A performance measure for constrained model predictive controllers

Model predictive control (MPC) is the dominant control technology for constrained, multivariable chemical processes. Performance assessment of MPC controllers is an important problem that has received little attention. Most multivariable controller assessment techniques are based on comparing the actual performance to that achievable under minimum variance control (MVC). MVC measures not well suited for MPC controllers because inherent control limitations imposed by constraints are neglected and minimum variance performance may be unachievable by any MPC controller even in the absence of constraints. We propose a performance measure for MPC controllers that utilizes the available process model to determine control system performance under constraints. The measure is computed as the ratio of the expected performance to the actual performance over a moving horizon of past information. The MPC performance measure is compared to a MVC measure using a linear distillation column model.