An improved inverse neural control structure for nonlinear dynamic systems

Neural networks with their inherent parallelism and their ability to learn has been seen by many authors in the field of system control, as an exciting possibility to design adaptive controllers. This paper focuses on the capabilities and performances of the inverse neural control structure. Usually, a traditional inverse neural controller performs very well on setpoint changes, but is not so good on the disturbance rejection. In the paper, we propose an improved structure of inverse neural control, and we are concerned mainly with two aspects: disturbance rejection, and the control system behaviour with regard to the process parameters variation and to the manifestation of the unmodeled dynamics