Machine learning for real-time green carbon dioxide tracking in refinery processes

The global increase in greenhouse gas emissions presents an urgent environmental challenge, demanding innovative strategies for emission reduction and a fundamental shift in energy consumption practices. Co-processing biogenic feedstocks, such as used cooking oils and biocrudes derived from forest and agricultural residues, within existing oil refineries has been demonstrated as a cost-effective, scalable approach to producing low-carbon fuels, quickly helping the oil refiners to mitigate carbon dioxide emissions, leveraging the existing infrastructures. Despite its potential, monitoring the ”green” CO2 emissions originating from biogenic feedstocks during co-processing remains challenging. The molecular structure of biogenic components becomes indistinguishable from fossil-based molecules, necessitating costly, labor-intensive, and time-consuming sample collection and testing procedures, often involving isotope carbon analysis. This work proposes a new approach by applying artificial intelligence to model green CO2 emissions in real-time. By analyzing over 102,000 samples of industrial data from a commercial FCC unit, a robust machine learning framework is developed to provide continuous, cost-effective, and accurate green CO2 monitoring. The methodology encompasses a comparative analysis of ten input analysis techniques and five regression models to model emissions, achieving an average error margin of just 2.66% compared to traditional laboratory measurements. This AI-driven approach offers refiners and policymakers a practical tool for assessing the environmental performance of biogenic feedstock co-processing, facilitating informed decision-making in renewable fuel production.