Nonlinear System Identification and Control Using an Input-Output Recurrent Neurofuzzy Network

In this work we develop an input-output recurrent neurofuzzy network in discretetime for identification and control of nonlinear systems. The structure is linear in the consequent parameters and nonlinear in the antecedent ones. The training of the antecedent parameters is achieved by linearizing them around a suboptimal value, assuming that the only known data are input-output signals obtained directly from measurements of the system, as well as some information about its structure (local stability and time delays). The training algorithm is based on a Kalman filter, stable under certain assumptions. It is also presented a theorem to check the stability of the resulting network in the Lyapunov sense, and a predictive control design. The performance of the network is shown by the identification and control of a nonlinear benchmark system.