3D inversion of DC data using artificial neural networks

In this paper, we investigate the applicability of artificial neural networks in inverting three-dimensional DC resistivity imaging data. The model used to produce synthetic data for training the artificial neural network (ANN) system was a homogeneous medium of resistivity 100 Ωm with an embedded anomalous body of resistivity 1000 Ωm. The different sizes for anomalous body were selected and their location was changed to different positions within the homogeneous model mesh elements. The 3D data set was generated using a finite element forward modeling code through standard 3D modeling software. We investigated different learning paradigms in the training process of the neural network. Resilient propagation was more efficient than any other paradigm. We studied the effect of the data type used on neural network inversion and found that the use of location and the apparent resistivity of data points as the input and corresponding true resistivity as the output of networks produces satisfactory results. We also investigated the effect of the training data pool volume on the inversion properties. We created several synthetic data sets to study the interpolation and extrapolation properties of the ANN. The range of 100–1000 Ωm was divided into six resistivity values as the background resistivity and different resistivity values were also used for the anomalous body. Results from numerous neural network tests indicate that the neural network possesses sufficient interpolation and extrapolation abilities with the selected volume of training data. The trained network was also applied on a real field dataset, collected by a pole-pole array using a square grid (8 ×8) with a 2-m electrode spacing. The inversion results demonstrate that the trained network was able to invert three-dimensional electrical resistivity imaging data. The interpreted results of neural network also agree with the known information about the investigation area.     

电阻率二维神经网络反演

由于非线性特性地球物理反演一直以来都是一个比较困难的问题. 近十年来,非线性反演方法如人工神经网络、遗传算法在地球物理数据解释中得到越来越多的应用,但目前基本仍限于一维反演问题. 对于二维反问题,反演参数较多,神经网络反演运用较少. 本文利用BP神经网络优化方法,实现了电阻率二维非线性反演. 与传统线性化的迭代反演比较,神经网络反演能够克服传统方法的不足、获得更好的反演结果.     

基于遗传神经网络的大地电磁反演

为进一步提高大地电磁非线性反演的稳定性、运算效率及准确度,将遗传神经网络算法引入大地电磁反演.首先针对大地电磁二维地电模型建立BP（Back Propagation）神经网络基本框架进行学习训练,网络输入为已知地电模型的视电阻率参数,输出为该地电模型参数;再利用遗传算法对神经网络学习训练过程进行优化,计算出多种地电模型网络连接权值和阈值的最优解;最后将最优连接权值和阈值对未知模型进行反演测试,网络输入为未知地电模型的视电阻率参数,输出为该地电模型参数.模型实验表明：遗传神经网络算法充分结合了遗传算法的全局寻优性和神经网络的局部寻优性,相比单一神经网络算法,在网络学习训练中提高了解的收敛成功率和计算速度,在反演测试中能更准确地逼近真实模型.将遗传神经网络算法与最小二乘正则化反演进行对比,理论模型和实测数据都验证了遗传神经网络算法在大地电磁反演中的可行性和有效性.     

Pitfalls of 1D inversion of small-loop electromagnetic data for detecting man-made objects

In order to interpret field data from small-loop electromagnetic (EM) instruments with fixed source–receiver separation, 1D inversion method is commonly used due to its efficiency with regard to computation costs. This application of 1D inversion is based on the assumption that small-offset broadband EM signals are insensitive to lateral resistivity variation. However, this assumption can be false when isolated conductive bodies such as man-made objects are embedded in the earth. Thus, we need to clarify the applicability of the 1D inversion method for small-loop EM data. In order to systematically analyze this conventional inversion approach, we developed a 2D EM inversion algorithm and verified this algorithm with a synthetic EM data set. 1D and 2D inversions were applied to synthetic and field EM data sets. The comparison of these inversion results shows that the resistivity distribution of the subsurface constructed by the 1D inversion approach can be distorted when the earth contains man-made objects, because they induce drastic variation of the resistivity distribution. By analyzing the integrated sensitivity of the small-loop EM method, we found that this pitfall of 1D inversion may be caused by the considerable sensitivity of the small-loop EM responses to lateral resistivity variation. However, the application of our 2D inversion algorithm to synthetic and field EM data sets demonstrate that the pitfall of 1D inversion due to man-made objects can be successfully alleviated. Thus, 2D EM inversion is strongly recommended for detecting conductive isolated bodies, such as man-made objects, whereas this approach may not always be essential for interpreting the EM field data.     

