A two-layer model predictive control system with adaptation to variables status values

This paper proposes a formulation of a linear Model Predictive Control system based on two layers. The formulation takes into account a status value associated with each variable. At the lower layer, a basic Model Predictive Control module minimizes a constrained quadratic cost function for the computation of the future control moves. At the upper layer, an additional module, based on a Linear Programming problem, searches for optimal steady state targets. To obtain resilience of the control system in particular conditions, at each control instant, the introduction of a parameter, the status value, associated to each process variable, is proposed. At this purpose, a suitable additional module has been introduced in the considered architecture that contributes to the definition of the status of each manipulated or controlled variable. Ad hoc modifications in the mathematical formulations of the optimization problems of the two layers are proposed to include the information provided by the status values. This strategy has been introduced in a control system that optimizes a pusher type reheating furnace of an Italian steel plant. Simulation results show the validity of the proposed approach. Field results prove the reliability of the designed control scheme, in terms of economic optimization, environmental impact reduction and productivity/product quality maximization.