SVR和AdaBoosting算法在孔隙型碳酸盐岩储层孔隙度解释中的应用  

Application of SVR and AdaBoosting arithmetic in porosity interpretation of porous carbonate reservoir

作　　者：王迪 程洪亮[1] 丁蔚楠[1] 李定军 刘昊年 WANG Di;CHENG HongLiang;DING WeiNan;LI DingJun;LIU HaoNian(Exploration and Development Research Institute,Sinopec Southwest Oil and Gas Company,Chengdu 610041,China;Geology Experiment Center,Sinopec Southwest Oil and Gas Company,Chengdu 610081,China)

机构地区：[1]中国石油化工股份有限公司西南油气分公司勘探开发研究院,成都610041 [2]中国石油化工股份有限公司西南油气分公司地质中心实验室,成都610081

出　　处：《地球物理学进展》2023年第6期2576-2587,共12页Progress in Geophysics

基　　金：国家科技重大专项“川西山前带潮坪相白云岩储层地震预测应用研究”(2017ZX05005-004-010)资助。

摘　　要：孔隙度是地层评价、储量研究中的重要参数,对于四川盆地C气田雷四上亚段气藏来说,由于矿物组分多样、密度曲线受扩径影响较大、开发井测井资料丰富程度不足等原因,导致使用体积物理模型方法求取孔隙度时,在开发井模型搭建、解释精度和效率等方面,都难以获得满意的结果.为更准确地计算孔隙度,在敏感测井响应分析基础上,尝试使用了SVR和AdaBoosting算法:SVR算法将低维度数据映射到高维空间,满足了把与孔隙度呈复杂非线性关系的电阻率曲线纳入至模型中的需要,和常规的多元、多项式回归相比,提升了模型准确度和稳定性;AdaBoosting算法采用了“集成学习”的思维,通过对简单模型进行迭代,将多个形式相同的简单模型提升为一个复杂的学习器,从而克服了使用单一模型算法灵活性差、精度低的缺点.使用上述方法进行孔隙度解释,并从误差和分布范围两个角度对模型进行了评价,认为相较于传统的体积物理模型,上述算法结果具有更高的精度、更强的稳定性,更能满足储层评价的需要.Porosity is not only the key research goal of logging interpretation,but also an important parameter in formation evaluation and reserves research.For the gas reservoirs in the upper submember of Leikoupo Formation in C gas field of Sichuan Basin,due to the diversity of skeleton mineral components,the large influence of exploration well density curve by diameter expansion,and the insufficient abundance of development well logging data,it is difficult to obtain satisfactory results in the construction of development well model,interpretation accuracy and interpretation efficiency when using the optimization method based on volume physical model to calculate porosity.Based on the above problems,on the basis of detailed analysis of sensitive logging response,a statistical model-based machine learning algorithm(SVR,AdaBoosting)is proposed to calculate porosity:SVR algorithm can map low-dimensional data to high-dimensional space,which meets the need of bringing resistivity curve that has complex non-linear relationship with porosity into the model,at the same time,precision and stability is improved compared with conventional multiple regression and polynomial regression.Adaboosting algorithm adopts the idea of ensemble learning which integrates and upgrades multiple simple models in the same form to a complex learner by the method of iteratively iteration of simple models,thus overcoming the shortcomings of poor flexibility and low accuracy of single simple model algorithm.Above two methods are used for porosity interpretation,and the model is evaluated from the perspectives of error and distribution range:it is believed that compared with the traditional volume physical model,the results of machine learning algorithm have higher accuracy,stronger stability and higher flexibility,which can better meet the needs of formation evaluation and reserves research.

关 键 词：孔隙型储层 孔隙度 支撑向量回归 集成学习 AdaBoosting算法 

分 类 号：P631[天文地球—地质矿产勘探]