关键词: 油气大模型；油气生产；勘探开发；数字化；智能化；信息交互

Abstract: To promote the transition of corporate digitalization to intelligence, Shengli Oilfield was committed to accelerate R&D and application of large language model technology in the area of exploration and development. In 2023, it successfully launched the “Shengxiaoli” intelligent assistant online to serve thousands of thousands of users safely. This article briefs about the AI and large language model conception as well as the current development conditions of large language model in the oil and gas sector. Focusing on some bottlenecks in the construction process of the large language model, such as integration and processing of a huge magnitude of multi-source heterogeneous data and a high demand for calculation resources and training efficiency, the “Shengxiaoli” large language model adopts the latest adaptation technology in construction of specialized knowledge, configuration of system technology and optimization of training and reasoning speed. It also learns millions of millions of specialized language materials and knowledge in the oil and gas area, creates the oilfield information assets intelligent service platform based on a large model with billions of natural language resources, and brings about application of high-efficiency exchanged data and sharing of specialized knowledge. Based on operation of the tailored knowledge and construction of the specialized large language model assistant, a dozen of large language models for virtual expert application are established to save users a large amount of time for writing, investigation, consultation of the data system, obviously improving work efficiency. This is a meaningful effort for realization ion of the“ one model for one enterprise and one assistant for one person” goals.

Key words: large oil and gas language model; oil and gas production; exploration and development; digitalization; intelligence;exchange of information