CLOSED-LOOP NONLINEAR PROCESS IDENTIFICATION USING INTERNALLY RECURRENT NETS

The feasibility of using an internally recurrent network (IRN) as a no nlinear dynamic model for a plant and directly identifying the model u sing the plant input and output data from the controlled plant is disc ussed in this paper. It is shown that if the setpoint signal is employ ed as the excitation to the plant, an open loop model can be identifie d by the direct identification method from closed loop data. Simulatio ns show that the IRN structure is a satisfactory nonlinear dynamic mod el structure for identifying nonlinear plants under closed loop contro l, and the long term prediction. performance of an identified IRN mode l is generally good. In our investigation, we used nonlinear programmi ng for IRN training, and it proved to be a good method for off-line ne twork training. (C) Elsevier Science Ltd.