Neural network based Model Predictive Control

Model Predictive Control (MPC), a control algorithm which uses an optimizer to solve for the optimal control moves over a future time horizon based upon a model of the process, has become a standard control technique in the process industries over the past two decades. In most industrial applications, a linear dynamic model developed using empirical data is used even though the process itself is often nonlinear. Linear models have been used because of the difficulty in developing a generic nonlinear model from empirical data and the computational expense often involved in using nonlinear models. In this paper, we present a generic neural network based technique for developing nonlinear dynamic models from empirical data and show that these models can be efficiently used in a model predictive control framework. This nonlinear MPC based approach has been successfully implemented in a number of industrial applications in the refining, petrochemical, paper and food industries. Performance of the controller on a nonlinear industrial process, a polyethylene reactor, is presented.