Data-driven dynamic modeling of renewable CO2 emissions in multimode industrial co-processing processes

Accurate modeling and real-time monitoring of renewable CO${}_2$ emissions in biofeedstock co-processing technologies are critical yet challenging, hindered by limited experimental data and static operational assumptions. This study introduces a novel data-driven dynamic modeling approach using an extensive dataset comprising 43,662 samples from the Parkland refinery. We implement change point detection algorithms to automatically partition the data into segments corresponding to different operating conditions and develop segment-specific robust regression models to predict CO${}_2$ emissions. The proposed framework uniquely integrates change point detection with robust regression, forming a dynamic monitoring system that continuously adapts to multimode industrial processes while balancing numerical accuracy, interpretability, and computational efficiency. These findings reveal that the CO${}_2$ emission ratio per unit of biofeedstock to fossil fuels fluctuates between 51% and 82% under varying operating conditions. The dynamic model exhibits strong agreement with experimental data, providing refineries with a practical, reliable tool for real-time emissions monitoring and regulatory compliance in industrial co-processing applications.