Inversion Of Gravity Anomaly Based On Deep Learning

As an important tool in the intelligent age,deep learning has excellent performance in processing big data and establishing accurate learning models.In the field of geophysical exploration,deep learning has been initially applied,successfully solving the problems of seismic velocity estimation,fault tracking,and wave impedance inversion,but in the field of gravity and magnetic data inversion,there are still few research results.This paper studies the gravity inversion problem based on deep learning,and uses deep neural network to establish the mapping relationship between the gravity anomaly data observed on the surface and the density distribution of the underground medium,so as to realize the gravity anomaly inversion.The deep learning method relies on the "learning" function of the deep neural network.This method is based on a large amount of data training.The training phase is time-consuming,but after the training,a general deep neural network reflecting the mapping relationship between gravity anomalies and the underground medium model will be obtained.,The calculation time of gravity inversion using this network can be greatly reduced.This paper mainly studies the two-dimensional(2D)gravity inversion problem,divides the underground medium into uniform grid cells,and performs forward simulation by establishing different types and sizes of special underground geological bodies to obtain a large amount of data.To train a deep neural network(DNN)model to establish a linear mapping from the surface gravity anomaly to the density distribution of the underground medium.The trained deep neural network can estimate the corresponding density distribution of underground media based on the gravity anomaly data.In the selection of deep neural networks,this paper chooses two different DNN structures of Fully Connected(FC)and Convolutional Neural Networks(CNN)for testing.The inversion results show that the predicted distribution of geological body density is concentrated and has good results.The resolution to determine the boundary of the anomalous body fully proves the feasibility of deep learning in gravity inversion.Deep neural network has good non-linear ability,has the advantages of strong learning ability,wide coverage,strong adaptability,etc.The application of deep learning to the inversion of gravity anomaly not only makes the inversion result more accurate,but also saves a lot of manpower Material resources provide more possibilities for geophysical inversion.