基于多元回归模型的钻头前方地层速度预测方法  

Prediction of layer velocity ahead of bit based on multiple regression model

作　　者：郭书生 朱江梅[2] 王世越 赵前华[2] 王晓飞[2] GUO Shusheng;ZHU Jiangmei;WANG Shiyue;ZHAO Qianhua;WANG Xiaofei(CNOOC China Limited,Hainan Branch,Haikou,Hainan 570311,China;Well Tech Department,China Oilfield Services Limited,Langfang,Hebei 065201,China)

机构地区：[1]中海石油(中国)有限公司海南分公司,海南海口570311 [2]中海油田服务股份有限公司油田技术事业部,河北廊坊065201

出　　处：《中国海上油气》2022年第3期49-54,共6页China Offshore Oil and Gas

基　　金：中海石油(中国)有限公司科技项目“钻头前地层压力随钻预测方法研究(编号:ZYKY-2020-ZJ-03)”;中海油田服务股份有限公司科技项目“异常压力地层速度及孔隙压力随机反演预测技术研究(编号:YJB21YF016)”部分研究成果。

摘　　要：为了应对中深地层高温、超高压等特点所带来的钻井工程风险,准确预测钻头前方压力变化界面深度是其中的关键点之一。因此钻头前方地层速度预测成为了一项重要的研究课题,备受国内外学术界和工业界关注。本文提出利用多元回归模型量化刻画客观因素对地层速度的影响,从而将钻头前方地层速度预测转化为基于非线性多元回归的有监督机器学习问题。主要步骤包括:首先基于VSP走廊叠加资料的多属性信息构建多元回归模型;然后利用反演来预测中频地层速度;再利用粒子滤波法来求解该多元回归模型的状态参数,该模型将中频地层速度等因素特征作为输入,通过求解从而有效地预测钻头前方地层速度。实际应用成效表明该方法可预测钻头前200 m的高精度地层速度,从而验证了本文方法的有效性和推广应用价值。In order to mitigate the risk of drilling operation when drilling in high temperature and ultra-high pressure middle-deep formation,it is necessary to accurately predict the pressure change ahead of the bit and the depth at which the pressure is changed.Therefore,prediction of the layer velocity ahead of bit has become an important study subject,which has attracted much attention of the personages from academic and industrial worlds in China and the World.In this paper,a multiple regression model is proposed to quantify the influence of objective factors on layer velocity,so as to transform the predicted layer velocity ahead of bit into a supervised machine learning item based on nonlinear multiple regression.The work flow of the model is as follows:at first,a multiple regression model is constructed based on the multi-attribute information of VSP corridor stack data;secondly,the inversion is used to predict the intermediate frequency layer velocity;then the particle filter algorithm is used to solve the state parameter of the multiple regression model,finally the feature parameters as intermediate frequency layer velocity are input in the model and the layer velocity ahead of bit can then be effectively predicted by solving of the model.The results of field application show that this method can predict the high-precision layer velocity of 200 m ahead of bit,thus verifying the effectiveness and good value for popularization of this method.

关 键 词：钻头前方地层速度 多元回归模型 有监督机器学习 VSP走廊叠加资料 粒子滤波法 

分 类 号：TE132.1[石油与天然气工程—油气勘探]