基于密度聚类的K近邻法在储层流体识别中的应用

摘要

Reservoir fluid identification is an indispensable link in logging interpretation. In order to remove the defects of tra-ditional approaches, such as unsatisfying accuracy, excessive computation, undue dependence on personal experience, a density clustering based K-nearest neighbor method was proposed. According to the spatial distribution of the interval logging data under test, data clusters are formed based on relative density. And then with K-nearest neighbor voting method, the cat-egories of all clusters become available. Comparing with other commonly used identification methods, tested on the carbonate reservoir of Ordovician Yingshan Formation in an oil field, this approach shows a high accuracy, strong generalization and ro-bustness, as well as better effects on oil-water layer identification which is usually difficult for the compared methods. The method has a good application prospect and provides a new thought on solving complex problems in oilfield exploration and development with data mining methods.%针对传统储层流体识别方法识别精度低、运算量大、过于依赖个人经验的缺点，提出基于密度聚类的K近邻法，根据待测层段测井数据的空间分布规律，将样本按相对密度聚类成数据簇，并利用K近邻投票获得各簇所属类别。将该方法应用在某油田奥陶系鹰山组碳酸盐岩储层识别中。结果表明，较之其他常用识别方法，该算法识别精度高，泛化性和鲁棒性强，在处理大数据分类问题时具有明显优势，且在识别常规方法难以识别的油水同层时取得了较好的效果，具有良好的应用前景，为利用数据挖掘方法解决油田勘探开发中的复杂问题提供了新思路。