一种辅助决策智能体及与生成式AI的联合应用

A Decision Support Agent and Allied Applications with Artificial Intelligent Generative Content

现阶段,受限于数据的私域属性、数据治理品质的差异,垂直领域细分场景较高的复杂机理和专业知识门槛,以及较高的后服务成本,面向通识应用的生成式人工智能大模型或经小规模数据微调的大模型,仍难以在工业制造、能源生产、资产管理等场景下,解决用户希望结合现象,即时获得相应的推理决策结论,并实现一定程度的自主持续学习提升的需求。为此,该文提出了一种适用油气行业的自监督辅助决策智能体系统,基于多个不同角色智能体间的反馈和自监督,在问答决策类场景中,弥补大模型在专业语义理解、多轮自交互、判断决策及自主优化方面的短板,同时探索应用中等规模预训练大模型,面向垂直场景,实现类似超大规模基座模型表现的可行性。通过在钻井井控、装置资产运维管理、炼化装置操作指引三个高频场景中与AIGC大模型的联合应用,论证了该方法对降低大模型的工程应用门槛、可解释、降低应用成本等方面的有效性。

At this stage,artificial intelligence generative content(AIGC)based large models that have undergone professional data fine-tuning still have difficulty in addressing users'needs for immediate reasoning and decision-making in industrial manufacturing,energy production,asset management,and other scenarios due to the private attributes of data asset privacy,variation of quality of data administration and processing,expertise entrance regarding to industrial mechanism as well as complicate continuous operation after AIGC model been deployed.We propose a multi-role,self-closed-loop intelligent agent collaborative framework that can compensate for the shortcomings of large models in professional semantic understanding,multi-round self-interaction,and decision-making based on feedback and self-supervision between multiple different role agents in question-answering decision-making scenarios.At the same time,it explores the feasibility of applying medium-sized large-scale AI models to vertical scenarios to achieve performance similar to that of very large-scale base models.Through allied application with AIGC large models in three high-frequency scenarios:drilling well control,device asset operation and maintenance management,and refining device operation guidance,it is proven that the proposed method is effective in reducing the engineering application threshold of large models,explaining and reducing application costs.