机理-数据融合的井下溢流、漏失风险智能诊断数据集

本数据集主要面向智能钻井技术研发，以及溢流与漏失诊断算法建模、复杂安全预警等研究需求建设。油气钻井中溢流、漏失是诱发井喷等事故的关键因素，而现有数据集存在规模有限、标注质量不均等问题，难以支撑鲁棒性模型构建，随着高频传感设备普及率提升，亟需高质量数据资源填补空白。该数据集由中国石油集团工程技术研究院有限公司于2020-2024年构建，基于钻井现场多类型传感器实时采集产生，涵盖地面机械参数传感器、流量压力等流体参数监测仪器、PWD测量工具等，采集系统符合工业通信标准并经过出厂、现场、定期三级标定。数据产生过程包含严格的质量控制：经缺失值插值、异常值检测、多通道同步校正、特征工程构造及平滑滤波等预处理，结合现场工程师交叉验证手动标注事件标签。主要内容涵盖井深、钻压、转速、立压、钻井液进/出口流量、钻井液密度、井口回压等50余项参数，衍生出ECD、压降等高阶特征，包含正常钻井、溢流、漏失三类标签（0为正常，1为溢流或漏失），空间覆盖井眼轨迹，时间精度达分秒级。数据体量为13.7MB。

This dataset is developed to support research on intelligent drilling technology R&D, kick and lost circulation diagnosis algorithm modeling, complex safety early warning and other related studies. Kicks and lost circulation in oil and gas drilling are key factors triggering accidents such as blowouts. However, existing datasets have problems including limited scale and uneven annotation quality, which makes it difficult to build robust models. With the growing popularity of high-frequency sensing equipment, high-quality data resources are urgently needed to fill this gap. This dataset was constructed by CNPC Engineering Technology Research Institute Co., Ltd. from 2020 to 2024. It is generated based on real-time data collected by various types of sensors at drilling sites, including surface mechanical parameter sensors, fluid parameter monitoring instruments (such as flow and pressure sensors) and PWD (Pressure While Drilling) measurement tools. The data acquisition system complies with industrial communication standards and has undergone three-level calibration: factory calibration, on-site calibration and regular calibration. The data generation process incorporates strict quality control: preprocessing operations such as missing value imputation, outlier detection, multi-channel synchronization correction, feature engineering construction and smoothing filtering are performed, and event labels are manually annotated with cross-validation from on-site engineers. The main content covers more than 50 parameters including well depth, weight on bit, rotational speed, standpipe pressure, drilling fluid inlet/outlet flow rate, drilling fluid density, wellhead back pressure and so on. It also derives high-order features such as ECD (Equivalent Circulating Density) and pressure drop. The dataset includes three types of labels: normal drilling, kick and lost circulation (0 represents normal, 1 represents kick or lost circulation). It spatially covers wellbore trajectories, with a temporal accuracy up to the second level. The total size of the dataset is 13.7 MB.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于井下溢流和漏失风险智能诊断的机理-数据融合数据集，主要服务于智能钻井技术研发和算法建模。它基于2020-2024年钻井现场多传感器实时采集，涵盖50余项参数，包含正常、溢流、漏失三类标签，数据体量13.7MB，以CSV格式提供，旨在填补高质量数据资源空白，支持鲁棒性模型构建。