AbstractA two‐level approach for optimal final‐value control of non‐linear systems is considered. On the higher level, an on‐line optimization (using a Ritz parameterization of the control functions1) is performed during each major sampling interval in order to compute the control functions for the next intervals. On the lower level, an explicit model‐following control is used with a minor sampling interval. This allows, even for heavily disturbed systems, sufficiently long major intervals for the on‐line optimization. An ammonia reactor2is used as an example, and the application of this procedure is discussed. A realization of the controller using a hierarchical computer configuration gives an indication that, with a reasonable amount of hardware, sampling intervals of about one minute for the on‐line optimization and well below one second for the model‐following control