Physics-Informed Digital Twin Architecture for Real-Time Environmental Compliance and Proactive Emission Control in Petroleum Refineries

Djamila Bouchaour, Mokhtar Benalia, Ahmed Abdelmouiz

With the tightening of regulations related to environmental protection, contemporary refinery plants are expected not only to increase their economic throughput but also to maintain strict control of emissions (NOx, CO2). It is impossible to solve the problem of nonlinear, multi-variable regulation using conventional approaches to control and management (PID/DCS). In this regard, the concept of Digital Twin (DT) opens the way to implement a holistic approach based on the use of new solutions and technologies. The key innovation of the current study lies in the elaboration of a Physics-Informed Hybrid architecture that will enable the connection between basic transport equations and deep learning models, providing thermal consistency. Moreover, the authors suggest integrating Explainable Artificial Intelligence (XAI) algorithms into the ecosystem of the digital twin, solving the problem of the "black box". Thus, the proposed DT solution allows achieving a prediction accuracy of 1.06% using real-time online detection data and soft sensors. Besides, it provides the ability to use proactive optimization measures that will help decrease NOx emissions up to 45.96%. The proposed operational approach acts as a crucial leverage point in moving towards Industry 5.0.