Research On Gravity And Magnetic Data Processing Based On Deep Learning

Gravity and magnetic method are an important way to understand the distribution of underground density and magnetic susceptibility.It is widely used in mineral and oil explorations and in engineering and environmental problems.With the improvement of gravity and magnetic exploration technology,we not only need to develop the classic data processing technology but also need to find new data processing technology.Among them,deep learning in artificial intelligence is undoubtedly outstanding.In recent years,deep learning with the rapid development of theoretical methods,supporting software and hardware has been applied in many fields and achieved great success.Although the application of deep learning in geophysical is late,new achievements and breakthroughs are endless,which makes the deep learning research a hot topic.The processing of gravity and magnetic data need personal experience and high-intensity repetitive operations.In this paper,we introduce the deep learning technology into potential field data processing which can not only reduce the working pressure of staff but also make improvements.In this paper,the deep learning method is used to study denoising and inversion of potential field data.The corresponding research results can be summarized as follows:1)Traditional denoising methods usually are based on noise level estimation and require heavy manual work to select the parameters.Moreover,field data often includes various noise distribution.In this cases,traditional methods could not eliminate noise effectively.To solve the above problems,we apply the Dn CNN network which is one of deep learning tool to solve the potential field data denoising problem.The proposed method could remove noise adaptively.Comparing to traditional method,such as wavelet denoising,the deep learning method could obtain cleaner records and preserve more information of signals.Finally,we test the network using synthetic and actual data to validate the effectiveness of this method.2)At present,the inversion methods of potential field can be divided into linear and nonlinear inversion.There are several drawbacks for linear inversion method,such as sensitive to the initial model,easy to fall into the local optimal solution,need to calculate the complex sensitivity matrix,ill conditioned equations and poor stability of the solution.The main problem of non-linear inversion methods is the long calculating time and large memory occupying when the quantities of inverse data are large.Especially for 3D inversion,nonlinear inversion method is difficult to work.To solve the above problems,this paper proposes a nonlinear inversion method based on deep learning.This method can avoid the complex forward calculation,and has the characteristics of strong fault tolerance,processing highly nonlinear problems and powerful parallel operation.It is very suitable for solving nonlinear problems that are difficult to be described by explicit functions in geophysics.The deep learning inversion network is mainly based on U-Net.In order to verify the effect of this method,different geological models are designed.The model is mainly composed of prisms and dipping slabs,and the size,depth,location,and physical parameters of each anomalous body are different.Based on the inversion result,3D rendering and profile map are drawn to verify the feasibility of deep learning inversion.Finally,the noise experiment shows that the network has the ability of resisting noise.