摘要
为实现复杂介质中随钻电磁波电阻率测井响应的快速模拟,加速随钻电磁波电阻率测井数据反演,基于物理信息神经网络(Physics-Informed Neural Network,PINN)模拟了随钻电磁波电阻率测井响应。PINN将控制方程纳入损失函数,将偏微分方程求解问题转化为最优化问题,实现偏微分方程的求解。在数值算例中通过PINN方法得到了电场散射场,研究了采样方式、激活函数、网络结构对PINN结果精度的影响。在高电阻率地层和侵入地层模型中,使用PINN方法计算了随钻电磁波电阻率测井响应。数值算例结果表明,基于PINN模拟的随钻电磁波测井响应与有限元方法结果一致,利用该方法可以实现随钻电磁波电阻率测井响应的精确求解。
In order to simulate the response of electromagnetic wave resistivity logging while drilling efficiently in complex media and accelerate the inversion of logging data,the physical-informed neural network(PINN) is used to simulate the response of electromagnetic wave resistivity logging while drilling.PINN incorporates the governing equation into the loss function and transforms the problem of solving partial differential equation into optimization problem.PINN realizes the solution of partial differential equation.In numerical examples,the scattered field is obtained by PINN method.The influence of sampling method,activation function and network architecture on the accuracy of PINN results are studied.The PINN method is used to calculate the response of electromagnetic wave resistivity logging while drilling in high resistivity and intrusive formation models.The numerical results show that the response of electromagnetic wave logging while drilling based on PINN simulation is consistent with the finite element results.This method can be used to accurately solve the response of electromagnetic wave resistivity logging while drilling.
作者
刘阳
王健
徐德龙
LIU Yang;WANG Jian;XU Delong(China State Key Laboratory of Acoustics,Institute of Acoustics,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《测井技术》
CAS
2023年第6期653-661,共9页
Well Logging Technology
基金
国家自然科学基金面上项目“内嵌物理知识的深度神经网络电阻率三维反演研究”(42274177)
国家自然科学基金项目“基于异常电磁扩散理论的感应测井裂缝密度评价研究”(41604123)。
关键词
物理信息神经网络
随钻电磁波电阻率测井
测井响应
电磁场模拟
physics-informed neural network
electromagnetic wave resistivity logging while drilling
logging response
electromagnetic field simulation
作者简介
第一作者:刘阳,女,1999年生,硕士研究生,主要从事电磁场模拟、物理信息神经网络正演的研究。E-mail:liuyang2022@mail.ioa.ac.cn;通信作者:王健,男,1987年生,副研究员,主要从事电磁波场数值模拟、反演算法和电磁法专用装备研究。E-mail:wangjian@mail.ioa.ac.cn。
