ABSTRACT
Several investigators have shown that nonparametric model building via artificial neural networks (ANN) can be used with nonlinear programming to implement nonlinear model predictive control (MPC). The use of nonlinear combinations of basis functions in an ANN can often provide models for MPC superior to conventional methods for nonlinear processes. We will discuss how to select an appropriate set of data to be used in training an externally recurrent ANN for modeling nonlinear dynamic processes as well as issues of model validation. In addition, we illustrate the use of an ANN as a process model in an MPC control scheme for parts of the Tennessee Eastman (TE) Test Problem.
