基于U-Net卷积神经网络的直流电阻率法数据重构  

Data Reconstruction of Direct Current Resistivity Method Based on U-Net Convolutional Neural Network

作　　者：李泽扬 马欢 张浩楠 代一龙 李阳 杨瀛彧 LI Ze-Yang;MA Huan;ZHANG Hao-Nan;DAI Yi-Long;LI Yang;YANG Ying-Yu(School of Geoscience,Institute of Disaster Prevention,Sanhe 065201,Hebei,China;Institute of Disaster Prevention,Langfang 065000,Hebei,China;School of Information Engineering,Institute of Disaster Prevention,Sanhe 065201,Hebei,China)

机构地区：[1]防灾科技学院地球科学学院,河北三河065201 [2]河北省地震动力学重点实验室,河北廊坊065000 [3]防灾科技学院信息工程学院,河北三河065201

出　　处：《华南地质》2025年第1期240-248,共9页South China Geology

基　　金：中央高校科研业务费专项(ZY20240305);廊坊市青年拔尖人才项目(XY202304);廊坊市科学技术研究与发展计划自筹经费项目(NO.2023013173)。

摘　　要：本文针对地球物理电阻率法实测数据采集中因人文噪声引起的数据突变问题,提出了一种基于卷积神经网络(CNN)的数据重构算法。鉴于电阻率法控制方程的非线性特征,传统的线性插值技术可能导致反演结果精度下降。本研究首先构建了CNN模型,并通过三维有限差分法进行电阻率法正演数值模拟,生成训练集和测试集。利用训练集对CNN模型进行训练,并基于损失函数结果优化U-Net网络参数。通过对比线性插值技术和CNN重构合成突变数据的反演结果,验证了CNN在数据重构中的有效性和优越性。研究结果表明,U-Net-CNN可以有效重构非线性直流电阻率法数据,为提高地球物理数据采集精度和反演结果的可靠性提供了新的技术途径。In this paper,a data reconstruction algorithm based on Convolutional Neural Networks(CNN)is proposed for the issue of data mutation caused by humanistic noise in the geophysical resistivity method data acquisition.Given the nonlinear characteristics of the resistivity method control equation,the traditional linear interpolation techniques may lead to a decline in the accuracy of the inversion results.In this study,a CNN model is first constructed and numerical simulation of the resistivity method inversion is carried out by the three-dimensional finite difference method to generate a training dataset and a test dataset.The CNN model is trained using the training set and the U-Net network parameters are optimized based on the loss function results.By comparing the inversion results of data reconstructed with linear interpolation technique and CNN reconstruction of synthetic mutant data,the effectiveness and superiority of CNN in data reconstruction are verified.The results demonstrate that U-Net-CNN can effectively reconstruct nonlinear direct current resistivity method data,providing a new technical approach to improve the accuracy of geophysical data acquisition and the reliability of inversion results.

关 键 词：卷积神经网络 直流电阻率法 数值模拟 数据重构 

分 类 号：P3[天文地球—地球物理学]