Neural-network-based model predictive control: a case study

Abstract

This paper presents a specific example of model predictive control (MPG) of an Ultra-High Temperature (UHT) milk treatment plan using a Artificial Neural Network (ANN) as the model. Single-network and composite-network models were trained on plant data with the composite-network model performing better. Simulations of a MPC scheme using the composite network model as a prediction model show that the scheme does not perform as well as a PI controller. Some pitfalls and possible improvements are noted.