Monitoring and Optimization of a Catalytic Process: Application to an Isomerization Unit at Sonatrach

Abstract

The petroleum refining industry faces increasing challenges in maximizing the production of high-quality, environmentally compliant fuels while maintaining operational efficiency. Catalytic isomerization units play a crucial role in this objective by converting low-octane light naphtha into high-octane isomerate. However, the complex nonlinear behavior of the process, coupled with continuous catalyst deactivation, makes conventional optimization approaches difficult to implement and maintain. This project presents a data-driven Digital Twin framework for the prediction and optimization of the catalytic isomerization process. The proposed approach leverages historical data collected from the Distributed Control System (DCS) to develop surrogate models capable of accurately predicting the Research Octane Number (RON) of the produced isomerate. Several machine learning and deep learning algorithms were evaluated and compared to identify the most suitable predictive model for representing the process behavior. Building upon the predictive model, an optimization framework was developed to maximize RON by determining the optimal inlet temperature setpoints of reactors R-001 and R-002 while accounting for catalyst aging effects. The Digital Twin enables rapid what-if simulations, performance monitoring, and optimization without interfering with plant operation. By combining machine learning-based prediction with advanced optimization techniques, the proposed framework provides a practical decision-support tool for refinery engineers, facilitating improved process performance, enhanced operational efficiency, and increased economic value