基于粒子群优化的相关向量机算法的井震联合储层参数预测与不确定性评估  

Seismic and well logs integration for reservoir parameter prediction and uncertainty evaluation based on relevance vector machine optimized by particle swarm optimization

作　　者：代仕明 李敏 唐金良 朱童 李京南 胡华锋 DAI Shiming;LI Min;TANG Jinliang;ZHU Tong;LI Jingnan;HU Huafeng(SINOPEC Geophysical Research Institute Co.,Ltd.,Nanjing,Jiangsu 211103,China)

机构地区：[1]中石化石油物探技术研究院有限公司,江苏南京211103

出　　处：《石油地球物理勘探》2023年第6期1436-1445,共10页Oil Geophysical Prospecting

基　　金：国家自然科学基金项目“海相深层油气富集机理与关键工程技术基础研究”(U19B6003)资助。

摘　　要：目前有岩石物理、地质统计学及地震多属性三类方法预测孔隙度和饱和度。第一类应用广泛且物理意义明确,但预测结果具有一定局限性。第二类可以获得较常规方法分辨率更高的预测结果,但在构造复杂区域预测储层参数存在一定困难。第三类中的支持向量机算法(Support Vector Machine,SVM)的计算复杂度随着样本增加而增加,且难以评估预测结果的不确定性;第三类中的相关向量机算法(Relevance Vector Machine,RVM)没有明确的理论指导核参数选取。为此,利用粒子群算法(Particle Swarm Optimization,PSO)指导核参数选取,在获取最优核参数基础上定量预测储层参数;同时考虑到变异系数可消除量纲的影响,引入变异系数评估预测结果的不确定性,结合逐步回归算法优选地震属性,提出了一种基于粒子群优化的相关向量机算法(PSO-RVM)的孔隙度与饱和度定量预测方法。数值模拟和实际数据应用结果表明:①PSO-RVM具有较好的学习性能和泛化能力,且具备一定的抗噪能力;PSO-RVM预测结果的均方根误差低于RVM,预测精度更高,说明PSO可以有效指导RVM核参数选择,进而提高算法性能。②PSO-RVM给出了预测结果后验概率,通过引入变异系数可量化不确定性。③以井震数据为基础,基于PSO-RVM定量预测了孔隙度与含气饱和度,预测精度较高,且孔隙度预测精度更高,不确定性更低。There are three types of methods for predicting porosity and saturation,which are rock physics,geostatis-tics,and seismic multi-attribute.The first type with clear physical meaning is widely used,but it has certain limita-tions.The second type can improve resolution compared with conventional methods,yet it is difficult to predict reservoir parameters in complex structural areas.The support vector machine(SVM)belongs to the third type.Its computational complexity increases with the rise of sample quantity.Meanwhile,it is difficult to evaluate the uncer-tainty.The relevance vector machine(RVM)in the third type lacks a clear theory for selecting kernel parameters.To improve this,particle swarm optimization(PSO)is applied to guide the selection of kernel parameters.The reservoir parameters are quantitatively predicted on the basis of obtaining the optimal kernel parameters.Then,the coefficient of variation is introduced to eliminate the influence of dimension and quantify the uncertainty of prediction results.With the help of a stepwise regression algorithm to screen seismic attributes,this paper proposes a quantita-tive porosity and saturation prediction method based on RVM optimized by PSO(PSO-RVM).The results of nu-merical simulation and field application show that:①PSO-RVM has good learning performance,satisfying genera-lization ability,and a certain ability of anti-noise.The RMS error of PSO-RVM prediction results is lower than that of RVM,and the prediction accuracy is higher,which indicates that PSO can effectively guide the selection of RVM kernel parameters and improve the algorithm performance.②PSO-RVM provides a posterior probability,and it can quantify uncertainty by introducing a coefficient of variation.③From seismic and well logs data,the porosity and gas saturation are quantitatively predicted by PSO-RVM with high prediction accuracy.Additionally,the accuracy of porosity prediction is higher,and the uncertainty is lower.

关 键 词：储层参数 地震属性 PSO RVM 变异系数 不确定性评估 

分 类 号：P631[天文地球—地质矿产勘探]