T.Y. Shevgunov

This paper introduces the structure of dynamical series-parallel neural network for linear ARMA model estimation based on time domain measured input and output signals of the system. The key feature of the offered structure is that it supports the simultaneous estimation procedures for the models of different orders whereas the choice of more appropriate model provides after the adaptation process. The model can be chosen based on two criteria - the relative root mean square error (RMSE) and information criterion MDL. The examples presented in the paper shows the numerical experiments for identification of the system with known transfer function driven by Gaussian random noise.