Identifying structural mismatch for optimization models
Major successes have been achieved with the application of on-line optimization applications particularly real-time optimization (RTO). Although there has been much success with RTO, it has been shown that the performance of an RTO application tends to decrease over time due to changes in the process that are unaccounted for in the optimization models. The main objective of this work is to develop a procedure that can be used in an industrial environment to identify structural model mismatch between the actual process units and the optimization models.
RTO uses complex process models to determine the optimum operating conditions of a process and these models are updated using process measurements so that they provide the most accurate predictions. It is shown in this study that structural model mismatch can be identified by looking at the variation in the calculated model parameters when there are variations in the process operating conditions and sensor noise is not excessive. The effect of model mismatch on calculated model parameters is studied for a heat exchanger, CSTR, distillation column, and an ethylene furnace in this work.
Because RTO applications use process measurements to update model parameters a better measurement should lead to an improvement in the performance of the application. It is shown in this study that decreasing the amount of noise associated with a sensor used by an optimization application leads to a small increase in overall profit, less than 0.4% for the cases considered here, and the increase in profit depends on the shape of the optimization curve about the optimum.