测井技术2025，Vol.49Issue(5)：696-703,8.DOI:10.16489/j.issn.1004-1338.2025.05.005
杭锦旗致密气层测井智能识别方法研究
Intelligent Recognition Method of Tight Gas Reservoirs in Hangjinqi Gas Field
摘要
Abstract
The tight gas reservoir of Hangjinqi exhibits small differences in logging responses between gas layers and water layers.To address the fluid identification challenges in this area,the difficulties in identifying gas layers through logging are thoroughly analyzed.A dataset is created using 911 depth points from 51 test layers in 35 wells,and a support vector machine(SVM)based intelligent fluid identification model is developed using logging curves.To further enhance the model's accuracy,a set of composite sensitive identification indices is constructed based on logging interpretation theory and the response characteristics of logging curves to fluid properties.This added additional input features to the intelligent model,establishing a"knowledge+data"dual-driven intelligent logging fluid identification model.The precision of the intelligent model under different input features is compared,with the results showing:①Effective samples are selected through data processing,with 531 samples from 23 wells designated as the training set and 380 samples from 12 wells as the testing set.The separation of training and testing sets from different wells ensured the model's generalization capability.②The SVM model with logging curves as inputs achieved a prediction accuracy of 81.1%on the testing set when the penalty factor is 900 and the kernel function parameter is 0.001.③The"knowledge+data"dual-driven model(incorporating 7 logging curve features and 5 composite features)improved the testing set prediction accuracy to 86.3%when the penalty factor is 57 and the kernel function parameter is 0.000 1.The gas layer identification accuracy increased from 87.0%to 90.2%,while the dry layer identification accuracy improved from 74.5%to 92.7%.④The intelligent model is applied to 60 actual wells in the study area,achieving a fluid identification compliance rate of 89.4%,demonstrating excellent application results.The methodology developed in this study,including data construction,sensitive feature development,model optimization,and evaluation,provides valuable references for well logging evaluation in other complex reservoirs.关键词
致密气层/测井解释/支持向量机/流体识别/"知识+数据"双驱动/智能模型Key words
tight gas layer/well log interpretation/support vector machine/fluid property identification/"knowledge+data"dual-driven/intelligent model引用本文复制引用
张军..杭锦旗致密气层测井智能识别方法研究[J].测井技术,2025,49(5):696-703,8.基金项目
中国石油化工股份有限公司科技攻关项目"基于人工智能的测井评价方法研究与应用"(P21006) （P21006）