Two-dimensional joint inversion of radiomagnetotelluric and direct current resistivity data

An algorithm for the two-dimensional (2D) joint inversion of radiomagnetotelluric and direct current resistivity data was developed. This algorithm can be used for the 2D inversion of apparent resistivity data sets collected by multi-electrode direct current resistivity systems for various classical electrode arrays (Wenner, Schlumberger, dipole-diplole, pole-dipole) and radiomagnetotelluric measurements jointly. We use a finite difference technique to solve the Helmoltz and Poisson equations for radiomagnetotelluric and direct current resistivity methods respectively. A regularized inversion with a smoothness constrained stabilizer was employed to invert both data sets. The radiomagnetotelluric method is not particularly sensitive when attempting to resolve near-surface resistivity blocks because it uses a limited range of frequencies. On the other hand, the direct current resistivity method can resolve these near-surface blocks with relatively greater accuracy. Initially, individual and joint inversions of synthetic radiomagnetotelluric and direct current resistivity data were compared and we demonstrated that the joint inversion result based on this synthetic data simulates the real model more accurately than the inversion results of each individual method. The developed 2D joint inversion algorithm was also applied on a field data set observed across an active fault located close to the city of Kerpen in Germany. The location and depth of this fault were successfully determined by the 2D joint inversion of the radiomagnetotelluric and direct current resistivity data. This inversion result from the field data further validated the synthetic data inversion results.     

Fracture-frequency prediction from borehole wireline logs using artificial neural networks

Borehole-wall imaging is currently the most reliable means of mapping discontinuities within boreholes. As these imaging techniques are expensive and thus not always included in a logging run, a method of predicting fracture frequency directly from traditional logging tool responses would be very useful and cost effective. Artificial neural networks (ANNs) show great potential in this area. ANNs are computational systems that attempt to mimic natural biological neural networks. They have the ability to recognize patterns and develop their own generalizations about a given data set. Neural networks are trained on data sets for which the solution is known and tested on data not previously seen in order to validate the network result. We show that artificial neural networks, due to their pattern recognition capabilities, are able to assess the signal strength of fracture-related heterogeneity in a borehole log and thus fracture frequency within a borehole. A combination of wireline logs (neutron porosity, bulk density, P-sonic, S-sonic, deep resistivity and shallow resistivity) were used as input parameters to the ANN. Fracture frequency calculated from borehole televiewer data was used as the single output parameter. The ANN was trained using a back-propagation algorithm with a momentum learning function. In addition to fracture frequency within a single borehole, an ANN trained on a subset of boreholes in an area could be used for prediction over the entire set of boreholes, thus allowing the lateral correlation of fracture zones.     

1D and 2D Cole-Cole-inversion of time-domain induced-polarization data

A new method for the 2D inversion of induced polarization (IP) data in the time domain has been developed. The entire IP transients were observed and inverted into 2D Cole-Cole earth models, including resistivity, chargeability, relaxation time and the frequency constant. Firstly, a modified 1D time-domain electromagnetic algorithm was used to calculate the response of a layered polarizable ground. The transient signals were then inverted using the Marquardt method to derive the Cole-Cole parameters of each layer. However, model calculations showed that the EM effects could be neglected for the time range (>1 ms) and for the transmitter–receiver distances (<50 m) used in this study. Therefore, the induction effects were not considered for the solution of the 2D inverse problem and a DC solution was applied. An approximative forward algorithm was introduced in order to calculate the IP transients directly in the time domain and in order to speed up the inverse procedure. The approximation is highly accurate, and this is demonstrated by comparing the approximations with their exact solutions up to 3D. The inverse algorithm presented consists of two steps. The transient voltages of an array data set were inverted separately into a two-dimensional resistivity model for each time channel. The time-dependent resistivity of each cell was then interpreted as the response of a homogeneous half-space. In the 2D inversion algorithm, a 3D DC algorithm was used as a forward operator. The method only requires a standard 2D DC inversion and a homogenous half-space Cole-Cole inversion. The developed algorithm has been successfully applied to synthetic data sets and to a field data set obtained from a waste site situated close to Düren in Germany.     

