川东北元坝地区长兴组岩相及裂缝表征预测  

Characterization and prediction of lithofacies and fractures in Changxing Formation, Yuanba area, northeastern Sichuan Basin

作　　者：王浩[1] 陈志强 刘远洋[1] Wang Hao;Chen Zhiqiang;Liu Yuanyang(Exploration and Development Research Institute,Sinopec Southwest Oil&Gas Company,Chengdu,Sichuan 610041,China;Xinjiang Branch,CNPC Logging Company Limited,Changji,Xinjiang 831511,China)

机构地区：[1]中国石化西南油气分公司勘探开发研究院 [2]中国石油集团测井有限公司新疆分公司

出　　处：《天然气勘探与开发》2018年第3期26-32,共7页Natural Gas Exploration and Development

基　　金："十三五"国家科技重大专项"超深层复杂生物礁底水气藏高效开发技术"(编号:2016ZX05017-005)

摘　　要：川东北元坝地区二叠系上统长兴组储层具有储层薄、非均质性强、裂缝发育且类型多样的特点。常规叠后波阻抗反演技术难以展示岩性非均质性及识别薄储层,因此,在叠后波阻抗反演基础上,充分利用叠后地震数据及井点处的岩相信息,在井点处建立地震数据与岩性的对应模式,引入深度神经网络预测技术,将井点岩相训练结果推广至井间,再通过贝叶斯判别其对应的岩性概率,实现井间岩性预测;对比相干法裂缝预测,开展了最大似然法裂缝预测,计算每个样点的相似性,保留最小相似值。结果表明:(1)深度神经网络技术预测岩相与实钻吻合度高,能提高储层横向分辨能力;(2)最大似然法预测裂缝与相干法预测裂缝相比,可展示裂缝发育细节,表征中小尺度裂缝,与实钻结果具有更高的吻合度。With variously developed fractures,some reservoirs of Upper Permian Changxing Formation,Yuanba area,northeastern Sichuan Basin,are featured by small thickness and strong heterogeneity.However,because those conventional post-stack wave impedance inversion technologies can't reveal strong heterogeneity and identify thin reservoirs,they are not suitable for these reservoirs.So,based on post-stack wave impedance inversion,a corresponding model between seismic data and lithology was established by making full use of post-stack seismic data and lithofacies information.Then,a technology of deep neural network prediction was introduced to extend lithofacies training results at well point to certain area among wells.Afterwards,the corresponding lithology probability was discriminated by means of Bayesian technique and consequently inter-well lithology prediction was realized.Finally,the maximum likelihood method was adopted for fracture prediction and compared with coherence method.By means of the former method,the similarity of each sample point was calculated to keep the minimum similarity value.Results show that,(1)lithofacies predicted by deep neural network technology is highly coincident with actual drilling,and this technology can improve lateral reservoir resolution;and(2)the maximum likelihood method can reveal more details on fracture development and characterize middle-and small-sized fractures,and its prediction results are of higher coincidence with drilling.

关 键 词：深度神经网络 岩相预测 最大似然 相干 裂缝预测 晚二叠世 

分 类 号：P618.13[天文地球—矿床学]