Enhancing Large Language Models with In-Context Learning for Reservoir Management Tasks

With the fast-paced advancements in generative AI, large language models (LLMs) are becoming essential tools in assisting our daily tasks. While these models are being exploited in various fields, it is important to assess their potential in oil and gas applications. This paper demonstrates enhancing a foundation model and evaluates the resulting capabilities in addressing problems in reservoir management. An LLM is adapted using in-context learning (ICL) on simulation data of a benchmarking reservoir model. The resulting model is evaluated through several tasks including forecasting future pressure and oil saturation spatial distributions, forecasting future production rates of different development scenarios, and optimizing the placement of a new well. Results are discussed and validated using numerical simulation. The paper also highlights the best practices and challenges in enhancing and using LLM models for reservoir management problems.