Performance assessment of model predictive control.

摘要

Model Predictive Control (MPC) technology has been successfully implemented in the petroleum, chemical, metallurgical and pulp and paper industries over the past three decades as an effective means to deal with multivariable constrained control problems. The primary focus of this thesis is the development of a practical technique for assessing the performance of single-input, single-output MPC schemes. A new benchmark curve is developed which represents the true design capability of an infinite-horizon Dynamic Matrix Controller. This performance standard is shown to be useful both as a model diagnostic and as a tuning guide during commissioning. A formal assessment procedure is presented which emphasizes the use of routine operating data plus knowledge of the deadtime to determine when it becomes feasible to invest in re-identification of the plant dynamics. The proposed method is illustrated through application to simulated processes and to experimental MPC data collected from a pilot scale stirred tank heater.