ONLINE STATISTICAL RESULTS ANALYSIS IN REAL-TIME OPERATIONS OPTIMIZATION

This paper presents methods for the statistical analysis of plant oper ations optimization results with special consideration for real-time o ptimization (RTO) applications. The key challenge is to determine whet her the results of an optimization calculation should be implemented i n the plant. Since feedback data used to correct the model include noi se and the effects of high-frequency disturbances, the results of the model-based optimization calculations are corrupted by a stochastic co mponent. The methods developed in this paper apply multivariable stati stical hypothesis tests based on control charts in order to distinguis h between high-frequency disturbances propagated through the calculati ons and significant changes in the plant optimization variables with t he goals of reducing the frequency of unnecessary changes in the imple mented independent optimization variables and increasing plant profits . Only the statistically significant results are implemented in the pl ant. Case studies indicate that increased profit can be obtained by im plementing fewer changes to the process because the preponderance of c hanges due to noise are rejected whereas most meaningful changes are i mplemented.