基于核贝叶斯判别法的储层物性参数预测  

Reservoir physical property prediction based on kernel-Bayes discriminant method

作　　者：刘兴业 陈小宏 李景叶 周林 郭康康 

机构地区：[1]中国石油大学油气资源与探测国家重点实验室中国石油大学海洋石油勘探国家工程实验室,北京102249

出　　处：《石油学报》2016年第7期878-886,共9页Acta Petrolei Sinica

基　　金：国家自然科学基金项目(No.U1262207);国家重大科技专项(2016ZX05033-003-008)资助

摘　　要：随着油田开发的持续深入,地震勘探技术在储层预测及储层描述方面的要求不断提高。储层的物性参数是描述储层特征的主要参数,但由于影响储层物性参数的因素众多,且关系复杂,为储层物性参数的准确预测带来了巨大困难。基于传统贝叶斯判别的物性参数预测方法能够综合考虑多种参数,在获得预测结果的同时能够给出预测结果的概率分布,从中提取最大后验概率,并对预测结果的不确定性进行定量评价。但在预测过程中条件概率密度函数比较难以估计,一般假设各参数服从特定分布,但当数据分布比较复杂时,不满足这种假设,限制了其应用效果。因此,基于贝叶斯定理,采用核函数估算的方法计算条件概率密度函数,提出了基于核贝叶斯判别法的储层参数预测方法。该方法不需要假设数据服从特定的分布,采用非参数估计方法获取条件概率密度函数,可以计算获得物性参数的最大后验概率,实现了多种物性参数的预测并提供预测结果的置信概率,可用于进行不确定性评价。模型数据和实际资料的应用效果很好地验证了该方法的有效性。该方法在储层物性参数预测、储层描述中有良好的应用前景。With the constant deepening of oil field development, it is also required to continuously improve the seismic exploration technology in reservoir prediction and description. Reservoir physical properties are the main factors for describing the reservoir characteristics. However, the diversified factors impacting reservoir physical properties with complicated relationships lead to a great difficulty in accurately predicting reservoir physical properties. The physical property prediction method based on traditional Bayes discriminant can not only take various parameters into account, but also present the probability distribution of prediction results while providing prediction results, so as to extract the maximum posterior probability value and quantitatively evaluate the uncertainties in prediction results. However, it is difficult to estimate the conditional probability density function in the prediction process. Generally, it is assumed that various parameters are subject to a specific distribution. This assumption cannot be satisfied duo to the complicated data distribution, thus limiting its application effect. Therefore, based on Bayes theorem, the conditional probability density function is calculated using the kernel function estimation method. On this basis, the reservoir parameter prediction method based on the kernel-Bayes diseriminant method is put forward. For this method, it is not required to assume that the data are subject to a specific distribution. Using non-parameter estimation method, the conditional probability density function can be obtained to further calculate the maximum posterior probability of physical property, thus achieving the prediction of multiple physical property parameters and providing the cotafidence probability of prediction results for uncertainty evaluation. The effectiveness of this method can be well validated by the application effect of model data and actual data, so that this method has a favorable application prospect in the prediction of reservoir physical pr

关 键 词：核函数 贝叶斯判别 条件概率 储层物性参数 不确定性 

分 类 号：P631.4[天文地球—地质矿产勘探]