Model Predictive Control of a Tubular Chemical Reactor

Abstract

The paper deals with two methods of model predictive control (MPC) for the nonlinear system represented by a plug-flow tubular chemical reactor. The mathematical model of the reactor is described by a set of nonlinear partial differential equations. The application of a neural network model based predictive control (NNPC) strategy and a gain-scheduled (GS) model predictive control for a tubular chemical reactor as a system with distributed parameters is studied. Simulation results obtained using two model predictive control strategies were compared with the results obtained by a conventional PID controller. The best control responses with the smallest overshoots and the best values of numerical quality criterion IAE were reached using NNPC. This controller also assured the smallest coolant consumption, thus ensuring the most economical operation with the lowest energy consumption needed for achievement the control goal. The GS design is a linear-model based strategy that can be applied to the nonlinear processes. The price for using the linear process model is that the model has to be linearized at each operating point, and the different controller parameters have to be set according to the operating point. As the linear control technique is applied, only local stability is ensured. Decisions to be made are system-dependent, including the choice of an appropriate scheduling variable and scheduling procedure. On the other hand, NNPC is a nonlinear-model based strategy that combines neural-network-based modelling with the predictive control strategy. The simulation results confirm that NNPC as a nonlinear-model-based strategy is a good tool for successful control of tubular reactors.