【作者】 何健；

【导师】 文晓涛；

【作者基本信息】 成都理工大学 ， 地球探测与信息技术， 2020， 硕士

【摘要】 根据叠前、叠后地震数据求取的地震属性与流体识别因子被广泛地应用于储层预测中,如裂缝带识别、流体识别等。但是随着勘探程度的深入和地下地质条件的日趋复杂,这些方法存在的问题也因此日益突出,主要有:(1)根据叠前、叠后地震数据可以求取众多地震属性与流体识别因子,但是想要从中筛选出对所研究区域的储层有显著响应的地震属性与流体识别因子就需要大量的人工参与。(2)仅利用单一地震属性或流体识别因子进行储层预测通常会带来多解性问题。针对这些问题,本文将人工智能领域中的随机森林算法引入储层预测,该算法具有泛化误差小、抗干扰能力强、不易产生过拟合等特点,能有效增强储层预测、流体识别的准确率与稳定性。本文的主要研究内容和认识如下:(1)根据泰勒中值定理对生成决策树的运算过程进行了简化,详细地介绍了随机森林的原理及其在储层预测中的构建与执行方法。利用理论数据和测井数据分别对随机森林算法的分类性能进行验证,为多参数储层预测奠定了基础。(2)研究了基于随机森林算法的裂缝发育带预测。首先从地震资料出发,计算能有效表征研究区裂缝带的地震属性数据体。然后基于井旁道地震属性与测井裂缝解释结果选择特征参数,建立地震属性与裂缝发育带程度的对应关系。最后应用随机森林算法对裂缝带进行了综合预测。与井资料及地质资料对比分析可发现:在随机森林算法的支撑下,利用多属性(或参数)输入进行裂缝带预测,可有效地减少预测的多解性。(3)建立了基于随机森林算法的储层流体预测流程。首先以实验验证结果为依据利用测井数据分别计算出纵横波速度比、杨氏模量、泊松比、泊松阻抗和流体属性5个用于流体识别的因子。然后按照测井流体分布情况提取特征参数,建立5种流体识别因子与流体分布信息之间的对应关系。再基于叠前道集,分别求取储层的密度和纵、横波速度等岩性参数,并结合相关公式计算流体识别因子。最后,应用随机森林算法对储层含流体情况进行综合预测。实例表明,该方法预测的含气储层能与井资料吻合,说明利用多种流体识别因子进行储层综合预测能有效减少多解性。更多还原

【Abstract】 Seismic attributes and fluid identification factors obtained from pre-stack and post-stack seismic data are widely used in reservoir prediction,such as fracture zone prediction,fluid identification etc.However,with the deepening of exploration and the increasing complexity of underground geological conditions,the problems of these methods have become increasingly prominent.(1)Many seismic attributes and fluid identification factors can be obtained from pre stack and post stack seismic data.But,it requires a lot of manual participation to select the seismic attributes and fluid identification factors that have a significant response to the reservoir characteristics of the studied area.(2)The prediction of reservoir only based on single seismic attribute or fluid identification factor usually leads to multiple solutions.To solve these problems,this paper introduces the random forest algorithm,which has the characteristics of small generalization error,strong anti-interference ability,and difficulty in overfitting.And it can effectively improve the accuracy and stability of storage layer prediction and fluid identification.The main research contents and understanding here are as follows:(1)The Taylor Median Theorem is used to simplify the computation in the decision tree generation process of random forest algorithm.The basic principles of the random forest algorithm are introduced by introducing the construction and execution processes of the forest.Besides,the classification performance of the random forest algorithm is verified by theoretical data and drilling data respectively,which would lay a foundation for multi parameter reservoir prediction.(2)The prediction of fracture zone based on random forest algorithm is studied.Firstly,based on the post-stack seismic data,the seismic attribute data volumes which can effectively represent the fracture zone in the study area are calculated.Then,based on the seismic attribute of the sidetrack and the interpretation result of the logging fracture,the characteristic parameters are selected,and the corresponding relationship between the seismic attribute and the fracture development zone degree is established.Finally,the random forest algorithm is used to comprehensively predict the fracture zone.Compared with well data and geological data,it can be found that under the support of random forest algorithm,multi-attribute(or parameter)input can be used to predict fracture zone,which can effectively reduce the multi solution of prediction.(3)The flow of reservoir fluid prediction based on random forest algorithm is established.First,based on the experimental verification results,five factors used for fluid identification are calculated by using drilling data,respectively,the longitudinal wave velocity ratio,Young’s modulus,Poisson’s ratio,Poisson’s impedance and fluid properties.Then extract the characteristic parameters according to the drilling fluid distribution and establish the correspondence between the five fluid identification factors and the fluid distribution information.Then,based on pre-stack trace set,the lithology parameters such as reservoir density,P-wave and S-wave velocity are obtained respectively,and the fluid recognition factor was calculated by combining the relevant formula.Finally,the random forest algorithm is used to comprehensively discriminate the reservoir fluid.The example shows that the gas reservoir predicted by this method can be consistent with the well data,which indicates that the reservoir comprehensive prediction by using multiple fluid identification factors can effectively reduce the multiple solutions.更多还原