Owner/operators of industrial processes have, over the years, invested significant amount of resources in the instrumentation, control and optimization of their facilities. Consequently, most modern industrial processes are well instrumented and many of the processing units have well functioning Model Predictive Controls (MPC); however success with Real Time Optimization (RTO) of processes has been somewhat muted. This paper investigates some of the underlying causes of this dichotomy and presents a practical synergistic approach to RTO.

MPC is multivariable control of process that relies on multi-input multi-output, dynamic model of the process. MPC models are usually empirical and are identified by step testing of the process. Such models are linear and are good representation of the process dynamic behavior around the normal operating conditions. Usually, MPC executes on a sub-minute frequency and continuously implements manipulated variable moves.

RTO also relies on a multi-dimensional model of the process. Traditionally, RTO models are first principles and are based on steady state conservation of mass, energy and momentum. Such models are non-linear and are good representation of the process steady state behavior over the entire operating region. Usually, RTO executes on a much slower frequency as a supervisory layer giving setpoints to MPC for implementation of the optimum.

The development and maintenance of two separate models for MPC and RTO is inherently inefficient. An alternate formulation of RTO that leverages MPC models is possible and much more efficient. One such approach and its application to industrial processes is presented in this paper.