一种基于知识图谱和随机森林算法的致密气井产能预测方法  

A Productivity Prediction Method for Tight Gas Wells Based on Knowledge Graph and Random Forest Algorithm

作　　者：李文倚 侯明雨 全航 余杰 Li Wenyi;Hou Mingyu;Quan Hang;Yu Jie(CNOOC Research Institute Ltd.,Beijing 100028,China;Chengdu University of Technology,Chengdu,Sichuan 610059,China)

机构地区：[1]中国海油研究总院有限责任公司,北京100028 [2]成都理工大学,四川成都610059

出　　处：《特种油气藏》2024年第5期77-84,共8页Special Oil & Gas Reservoirs

基　　金：国家自然科学基金面上项目“页岩储层纳米孔隙结构表征及渗流机理研究”(51674044)。

摘　　要：气井产能预测受地质、工程等多种因素影响,传统的数学解析、数值模拟等方法难以快速准确预测致密气井产能。针对上述问题,基于大数据及机器学习的思想,创新性地融合了知识图谱和随机森林算法,形成了一种针对致密气井的产能预测方法。通过数据预处理对不同类型的基础数据进行规范化处理,采用实体识别和链接技术将不同数据源的实体整合到知识图谱中。使用关系抽取和建模技术,建立实体之间的关系和属性,形成完整的知识图谱,准确预测产能。在此基础上,依托随机森林机器学习算法建立致密气井产能预测模型,利用模型对秋林区块致密气井产能进行预测,预测精度达到89.7%。该方法可以在开发前期快速准确预测气井产能,大幅度提高预测的准确度,为致密气开发产能部署和高产井的发掘提供决策支持。The productivity prediction of gas well is influenced by various factors such as geology and engineering.Traditional methods like mathematical analysis and numerical simulation struggle to quickly and accurately predict the productivity of tight gas wells.To address this issue,an innovative method combining knowledge graph and the random forest algorithm is proposed based on big data and machine learning concepts to develop a productivity prediction method of tight gas wells.Data preprocessing standardizes different types of basic data,and entity recognition and linking technologies integrate entities from various data sources into the knowledge graph.Relationship extraction and modeling techniques are used to establish relationships and attributes among entities,developing a complete knowledge graph for accurate productivity prediction.On this basis,a productivity prediction model for tight gas wells is developed using the random forest machine learning algorithm,and the model predicts the productivity of tight gas wells in the Qiulin Block with an accuracy of 89.7%.This method allows for rapid and accurate productivity predictions in the early stages of development,significantly improving prediction accuracy and providing decision support for productivity deployment and high-yield well cultivation in tight gas development.

关 键 词：致密气 机器学习 知识图谱算法 随机操作算法 产能预测 

分 类 号：TE312[石油与天然气工程—油气田开发工程]