Validation of joint inversion of direct current and electromagnetic measurements

One of the ways to improve the information content of a set of field data is that of combining the interpretation of disparate data sets. Electromagnetic and direct current resistivity methods suffer from inherent equivalence problem. Application of joint inversion for these measurements can overcome the problem of equivalence very well. In the present work, synthetic data from vertical electrical sounding (VES) and horizontal coplanar low-frequency induction sounding (EMHD) are inverted individually and jointly over different types of 1D earth structures. Global optimization with Monte Carlo Multistart algorithm was used in the calculations. The results obtained from the inversions of synthetic data indicate that the joint inversion significantly improves the solution reducing the ambiguity of the models.     

2.5维井间电磁反演在中国东部孤岛油田的应用

In this study, we present a practical technique of transforming cross-hole EM data into the inter-well resistivity distribution. The a priori information constraint is incorporated into an iterative regularized inversion procedure and a variable roughness is added into the inversion process. Finite element approximation based on a two and a half-dimensional （2.5D） model has been developed for the forward problem and the ＂pseudo-forward＂ problem needed for constructing the sensitivity matrix and synthetic data set. The regularized least-squares inversion scheme, constrained with the a priori information obtained from well logs, was adopted to reconstruct the inter-well resistivity profile from two synthetic electromagnetic data sets and field data acquired in the Gudao Oil Field, East China. The partial derivatives of the sensitivity matrix were computed by the adjoint equation based on the reciprocity principle. Inversion results of the synthetic and field data examples suggest that our method is robust and stable in the presence of random noise in the field data and can be used for cross-hole EM field data interpretation.     

基于非结构网格的电阻率三维带地形反演

地表起伏地形在野外矿产资源勘察中不可避免,其对直流电阻率法勘探影响巨大.近年来,电阻率三维正演取得诸多进展,特别是应用非结构网格我们能够进行任意复杂地形和几何模型的电阻率三维数值模拟,但面向实际应用的起伏地形下电阻率三维反演依然困难.本文基于非结构化四面体网格,并考虑到应用GPS/GNSS时,区域地球物理调查中可非规则布设测网的实际特点,实现了任意地形(平坦或起伏)条件下、任意布设的偶极-偶极视电阻率数据的不完全Gauss-Newton三维反演.合成数据的反演结果表明了方法的有效性,可应用于复杂野外环境下的三维电法勘探.     

Common image gather conditioning using cycle generative adversarial networks

Seismically derived amplitude-versus-angle attributes along with well constraints are the base inputs into inverting seismic into subsurface properties. Conditioning the common image gathers is a common workflow in quantitative inversion and leads to a more accurate inversion product due to the removal of post-migration artefacts. Here, we apply a neural network to condition the post-migration gathers. The network is a cycle generative adversarial network, CycleGAN, which was designed for image-to-image translation. This can be considered the same problem as translating an artefact rich seismic gather to an artefact free seismic gather. To assess the feasibility of applying the network to pre-stack conditioning, synthetic data sets were generated to train different networks for different tasks. The networks were trained to remove white noise, residual de-multiples, gather flattening and a combination of the above for conditioning. The results show that a trained network was able remove white noise providing a more robust amplitude-versus-offset calculation. Another network trained using synthetic gathers with and without multiples assisted in multiple removal. However, instability around primary preservation has been observed so the network works better as a residual de-multiple method. For gather conditioning, a network was trained with the unpaired artefact-rich and artefact-free training data where the artefacts included complex moveout, noise and multiples. When applied to the test data sets, the networks cleaned the artefact-rich test data and translated complex moveout into flat gathers whilst preserving the amplitude response. Finally, two networks are applied to real data where a gather based on the well logs is used to quantify the match between the conditioned gathers and the raw gathers. The first network used synthetic data to train the network and, when applied to real data, provided a better tie with the well. The second network was trained with synthetic gathers whose properties were constrained by real seismic gathers from near the well. As anticipated, the network trained on the representative training data outperforms the network trained using the unconstrained data. However, the ability of the first network to condition the gather indicates that a sweep of networks can be trained without the need for real data and applied in a manner analogous to the way parameters are adjusted in traditional geophysical methods. The results show that the different neural networks can offer an alternative or augmentation to the existing geophysical workflow for conditioning pre-stack seismic gathers.     

