Temperature prediction and model predictive control (MPC) of a distillation column using an artificial neural network based model

This paper addresses the possibility of using neural network based model predictive control (NN - MPC) with receding horizon principle to control a distillation column system as well as many other nonlinear systems with complex mathematic model. A small, refined distillation column model is built using balance equations and vapor liquid equilibrium (VLE) theory to provide data set for training neural network and testing the controller. The neural network is established by measuring the system inputs and outputs from the simulating model of the plant. The neural network based model is then used to predict the future outputs for the MPC controller. The simulation results indicated that NN - MPC is capable to track the changing set-points and reject the changing disturbances with a negligible steady state error due to the estimation error of the neural network and linearization.