Adaptive predictive control using recurrent neural network identification

This paper presents a new adaptive predictive control algorithm which consists of an on-line process identification part and a predictive control strategy which is updated every time a process model change is identified. The identification method is based on recurrent neural network (RNN) nonlinear AutoRegressive with eXternal input (NARX) model derived from dynamic feedforward neural network by adding feedback connection between output and input layers. Two model-based predictive control strategies have been studied: the generalized predictive controller (GPC) and nonlinear adaptive model predictive controller (NAMPC). The neural network training and validation data are obtained from the open-loop simulation of a validated first principles plant model. The identified neural network (NN) model is validated using the following three different validation algorithms: (1) one-step ahead cross-correlation; (2) Akaike's final prediction error (AFPE) estimate; and (3) 5-step ahead prediction simulations. The algorithm has been applied to the temperature control of a fluidized bed furnace reactor of the steam deactivation unit of a fluid catalytic cracking (FCC) pilot plant used to evaluate catalyst performance. The validation results show that the RNN models the reactor to a high degree of accuracy. Simulation results show that the proposed NAMPC control strategy outperforms the GPC at the expense of extra computation time.