人工智能大模型在油气勘探开发领域的应用及挑战  

Application of Artificial Intelligence Pretrained Foundation Models in Exploration and Development and Challenges in This Area

作　　者：杨明澔 李小波[1,2,3] 刘兴邦 曾倩 Yang Minghao;Li Xiaobo;Liu Xingbang;Zeng Qian(AI Research Center,PetroChina Research Institute of Petroleum Exploration&Development,Beijing 100083,China;Artificial Intelligence Technology R&D Center for Exploration and Development,CNPC,Beijing 100083,China;National Key Laboratory for Green Mining of Multi-resource Collaborative Continental Shale Oil,Daqing 163712,China)

机构地区：[1]中国石油勘探开发研究院人工智能研究中心 [2]中国石油天然气集团有限公司勘探开发人工智能技术研发中心 [3]多资源协同陆相页岩油绿色开采全国重点实验室

出　　处：《石油科技论坛》2024年第6期107-113,125,共8页PETROLEUM SCIENCE AND TECHNOLOGY FORUM

基　　金：国家重点研发计划“二氧化碳驱油封存中的人工智能技术与应用研究”(编号:2023YFE0119600);中国石油天然气集团有限公司“十四五”重点科技项目“油气勘探开发人工智能关键技术研究”(编号:2023DJ84-06)。

摘　　要：自2022年底ChatGPT发布以来,人工智能正在从弱人工智能走向以大模型为代表的强人工智能阶段,大模型能力快速提升,行业场景落地应用成为产投研各方关注热点。通过调研语言、图像和多模态大模型在油气勘探开发领域知识管理、地震解释、实验图像分析、生成数据分析、设备预测性维护等方面的应用情况,显示人工智能大模型在油气勘探开发领域应用需求巨大。受油气勘探开发领域业务和数据特征影响,人工智能大模型的落地应用还面临专业数据融合、领域认知深度和大模型应用安全等多方挑战。随着人工智能大模型技术在多模态能力、上下文学习、智能体及具身智能等方面取得的新进展,有望进一步提升其在解决油气勘探开发领域复杂任务的能力,加速油气上游业务的数字化转型和智能化发展。Artificial intelligence(AI)has developed towards the powerful stage characterized for pretrained foundation models from the weak one since ChatGPT was released at the end of 2022.With the capability of pretrained foundation models rising rapidly,actual application of industrial scenarios has become a focus of concerns from the production,investment and research communities.Base on investigation of languages,images and multi-model state,pretrained foundation models can be used in the oil and gas exploration and development area for knowledge management,seismic interpretation,experimental image analysis,generation of data analysis and predictive maintenance of equipment.Obviously,the demand for application of AI pretrained foundation models is huge in the oil and gas exploration and development area.Influenced by the characterizations of oil and gas business and data,actual application of AI pretrained foundation models still faced some challenges,such as integration of specialized data,depth of domain understanding and safety for application of pretrained foundation models.With the new progress made in the areas of multi-model state capability,context learning,Agent and embodied artificial intelligence,AI pretrained foundation model technology will hopefully improve the ability for settlement of complicated tasks in the oil and gas exploration and development area and accelerate digital transformation and intelligent development of the oil and gas upstream business.

关 键 词：人工智能 大模型 勘探开发 应用场景 技术挑战 研发进展 

分 类 号：TE319[石油与天然气工程—油气田开发工程] TP18[自动化与计算机技术—控制理论与控制工程]