Sobre o Uso de Métodos de Predição de Erro para Identificar Módulos em Redes Dinâmicas

Abstract

In this paper, the problem of identifying dynamic modules that operate in a complex interconnect structure is discussed. This problem is presented as an extension of the classical closed-loop configuration and the goal is to consistently identify the transfer functions embedded in a dynamic network, based on the internal and external signals of the network. The objective of this paper is to provide a comparison of methods and apply them to identify modules in dynamic networks. Graph representation tools are presented to verify the network interconnection structure conditions. For a known topology, it is shown that classical prediction error methods for closed-loop identification can be generalized to provide estimates of the modulus, under certain experimental circumstances. The analysis and correction of the signals that will be used in the predictor constitute the adaptation for the case of networks and are discussed in this work. To attain that purpose, the direct method, which depends on noise models, and the two-stage method, which is independent of the noise model but depends on external excitation signals, are considered. The results show that the direct method can present bias and the two-stage method is able to reduce the variance of the parameter estimation. Therefore, the two methods presented were able to consistently estimate local modules in dynamic networks, with the two- stage method showing better indicators. The models were validated using a second set of input and output data. The results showed that the model was able to reproduce the dynamic behavior of the system with the validation data.