Model Accuracy Requirments For Economic Optimizing Model Predictive Controllers - The Linear Programming Case

Model predictive controllers have proven very successful in controlling multivariable systems. In some cases these controllers are implemented on processes where the number of manipulated and controlled variables are not the same. These "non-square" systems provide optimization opportunities, which have been addressed by a combination of steady-state economic optimization / model predictive control. The optimization is model-based and is usually coupled with a model updating scheme to account for plant / model mismatch. Such a system must be designed so that the model-based optimization yields operating conditions which are optimal for the true plant. This paper presents methods to determine whether the model-based optimization is capable of finding the true process optimum despite errors in the model parameters. Discussions are concluded with a demonstration of the methods on a real-time gasoline blending control and optimization problem.