基于随钻测井数据预测破裂压力  

Prediction of Fracture Pressure Based on Logging-while-drilling Data

作　　者：郭大立[1] 王玉基 张小栓 辛骅志 康芸玮 GUO Da-li;WANG Yu-ji;ZHANG Xiao-shuan;XIN Hua-zhi;KANG Yun-wei(School of Science,Southwest Petroleum University,Chengdu 610500,China;Xinjiang Oilfield Company,Karamay 834000,China)

机构地区：[1]西南石油大学理学院,成都610500 [2]新疆油田公司,克拉玛依834000

出　　处：《科学技术与工程》2023年第5期1923-1930,共8页Science Technology and Engineering

基　　金：国家科技重大专项(2016ZX05042,2016ZX05042_003)。

摘　　要：破裂压力是判断岩石是否起裂的重要依据,而现有的随钻测井资料仅能解释地应力、弹性模量等参数,缺乏对破裂压力的解释。为了准确预测破裂压力并降低施工压力和风险,基于准格尔盆地MH区块的流体扫描成像(flow scanner image,FSI)测试的产出剖面测井资料,区分泡酸与不泡酸两种方式,分别建立广义回归神经网络(generalized regression neural network,GRNN)模型,对该区块油井开发进行破裂压力预测,并运用交叉验证方法得出光滑因子,通过与真实破裂压力值对比验证模型的准确性,并与反向传播(back propagation,BP)神经网络和Hubbert-Willis(H-W)模型的预测结果进行对比分析,再基于此预测给出泡酸建议。结果表明:GRNN模型预测结果与实际破裂压力更接近,且均方根误差为4.54%,平均百分比误差为0.03%,均优于BP神经网络和H-W模型。GRNN模型不受地质条件影响且预测精度高,操作简便,可用于该地区破裂压力预测,也可作为后续井FSI测试的替代,不但可以为同类地区的施工提供借鉴,而且可以为同地区开发资源节约成本。Fracture pressure is an important basis to judge whether the rock is fractured or not,but the existing logging-while-drilling(LWD)logging data can only explain the in-situ stress,elastic modulus and other parameters,and lack the explanation of fracture pressure.In order to accurately predict the rupture pressure and reduce the construction pressure and risk,the generated profile logs tested by the flow scanner image(FSI)instrument in the MH block of the Junge Basin,a generalized regression neural network(GRNN)model was established to predict the fracture pressure of oil well development in this block by differentiating the two methods,and the smooth factor was obtained by cross-validation method.The accuracy of the model was verified by comparison with the real rupture pressure,and the prediction results of back propagation(BP)neural network and Hubbert-Willis(H-W)model were compared.Based on the prediction,some suggestions of paulic acid were given.The results show that the predicted results of GRNN model are closer to the actual rupture pressure,and the root mean square error is 4.54%,and the average percentage error is 0.03%,both of which are better than BP neural network and H-W model.GRNN model is not affected by geological conditions,has high prediction accuracy,and is simple to operate.It can be used to predict the fracture pressure in this area,and can also be used as the replacement of FSI instrument flow scanning in subsequent wells.It can not only provide reference for the construction of similar areas,but also save cost for the development of resources in the same area.

关 键 词：随钻测井 产出剖面测井 交叉验证 泡酸分析 GRNN模型 破裂压力 

分 类 号：TE321[石油与天然气工程—油气田开发工程]