地质学
地下室
盐丘
反演(地质)
地球物理学
人工神经网络
磁异常
海底管道
重力异常
油气勘探
反问题
接头(建筑物)
地震学
算法
计算机科学
人工智能
岩土工程
地貌学
石油工程
数学
工程类
数学分析
构造学
油田
建筑工程
土木工程
作者
Zahra Ashena,Hojjat Kabirzadeh,Jeong Woo Kim,Wang Xin,Mohammed Y. Ali
出处
期刊:Spe Journal
[Society of Petroleum Engineers]
日期:2023-10-25
卷期号:29 (02): 1015-1028
摘要
Summary By using a deep neural network (DNN), a novel technique is developed for a 2.5D joint inversion of gravity and magnetic anomalies to model subsurface salts and basement structures. The joint application of gravity and magnetic anomalies addresses the inherent nonuniqueness problem of geophysical inversions. Moreover, DNN is used to conduct the nonlinear inverse mapping of gravity and magnetic anomalies to depth-to-salt and depth-to-basement. To create the training data set, a three-layer forward model of the subsurface is designed indicating sediments, salts, and the basement. The length and height of the model are determined based on the dimensions of the target area to be investigated. Several random parameters are set to create different representations of the forward model by altering the depth and shape of the layers. Given the topography of the salts and basement layers as well as their predefined density and susceptibility values, the gravity and magnetic anomalies of the forward models are calculated. Using multiprocessing algorithms, thousands of training examples are simulated comprising gravity and magnetic anomalies as input features and depth-to-salt and depth-to-basement as labels. The application of the proposed technique is evaluated to interpret the salt–basement structures over hydrocarbon reservoirs in offshore United Arab Emirates (UAE). Correspondingly, a DNN model is trained using the simulated data set of the target region and is assessed by making predictions on the random actual and noise-added synthetic data. Finally, gravity-magnetic anomalies are fed into the DNN inverse model to estimate the salts and basement structures over three profiles. The results proved the capability of our technique in modeling the subsurface structures.
科研通智能强力驱动
Strongly Powered by AbleSci AI