基于可解释性机器学习的ECD敏感性分析与预测技术  

ECD Sensitivity Analyses and Prediction Based on Interpretable Machine Learning

作　　者：马磊[1] 周波[1] 张宁俊 杨恒 蔡新树 刘征 徐同台[3] MA Lei;ZHOU Bo;ZHANG Ningjun;YANG Heng;CAI Xinshu;LIU Zheng;XU Tongtai(PetroChina Tarim Oilfield Company,Korla,Xinjiang 841000;Kunlun Digital Technology Co.,Ltd.,Beijing 102206;Beijing Shida Huyang Petroleum Technology Development Co.,Ltd.,Beijing 102206)

机构地区：[1]中国石油塔里木油田分公司,新疆库尔勒841000 [2]昆仑数智科技有限责任公司,北京102206 [3]北京石大胡杨石油科技发展有限公司,北京102206

出　　处：《钻井液与完井液》2023年第5期563-570,共8页Drilling Fluid & Completion Fluid

摘　　要：当量钻井液循环密度(ECD)计算过于繁琐耗时,且其敏感性规律尚未得到明确认识。为此,采用ProHydraulic软件对克深区带的1928个数据点进行分析,以确定ECD的理论值,并建立了相关的特征参数。同时,利用可解释性机器学习方法SHAP对钻井液特性、钻进参数和环空容积等关键因素进行了敏感性分析。最终,利用线性回归构建克深区带计算ECD的经验公式,涵盖了12个主要特征参数。结果表明,该模型表现优异,测试集决定系数达到0.963,平均绝对误差仅为0.04,为实际工程应用推出了简明、高效的经验公式。The calculation of equivalent circulation density(ECD)of drilling fluids is very cumbersome and time-consuming,and the pattern of sensitivity of the ECD has not been quite understood yet.To solve this problem,1928 data from the Block Keshen were acquired and analyzed using the ProHydraulic software to calculate the theoretical ECD.Based on the calculation some characteristic parameters were determined.Key factors affecting the sensitivity of the ECDs of a drilling fluid,such as mud properties,drilling parameters and sizes of the annular spaces are analyzed using interpretable machine learning technology.Using linear regression,an empirical equation for calculating the ECDs of drilling fluids in the Block Keshen,which covers 12 main characteristic parameters,is constructed.ECD calculation with the model shows that the model gives results that fit the practical values with excellence.The coefficient of determination of the testing set is 0.963,and the average absolute error is only 0.04,indicating that this empirical equation is simple and efficient in ECD calculation in practical engineering application.

关 键 词：ECD 敏感性分析 可解释性机器学习 线性回归模型 

分 类 号：TE254[石油与天然气工程—油气井工程]