Identification and predictive control of a melter unit used in the sugar industry

Abstract This paper presents the real identification and non-linear predictive control of a melter unit; the unit is used in a sugar factory placed in Benavente (Spain). The proposed approach uses a specific recurrent neural network that allows us to identify a non-linear model of the process, providing a mathematical representation in the state space form. Output and state variables can be obtained from the inputs and measured disturbances acting on the system. The neural based predictive control is carried out through the optimization of a cost function that takes into account the output prediction errors from a reference trajectory and the future control efforts, by using the identified model as a prediction model for the system outputs. The solution to this problem provides the optimal set of future control actions, but only the first one is applied to the real process, and the optimization problem is solved again at time t + 1. The results show the good performance of neural predictive control and its suitability for applications in real systems, particularly in the process industry.