Técnicas de otimização aplicadas à Identificação de sistemas dinâmicos

Abstract

Building a reliable model is one of the pillars of controlling autonomous systems. Identification is a process that extracts characteristics of a process for further analysis and manipulation of the system. Mathematical models are a guide to the real behavior of a process. However, such models require the input of some specific parameters of the system operation. Parameter estimation is a form of identification that aims to fine-tune models so that they behave as closely as possible to reality. Therefore, estimation methods together with an efficient input signal can produce good results. SOESGOPE is a method that combines excitation signal generation with parameter estimation. Although it presents satisfactory results, it should be analyzed from an optimization point of view, that is, how reliable and robust it is against different configurations. Thus, this work presents analysis in this aspect and a proposal for alteration when a lack of convergence is verified. The applicability of the method is enhanced in situations where the model is a reasonable copy of real behavior. However, in real situations, it is not possible to guarantee that a mathematical model will produce outputs similar to the real process when faced with the same input. To circumvent this problem, this work also proposes a hybrid technique that is able to use the mathematical model as a guide and learn characteristics peculiar to that process. Such a hybrid model will be able to correct certain errors of the real system related to the model and achieve more reliable outputs. Furthermore, SOESGOPE can count as an ally to the learning process since it provides more efficient excitation signals. To confirm the applicability of the proposed method, an autonomous boat is used for application and verification of the effectiveness of the method.