Data-driven Self-optimizing Control

Abstract

Selection of controlled variables (CVs) plays a crucial role in overall operational performance. Existing approaches for CV selection based on the self-optimizing control (SOC) strategy are model-driven. Selection of CVs using rigorous nonlinear process models requires linearization around nominal operating point. Inevitably, these approaches not only result in large losses as a result of linearization errors, but also are difficult to be adopted for practical applications due to the requirement of a rigorous model. In this paper, a novel data-driven approach, where the necessary conditions of optimality (NCO) are directly approximated by CVs using operational data in a single regression step is proposed for selecting CVs based on SOC. The new approach does not require evaluation of derivatives so that process models associated with commercial simulators can be directly used for CV selection. The effectiveness of the proposed approach is demonstrated using a 3-stream heat exchanger network (HEN) case study.