A large-scale once-through-type ultrasupercritical boiler power plant is investigated for the development of an analyzable model for use in developing an intelligent control system. Using data from the power plant, a model is realized using dynamically recurrent neural networks (NN). This requires the partitioning of multiple subsystems, which are each represented by an individual NN that when combined form the whole plant model. Modified predictive optimal control was used to drive the plant to desired states; however, due to the computational intensity of this approach, it could not be executed quickly enough to satisfy project requirements. As an alternative, a reference governor was implemented along with a PID feedback control system that utilizes intelligent gain tuning, which, while more complicated, satisfied the computational speed required for the controller to be realized.