Meta-Analysis of Explainable AI Architectures for Optimized Real-Time NOx and CO2 Emission Control in Petroleum Refineries

Djamila Bouchaour, Mokhtar Benalia, Ahmed Abdelmouiz

The decarbonization of the petroleum refining industry calls for a paradigm shift from heuristically guided processes to intelligent automation. This paper conducts a comprehensive systematic review and metaanalysis of 64 papers, centered around artificial intelligence (AI)-driven prediction models in the domain of NOx and CO2 emissions reduction. While classical approaches have proven ineffective in accounting for the non-linearity of process dynamics, our research unveils a 54.7% prevalence of data-intensive techniques but also an urgent need for addressing the interpretability gap in order to facilitate implementation within the industry. This extended version introduces novel statistical findings concerning model consistency among various refining facilities. This paper groups the literature into three eras of control and predicts a natural evolution towards hybrid architectures to comply with ecological requirements. By summarizing the global state-of-the-art trend up until 2026, this paper provides an academic standard for designing autonomous refineries of the future without resorting to site-specific information.