Approximate neural economic set-point optimisation for control systems

Abstract

This paper describes a neural approach to economic set-point optimisation which cooperates with Model Predictive Control (MPC) algorithms. Because of high computational complexity, nonlinear economic optimisation cannot be repeated frequently on-line. Alternatively, an additional steady-state target optimisation based on a linear or a linearised model and repeated as often as MPC is usually used. Unfortunately, in some cases such an approach results in constraint violation and numerical problems. The approximate neural set-point optimiser replaces the whole nonlinear economic set-point optimisation layer. © 2010 Springer-Verlag.