基于机器学习的川中地区雷口坡组三段二亚段泥质灰岩储层分布预测OA北大核心

Recently,an industrial gas flow has been obtained from the second sub-member of the third member of the Leikoupo Formation(Lei32)at Well CT1 in the central Sichuan Basin,indicating favorable exploration prospects for marly limestone reservoirs.However,research on well-logging response mechanisms and predictive modeling for such reservoirs remains limited.This study integrates core descriptions,thin-section petrography,and a well-log response analysis to characterize the lithological features of the Lei32 sub-member.A long short-term memory(LSTM)machine learning model was established using well-log data for lithofacies identification of marly limestone.By integrating a single-factor mineral content analysis and sedimentary facies characteristics,a multi-scale predictive framework(point-line-surface)was constructed to forecast marly limestone reservoirs.The results show that the Lei32 sub-member is primarily composed of limestone,marly limestone,dolomite,and gypsum salt rocks.The marly limestone reservoir is characterized by nanoscale to micrometer-scale pores and microfractures,indicating a typical low-porosity,low-permeability system,with reservoir thickness ranging from 40 to 130 m.Compared with CIFLog software calculations and petrographic identifications,the LSTM model achieved a prediction accuracy of(87.3±0.5)%.Spatial prediction results indicate that the reservoir's thickness ranges from 80 to 120 m in the Xichong,Nanchong,and Yilong areas,which are favorable for exploration.In contrast,Zhongjiang,Ziyang,Anyue,and Hechuan exhibit reservoir thicknesses of 60 to 80 m,suggesting potential exploration targets.This study provides useful a reference for the hydrocarbon exploration of marly limestone reservoirs in the Lei32 sub-member of the central Sichuan Basin.