A Statistical Appreciation of Dynamic Matrix Control

The Dynamic Matrix Control (DMC) technique (representative of a new class of predictive-cum-optimizing control techniques that have recently made a noticeable impact on the process control community) has enjoyed great success in solving industrial process control problems where others have failed. To provide another vehicle for understanding the remarkable success of this technique, we examine, in this paper, the framework of DMC and draw attention to the striking resemblance it bears with certain time-proven statistical concepts although it is a strictly non-stochastic technique. The statistical concepts which feature prominently in the DMC framework are identified as the Levenberg-Marquadt non-linear regression method, Ridge Regression, Bayesian sequential decisions, and, of course, least squares estimation.