CISFA: A Decision-Support Agent Framework and its Allied Implementation with Generated AI in Oil and Gas Industry

With a rapid development of artificial intelligence generative content (AIGC), a set of human-machine interactive models have been changed. However, with limited understanding of the purposes of industry, even though fined-tuned by the professional data. The manuscript proposes a multi-role, self-closed-loop intelligent agent collaborative system framework (CISFA) that can compensate for the shortcomings of LLM in professional semantic understanding, multi-round self-interaction, and judgment and decision-making application scenes based on feedback and self-supervision between multiple role agents in multi-round Q&A based decision-making scenarios. Meanwhile, feasibility of applying medium-sized LLMs to aforementioned industry scenarios to achieve performance similar to that of very large-scale base models have also been considered. Through joint application with AIGC large models in three standard industry scenarios: drilling well control, device asset operation and maintenance management, as well as refining device operation guidance searching, it is proven that CISFA agent framework is effective in reducing the engineering application threshold of large models, simplify the prompt process and interpretation of industry mechanisms, and reducing application costs since medium-sized LLM been proven to show similar performance as very-large LLM by the allied application with CISFA.