AI-Driven Optimization of Refinery Configurations: A Novel Approach to Predicting Production Outputs and Capacities

Abstract In the context of conceptualization and operation of refineries, achieving an optimal configuration emerges as a critical endeavor. Our proposed methodology transcends conventional approaches by seamlessly integrating domain expertise, empirical insights, and cutting-edge artificial intelligence (AI). Our approach holistically captures the intricate linkages between different refinery process facilities (also referred as process units or modules in later part of this paper) to make their best use in ever changing market conditions. Key aspects include robust predictive modeling for individual modules, meticulously considering feedstock availability, product demand, price information, process interdependencies and historical data. These predictions serve as foundational inputs during the conceptualization and planning stages, guiding optimal configuration decisions. Additionally, our rigorous cost modeling extends beyond capital costs to encompass financial parameters sufficient for shortlisting of configuration options for detailed economic evaluation. The machine learning models are evaluated based on the error metrices and has confirmed the models performs with in the acceptable tolerance ranges. Moreover, our integration of Large Language Models (LLMs) adeptly transforms complex refinery configurations into actionable insights expressed in business language. Decision-makers gain a holistic view, aligning technical choices with overarching organizational goals. Ultimately, this pioneering approach eases conceptualization process of new refineries and empowers operating refineries toward sustainable production and refinery-wide optimization.