This work proposes a genetic algorithm (GA) to solve process estimation problems when the real process presents high orders polynomials (complexity model) or non-linearities, non-minimum phase behavior, etc. The algorithm finds the best linear model in the pole and zero form to represent the real plant using its input and output signals. A new chromosome representation was introduced and a new ''fitness'' function based on the tradeoff bias x variance was developed. To validate this genetic estimator, simulations studies were done and the GA performance was compared with one obtained by use of the traditional least square estimation method.
Identification; Least Square estimation; genetic algorithm; reduced order model
