Model assessment of MPCs with control ranges: An industrial application in a delayed coking unit

Poor model quality is one of the most frequent causes of performance deterioration in Model Predictive Controllers. As such, frequent model evaluation and correction is fundamental. Some assessment methods are reported in the literature, but most cannot deal with Model Predictive Controllers (MPCs) without fixed setpoints for controlled variables. Botelho et al. (2015, 2016a, 2016b) proposed a series of methods that include the controller tuning and the applied MPC implementation in the assessment procedure. Their main advantage is setpoint independence. This paper analyzes the application of these methods in an industrial MPC with control ranges. The system studied is an MPC of a fractionating column in a delayed coker unit of a refinery in Brazil. The results demonstrate that the method is capable of correctly quantifying the effect of modeling problems and identifying whether they are related to a model-plant mismatch or unmeasured disturbance.