Generating information for real-time optimization

Real-time optimization (RTO) applications monitor the behavior of processes, adjusting the setpoints of process controllers to track significant, low-frequency changes in the plant optimum. The performance of the optimizer depends on its ability to track these changes effectively and locate the true plant optimum operating conditions. The ability to track changes in turn depends on having sufficient plant information to update parameter estimates, improving the model predictions of the process behavior. This paper proposes an improvement to RTO performance by integrating information generation using experimental design techniques into the RTO algorithm to reduce uncertainty in the final optimization results. An expansion of the command conditioning (CC) subsystem evaluates when the predicted result from the economic optimizer will not generate a sufficient amount of information for updating. An A-optimal experimental design criterion is used to reduce uncertainty associated with decision variables by perturbing from the optimal point to another that generates more information. By sacrificing short-term profit, greater profit can be realized in future RTO intervals. (C) 2006 Curtin University of Technology and John Wiley & Sons, Ltd.