地震属性驱动的条件生成对抗网络沉积微相模型构建

打开文本图片集

中图分类号：TP391.41 文献标志码：A

Abstract：Duetothecomplexityand strong heterogeneityof stratigraphic structure，aswellasthelimitedavailabilityoflogging，core，andoil testing data，existing sedimentarymicrofacies modeling methods struggle toachieve acurateresults.To address this challenge，anew modeling approach basedon conditional generative adversarial networks（cGANs）was proposed.This methodutilizes greycorrelation analysis tocalculate thedegree ofcorelationbetween various seismicattributes andthesand-to-groundratio，thereby identifying atributes with strong predictiverelevance.These selectedseismicatribute imagesare thenusedas inputs toaconvolutional neural network，whichisemployed to constructaprediction model for the sand-to-groundratio.Theresulting predictions are visualizedasathermal map，which，combined with wellog phasediagrams，servesasajointconstraintfortraining thegenerativeadversarialnetwork．Basedonthis，asedimentarymicrofacies generation model is developed toenableaccurate modelingof sedimentarymicrofacies.This methodwasapplied toacase studyof an oilfieldin eastern China.Theresults demonstratethatthecGAN-based modelcan efectivelycapturecomplex geological patterns，achieving a well-point coincidence rate of 94.1%

Keywords：conditional generative adversarial network；deep learning；sedimentary microfacies；sand-to-groundratio；gre！ correlation；convolutional neural network

期川反阴段，佃气贝源八多分布仕发东、难川反时区域[3」，且受到测井、岩心、试油等数据不足的影响，现有沉积微相建模方法[4-7]难以实现精确建模。（剩余18008字）