摘要
目前,尚未存在全面的断控岩溶型溶洞训练数据集用于深度学习建模。本文采用基于露头资料、地震数据、可靠的地质模型以及基于目标的方法研制了断控岩溶型溶洞原型模型,对不同来源的原型模型集进行组合、旋转、裁剪和优选操作来构建可靠且多样的断控岩溶型溶洞相训练数据集,同时构建相应的虚拟井和概率体训练数据集,作为训练条件化生成对抗网络的数据输入。将训练好的生成器卷积神经网络应用于塔河油田TH12330井区,生成的多个断控岩溶型溶洞地质模型符合地质模式,吻合条件井、概率体数据,且与构造、裂缝和累产基本一致。本研究探索了断控岩溶型溶洞多源训练数据集的构建并在实际应用中取得了显著成果,同时也为其它类型储层深度学习建模中构建可靠且多样化的训练数据集提供了新思路。
Currently,there is no comprehensive training dataset available for the modelling of fault-controlled karst caves using deep learning.In this study,we constructed prototype models for fault-controlled karst caves using outcrop data,seismic data,reliable geological models,and object-based methods.We combined,rotated,cropped,and selected prototype models from different sources to create a reliable and diverse training dataset for fault-controlled karst caves.Additionally,we constructed corresponding virtual well and probability map training datasets,all of which were used to train conditional generative adversarial networks(GANs).The trained generator convolutional neural network was applied to TH12330 well block,Tahe Oilfield.The generated multiple geological models for fault-controlled karst caves were consistent with geological patterns, conditioning well data, conditioning probability mapdata, and aligned with fracture structures, fractures, and cumulative production. This research explores the construction of a multisourcetraining dataset for fault-controlled karst caves and has achieved significant success in a real application example. Furthermore,it provides new insights into building reliable and diverse training dataset for deep learning modelling in other types of reservoirs.
作者
胡迅
侯加根
刘钰铭
HU Xun;HOU Jiagen;LIU Yuming(College of Geosciences,China University of Petroleum-Beijing,Beijing 102249,China)
出处
《石油科学通报》
CAS
2024年第3期422-433,共12页
Petroleum Science Bulletin
基金
国家自然科学基金面上项目(42072146)资助。
关键词
断控岩溶型溶洞
训练数据集
生成对抗网络
深度学习
地质建模
fault-controlled karst caves
training dataset
generative adversarial networks
deep learning
geological modelling