Model predictive control of a chemical process using neural networks

A simulation study on the control of a multivariable chemical process by using a neural network model predictive control strategy is described in this paper. The laboratory process, in which temperature, pH and dissolved oxygen are involved, has characteristics typical of industrial processes. The main difficulties in control of this process are non-linearity, coupling effects among variables and long time-delay in heat exchange. Neural sub-system models are developed from real process data for the model predictive control strategy and also for use as a bank of parallel models to represent the process in the control simulation. The control simulations are performed before the online control to gain more insight of the process and to determine suitable controller parameters. The simulation results are demonstrated in the paper.