基于CNN的矿井富水区电磁反演研究  

Electromagnetic inversion of mine water-rich area based on CNN

作　　者：郭瑞 董振良 乔鹏举 闫涛 苗壮 田丰[3] GUO Rui;DONG Zhenliang;QIAO Pengju;YAN Tao;MIAO Zhuang;TIAN feng(China Coal Shaanxi Yulin Energy Chemical Co.,Ltd.,Yulin 719000,China;China Coal Energy Research Institute Co.,Ltd.,Xi’an 710054,China;College of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)

机构地区：[1]中煤陕西榆林能源化工有限公司,陕西榆林719000 [2]中煤能源研究院有限责任公司,陕西西安710054 [3]西安科技大学通信与信息工程学院,陕西西安710054

出　　处：《西安科技大学学报》2023年第6期1186-1194,共9页Journal of Xi’an University of Science and Technology

基　　金：陕西省重点研发计划项目(2020GY-029)。

摘　　要：针对传统卷积神经网络(CNN)在电磁反演中提取数据特征时冗余信息多,导致网络反演精度降低的问题,提出一种变步长卷积神经网络电磁反演方法,将输入数据拓展为一维行向量,在各层网络中交替使用不同步长的卷积核进行卷积运算提取数据特征,利用变步长卷积方式替代传统网络的池化层,完成对冗余信息的过滤和特征信息的选择,并通过小卷积核级联的方式增大网络感受野提高网络的非线性表达能力。通过二维时域有限差分法(2D-FDTD)对不同电磁参数的富水区模型进行正演计算,并根据计算得出的电场时域响应特征建立样本数据集;将变步长卷积神经网络应用于电磁反演研究,建立适用于富水区问题的变步长卷积神经网络电磁反演模型,并验证变步长卷积神经网络电磁反演方法的精度。结果表明:该方法对坐标位置的反演平均相对误差为2.85%,对相对介电常数的反演平均相对误差为6.07%,反演结果与实际模型吻合度较高。所提方法对提高矿井富水区电磁反演的精度和效率具有一定的理论参考价值。Aiming at the problem that the traditional convolutional neural network(CNN)has a lot of redundant information when extracting data features in electromagnetic inversion,which leads to the reduction of network inversion accuracy,this paper proposed a variable-step convolutional neural network electromagnetic inversion method.The method expands the input data into a one-dimensional row vector,and alternately uses convolution kernels with different convolution steps in each layer of the network to extract data features.The method is used to replace the pooling layer of the traditional convolutional neural network to complete the filtering of redundant information and the selection of feature information.And the network receptive field is increased by cascading small convolution kernels to improve the nonlinear expression ability of the network.The two-dimensional finite-difference time-domain method(2D-FDTD)is used to calculate the water-rich area model with different electromagnetic parameters,and the sample data set is established according to the calculated electric field time-domain response characteristics.The variable step-size convolutional neural network is applied to establish the electromagnetic inversion model suitable for the problem of water-rich area,and its accuracy is verified.The results show that the average relative error of the method for the inversion of coordinate positions is 2.85%,and for the inversion of relative permittivity is 6.07%,and the inversion results are in good agreement with the actual model.The method has some theoretical reference value for improving the accuracy and efficiency of electromagnetic inversion in water-rich area of mine.

关 键 词：矿井富水区 电磁探测 电磁反演 卷积神经网络 变步长CNN 

分 类 号：TD745[矿业工程—矿井通风与安全]