Thermal pyrolysis optimization and production forecasting in heterogeneous shale formations for enhanced oil recovery with large language models

Abstract

Oil shale, a significant unconventional fossil energy resource, plays a crucial role in global energy security. However, its complex composition and heterogeneous nature pose substantial challenges for efficient extraction and utilization, often resulting in suboptimal energy yields and environmental concerns. This study presents an innovative approach called Lingu-Graph Hybrid Network (LingGHN) to optimizing oil shale extraction and thermal utilization processes, addressing critical challenges in fossil energy efficiency and environmental sustainability. We develop a comprehensive framework that elucidates complex relationships among key parameters governing shale oil production, capturing the intricate dynamics of oil shale composition, reactive processes, and production indicators. Our method offers unprecedented insights into subsurface mechanics, demonstrating significant improvements in predicting oil yield and quality under various extraction conditions. Notably, this approach enables the identification of optimal operational parameters for maximizing energy efficiency and minimizing environmental impact in oil shale utilization. The integration of domain-specific knowledge enhances the framework’s ability to generate physically meaningful insights, bridging the gap between data-driven predictions and chemical engineering principles. Our findings contribute to the broader goal of optimizing fossil energy use while supporting the transition to more sustainable energy systems. This research not only advances the field of energy chemistry but also demonstrates the potential of innovative systems in addressing complex challenges in fossil fuel utilization, carbon management, and energy conversion technologies. Our relevant code can be utilized at https://github.com/AmbitYuki/Machine-Learning/tree/main/H-SRSF.