Model-plant mismatch compensation strategies in MPC
Model Predictive Control, MPC real time evaluation
The rise complexity in industrial processes led to the development of Advanced Controllers (APC), focused on improving the economic performance of systems. Model Based Predictive Control (MPC) is the most common advanced control strategy, having great applicability in several areas: petrochemicals, energy, food, automotive, aerospace, among others. This control strategy uses information from the process model to predict the future behavior of the system. The prediction is calculated minimizing a cost function which generates optimal control actions capable to lead the controlled variables to the desired setpoints values. An bad model prevents MPC from functioning properly, degrading its performance. Researchs in model quality evaluation have been performed, trying to detect and diagnose the occurrence of model-plant mismatches (MPM). There are also strategies that deal with these uncertainties, such as robust controllers and adaptive controllers. However, there is a demand for solutions that compensate MPM's effects. This work proposes two methodologies that use information from the model assessment, to try to temporarily correct the effects caused by MPM in the predictive controllers performance. A simple case study of a SISO system is presented, in order to show the consequences of the presence of these model-plants mismatches.