基于深度学习U-net网络的重力数据界面反演方法

重力数据的密度界面反演是位场数据解释中的一项主要工作, 在区域构造演化、深部莫霍面确定等领域的研究中发挥重要作用.近年来, 数据驱动的深度学习方法广泛地应用在地球物理数据处理与反演中, 本文提出一种基于深度学习U-net网络的重力数据密度界面反演方法.首先, 对半椭球体界面模型进行随机抽取和组合进而形成地下起伏界面数据集, 并基于Parker正演理论对界面数据集进行重力异常正演计算, 为深度学习网络模型的训练提供特征完备的数据源; 其次, 设计了基于U-net网络模型的深度学习界面反演算法, 在传统的损失函数基础上增加光滑损失项和过拟合抑制项, 提高重力界面反演结果的光滑性和收敛效率; 最后通过测试样本集进行反演预测, 验证建立深度学习网络模型的泛化性.本文通过理论模型和实际数据试验分析了本文方法在密度界面反演中的有效性和实用性, 基于改进损失函数约束的深度学习界面反演方法有效地提高了密度界面反演的收敛效率和计算稳定性.

Gravity data density interface inversion based on U-net deep learning network

Density interface inversion of gravity data is a major work in potential field data interpretation, which plays an important role in regional tectonic evolution and determination of deep Moho interface. In recent years, deep learning has been widely used in geophysical data processing and inversion. In this paper, a density interface inversion method based on deep learning U-net network is proposed. Firstly, the semi-ellipsoid interface model was randomly selected and combined to form the dataset of underground density interface. Based on Parker forward theory, the gravity anomaly forward calculation of the interface dataset was carried out to provide training dataset of deep learning network model. Secondly, a deep learning interface inversion algorithm based on U-net network was designed. The smooth loss function and overfitting suppression loss function were added to the traditional loss function to improve the smoothness and convergence efficiency of the inversion results of the gravity data. Finally, the generalization of deep learning network model is verified by inverse prediction of test sample set. The effectiveness and practicability of the proposed method in density interface inversion are analyzed through theoretical model and real measured data. The deep learning inversion method based on the improved loss function constraint effectively improves the convergence efficiency and computational stability of density interface inversion.