An approximate maximum likelihood estimation of nonlinear systems using extended neural networks

摘要

This paper proposes an identification method for nonlinear systems with colored noise, using extended neural networks. The extended neural networks consist of two sub-networks, the real-sub-networks which track the undisturbed output of a system and the noise-sub-networks which make the filtering estimation error white. As a result, approximate maximum likelihood estimator can be obtained by relatively simple algorithm without any calculation of inverse matrix. The effectiveness of this proposed method is shown with some numerical examples.