克拉玛依油田七区砂砾岩油藏智能岩性识别

Intelligent lithologic identification of sandy conglomerate reservoirs in District No.7 of Karamay oilfield

中国新疆克拉玛依油田七区砂砾岩油藏岩性多样,隔夹层发育,常规取心识别方法成本高,在非取心段识别精度低,导致储层划分困难.为实现岩性的快速准确识别,根据地质资料将研究区内岩性划分为泥岩、粗砂岩、中-细砂岩、粗砾岩、中-细砾岩和煤层6种.基于敏感性分析,优选测井参数提取主成分,建立岩性识别图版,识别准确率达81.37%.针对不均衡样本导致的少数类识别率低的问题,提出一种基于k均值聚类人工少数类过采样(k-means synthetic minority oversampling technique,KMSMOTE)与随机森林结合的智能岩性识别模型,通过对少数类样本过采样提升识别精度,该模型的识别准确率达到92.94%.将图版法和KMSMOTE-随机森林应用于邻井进行岩性识别并对比分析结果发现,KMSMOTE-随机森林识别准确率为95.71%,优于图版法的82.91%.同时,对各类岩性的识别准确率均高于传统的随机森林模型,证明KMSMOTE和随机森林结合的智能岩性识别模型在不均衡岩性样本识别问题上具有较好的适用性,泛化能力强,能够快速、准确地识别地层岩性.研究结果为不均衡岩性样本识别提供了智能化新思路.

The sandy conglomerate reservoirs in Karamay are characterized by diverse lithology and interlayers.The cost of the conventional coring methods is high,and the identification accuracy in non-coring section is low,which leads to difficulty in reservoir classification.In order to achieve rapid and accurate identification of lithology,the lithology of the target area is classified into mudstone,coarse sandstone,medium-fine sandstone,coarse conglomerate,medium-fine conglomerate,and coal seam based on geological data.Firstly,the principal component analysis method is adopted to establish the lithology identification cross plot based on sensitivity analysis of well log data with an accuracy rate of 81.37%.Secondly,a lithology identification model is proposed based on the combination of k-means synthetic minority oversampling technique(KMSMOTE)and random forest to improve minority identification accuracy.The model improves the identification accuracy to achieve the accuracy of about 92.94%for oversampling minority samples.The two methods are applied to adjacent wells for comparative analysis of lithology identification.The accuracy of the KMSMOTE random forest is 95.71%,better than that of the cross plot method of 82.91%.The accuracy of minority sample identification is higher than that of the traditional random forest model,with excellent applicability and generalization ability on imbalanced lithology identification.The research results provide a new intelligent idea for imbalanced lithology sample identification.