An Enhanced Dual-Layer Ensemble Framework for Porosity Prediction Utilizing Conventional Well Logging Data

作     者：Lu Qiao Shengyu Yang Qinhong Hu Huijun Wang Taohua He 

作者机构：National Key Laboratory of Deep Oil and Gas， China University of Petroleum (East China) Laboratory for Marine Mineral Resources， Qingdao National Laboratory for Marine Science and Technology Hubei Key Laboratory of Petroleum Geochemistry and Environment (Yangtze University)， Yangtze University 

出 版 物：《Journal of Earth Science》 (地球科学学刊)

年 卷 期：2025年

核心收录：

学科分类：12[管理学] 081801[工学-矿产普查与勘探] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081802[工学-地球探测与信息技术] 081803[工学-地质工程] 081104[工学-模式识别与智能系统] 08[工学] 0818[工学-地质资源与地质工程] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supports provided by the SINOPEC (31450008-22-ZC0607-0007 33550007-22-ZC0613-0040) Shandong Natural Science Foundation of China (ZR2021QD092) Young Taishan Scholars (tsqn201909066) 

摘      要：Porosity is the one of the most fundamental and critical parameters for reservoir characterization and is closely related to the hydrocarbon storage capacity. The interpretation methods of conventional well logs have poor generalization capabilities to achieve the desired accuracy in porosity prediction. The increasing adoption of machine learning has ameliorated this predicament to some extent， but it remains difficult to reconstruct the complex and nonlinear relationship between logging responses and porosity by relying on individual machine learning models alone. This work proposes a two-layer model ensemble framework to predicte porosity， which includes base regressors in the first layer and meta regressor in the second layer. The results indicate that the Light Gradient Boosting Machine (LightGBM) - Categorical-Features Gradient Boosting (CatBoost) ensemble model provides the highest prediction accuracy for the given dataset and is applied for porosity evaluation at other logged depths. This work also points out that the base regressor should be accurate and complex， while the structure of the meta-regressor should be simple in order to achieve the desired performance. Furthermore， the proposed model ensemble framework is versatile and applicable to reservoir evaluation， engineering development， and other potential fields due to its effectiveness and convenience.