2000 年 36 巻 11 号 p. 923-929
A Wiener system, i.e., a system comprising a linear dynamic subsystem and a nonlinear memoryless subsystem connected in a cascade, is identified. The linear subsystem is expressed by the ARX model and the order is unknown. The class of all possible nonlinearity is so wide that it cannot be presented in a specific form. The artificial neural networks which have the ability to learn complex nonlinear relationships provide an ideal means of modelling complicated nonlinear systems. The artificial neural networks are applied to represent these nonlinear elements. The connection coefficients of the artificial neural networks and the parameters of the ARX model are related each other, then they can not be obtained independently. So, they are estimated simultaneously based on the maximum likelihood method. The order of the linear subsystem and the structures of the neural networks are selected by the minimum description length. It is demonstrated in digital simulation that the proposed identification technique is efficient in the system with high level nonlinearities.