A performance monitoring algorithm for sustained optimal operation with economic MPC

This paper addresses the problem of performance monitoring for Economic Model Predictive Control (EMPC) in the presence of plant parameter changes. In order to cope with plant-model mismatch, we adopt a recently developed offset-free EMPC algorithm which requires the gradient of the plant input-output steady-state map. A subspace identification method is used in order to estimate this plant gradient from transient measurements. However, when the plant parameters change, this method may fail unless re-identification is performed. Hence, to start a new data collection for the identification an event-triggered mechanism is proposed, based on a suitable performance monitoring strategy. In this case this mechanism investigates a possible, more profitable, steady-state equilibrium and, if convenient, it re-identifies the plant gradient. The proposed monitoring technique is then successfully tested over an illustrative example of a chemical reactor.