This paper describes a Genetic Algorithm with Resolution Adaptation (GARA) used to solve System Identification (SI) problems. The algorithm is both hybrid, because it combines the genetic search with a fast local hill climbing method to improve the speed of convergence, and adaptive, because it uses information on the actual convergence of the population to modify the resolution of the binary representation of the search space. In particular, we introduce a new index of convergence of the population, which offers an effective estimate of the convergence with a small computation charge. When the convergence is reached, a heuristic strategy reinitializes the GA to obtain solutions that iteratively approach the global optimum with increasing speed and accuracy. In a set of case studies, GARA is used to estimate delays, gains, poles and zeros of discrete time linear systems, providing performances considerably better than other GAs previously used in SI problems. The reliability and the accuracy of GARA make it an interesting alternative to other conventional SI algorithms.