WOA优化LightGBM在火成岩测井岩性识别中的应用  

Application of WOA optimized LightGBM in lithology identification of igneous logging

作　　者：冯欢 张国强 曹军 任宏 万文春 刘迪仁[1] FENG Huan;ZHANG GuoQiang;CAO Jun;REN Hong;WAN WenChun;LIU DiRen(Key Laboratory of Exploration Technologies for Oil and Gas Resources(Yangtze University),Ministry of Education,Wuhan 430100,China;Tianjin Branch of CNOOC Ltd.,Tianjin 300452,China;The Fifth Oil Production Plant of Changqing Oilfield,PetroChina,Xi'an 710000,China)

机构地区：[1]长江大学,油气资源与勘探技术教育部重点实验室,武汉430100 [2]中海石油(中国)有限公司天津分公司,天津300452 [3]中石油长庆油田第五采油厂,西安710000

出　　处：《地球物理学进展》2025年第1期230-242,共13页Progress in Geophysics

基　　金：国家重点研发计划项目(2018YFC060330502)资助.

摘　　要：渤海南部莱州湾地区的火成岩岩性复杂多变,常规测井交会图识别效果差.为提升该地区火成岩岩性识别精度,结合全局优化能力强的鲸鱼优化算法(WOA)和高效的轻量级梯度提升机(LightGBM),提出了基于WOA-LightGBM的火成岩测井岩性识别方法.首先,通过分析岩性的测井响应特征,选择岩心和薄片等地质资料完整、常规九条测井曲线齐全的测井数据作为样本集;然后将样本集输入到WOA-LightGBM、WOA-AdaBoost、WOA-SVM、LightGBM、AdaBoost、SVM六种模型中进行识别,并将识别结果进行对比验证;最后将识别模型应用于15口井中.研究结果表明:当鲸鱼种群为50时,最佳超参数下的WOA-LightGBM模型的识别精度最高、泛化能力最好,在样本集中识别准确率达91.62%,ROC-AUC为0.9676,实例井中整体解释符合率达85%.WOA-LightGBM可作为利用测井曲线智能识别渤海火成岩岩性的有效方法,并为其他类似区块的火成岩岩性识别提供参考.The igneous rocks in Laizhou Bay,Southern Bohai Sea,exhibited complex and variable lithologies,posing significant challenges to accurately identify lithology using conventional logging cross-plots.To improve the precision of igneous rock lithology identification in study block A,a high-efficiency Light Gradient Boosting Machine(LightGBM)model was employed to identify lithology.Furthermore,the utilization of a greater number of hyperparameters by LightGBM necessitated the employment of the Whale Optimization Algorithm(WOA),which was renowned for its robust global optimization capabilities,to identify the optimal parameter solution.Consequently,a logging lithology identification approach was proposed based on WOA-LightGBM.Firstly,logging response of lithology was analyzed,and logging data with complete geological information,such as core and thin section,and complete regular nine logging curves were selected as the sample set.The sample set is then input into six models,namely,WOA-LightGBM,WOA-AdaBoost,WOA-SVM,LightGBM,AdaBoost,and SVM,for identification.And the results of identification process were compared and verified.Finally,the recognition models were applied to 15 wells.The results demonstrated that WOA-LightGBM model with optimal hyperparameters exhibited the highest recognition accuracy and the most robust generalization ability when the whale population was 50.The recognition accuracy in the sample set reached 91.62%,and macro-average F1-score was 87.41%,ROC-AUC was 0.9676,PR-AUC was 0.8726,Matthews Correlation Coefficient was 0.8902,and 0.3401 for Cross-entropy Loss.Thus,the WOA-LightGBM method can be employed as an effective means of intelligently recognizing the lithology of the igneous rocks in the Bohai Sea by utilizing logging curves.This approach can also serve as a reference for igneous lithology identification in other similar blocks.

关 键 词：莱州湾 火成岩 测井岩性识别 轻量级梯度提升机 超参数 鲸鱼优化算法 

分 类 号：P631[天文地球—地质矿产勘探]