Model predictive control performance assessment using a prediction error benchmark

Model predictive control technology can now be found widely in a variety of applications including petroleum, chemical and papermaking industries. An approach is proposed to decide a benchmark and monitor model predictive control performance on-line. A performance measure based on multi-step prediction error benchmark is shown to be more realistic without the requirement of process models or interactor matrix. A practical on-line monitoring strategy is presented which emphasizes the use of routine operating data plus the order of the interactor matrix to determine when it becomes worthwhile to re-identify the plant dynamics and re-install the model predictive control application.