The effect of model fidelity on real-time optimization performance

Abstract

The ability of a real-time optimization (RTO) system to track the changing optimum closely relies on an accurate model for representing the plant behavior. This paper investigates the effect of model fidelity on RTO performance using a simulated industrial boiler network case study. Three optimization approaches with very different modeling fidelity are investigated: (1) model-free direct search method; (2) model-based method using a simplified efficiency curve model; and (3) model-based method using a fundamental model. The model-free direct search method builds a locally linear model using plant data. It takes many steps to reach the optimum, which causes a significant profit loss during tracking. This tracking loss can be reduced by using the model-based RTO system. The RTO system with an updated, detailed fundamental model is able to track fast and large disturbances because the model is accurate in a large range of operation. The RTO system with a simplified efficiency model requires periodic experimentation to correct for the disturbances, which can cause a significant profit loss during experimentation and tracking. This study demonstrates how quantitative performance measures improve as higher fidelity models are used in real-time operations optimization.