The use of reference models from a priori data to guide 2D inversion of electrical resistivity tomography data

Clay-rich till plains cover much of the UK. Such sites are attractive locations for landfills, since clay aquitards lower the risk of landfill leachate entering groundwater. However, such tills often contain sand and gravel bodies that can act as leachate flow routes. Such bodies may not be detected by conventional site investigation techniques such as drilling boreholes and trial pitting. A method of guided inversion, where a priori data are used to construct structural reference models for use in inverting electrical resistivity tomography data, was proposed as a tool to improve the detection of sand and gravel bodies within clay-rich till sequences.
Following a successful 2D guided inversion synthetic modelling study, a field study was undertaken. Wenner 2D electrical resistivity tomography lines, resistivity cone penetrometry bores and electromagnetic induction ground resistivity data were collected over a site on the East Yorkshire coast, England, where sand and gravel lenses were known to exist from cliff exposures. A number of equally valid geoelectrical models were constructed using the electromagnetic and resistivity cone data. These were used as structural reference models in the inversion of the resistivity tomography data. Blind inversion using an homogenous reference model was also carried out for comparison.
It was shown for the first time that the best solution model produced by 2D inversion of one data set with a range of structural reference models could be determined by using the l ₂ model misfit between the solution models and associated reference models (reference misfit) as a proxy for the l ₂ misfit between the solution models and the synthetic model or 'best-guess' geoelectrical model (true misfit). The 2D methodology developed here is applicable in clay-rich till plains containing sand and gravel bodies throughout the UK.     

Nonlinear inversion of electrical resistivity imaging using pruning Bayesian neural networks

Conventional artificial neural networks used to solve electrical resistivity imaging (ERI) inversion problem suffer from overfitting and local minima. To solve these problems, we propose to use a pruning Bayesian neural network (PBNN) nonlinear inversion method and a sample design method based on the K-medoids clustering algorithm. In the sample design method, the training samples of the neural network are designed according to the prior information provided by the K-medoids clustering results; thus, the training process of the neural network is well guided. The proposed PBNN, based on Bayesian regularization, is used to select the hidden layer structure by assessing the effect of each hidden neuron to the inversion results. Then, the hyperparameter α _k , which is based on the generalized mean, is chosen to guide the pruning process according to the prior distribution of the training samples under the small-sample condition. The proposed algorithm is more efficient than other common adaptive regularization methods in geophysics. The inversion of synthetic data and field data suggests that the proposed method suppresses the noise in the neural network training stage and enhances the generalization. The inversion results with the proposed method are better than those of the BPNN, RBFNN, and RRBFNN inversion methods as well as the conventional least squares inversion.     

An approximation method for solving the inverse MTS problem with the use of neural networks

The work develops the approximation approach to solving the inverse MTS problem with the use of neural networks. The inverse problem is considered in model classes of parametrized geoelectric structures, whose electric conductivity is controlled by a few hundreds of macroparameters (N ∼ 300). An approximate inverse operator of the problem is constructed for each model class as a neural network, whose coefficients are determined in the process of training on a representative sample of standard examples of forward problem solutions. The problem of determination of the model class of geolectric structures corresponding to the presented input MT data is solved with the use of the neural network classifier constructed for the available set of model classes of structures. Regularizing factors and errors of the neural network method are analyzed. The operation of the algorithm is illustrated by examples of the 2-D inversion of synthetic MT data.     

Deep neural network application for 4D seismic inversion to changes in pressure and saturation: Optimizing the use of synthetic training datasets

In this work, we tackle the challenge of quantitative estimation of reservoir dynamic property variations during a period of production, directly from four-dimensional seismic data in the amplitude domain. We employ a deep neural network to invert four-dimensional seismic amplitude maps to the simultaneous changes in pressure, water and gas saturations. The method is applied to a real field data case, where, as is common in such applications, the data measured at the wells are insufficient for properly training deep neural networks, thus, the network is trained on synthetic data. Training on synthetic data offers much freedom in designing a training dataset, therefore, it is important to understand the impact of the data distribution on the inversion results. To define the best way to construct a synthetic training dataset, we perform a study on four different approaches to populating the training set making remarks on data sizes, network generality and the impact of physics-based constraints. Using the results of a reservoir simulation model to populate our training datasets, we demonstrate the benefits of restricting training samples to fluid flow consistent combinations in the dynamic reservoir property domain. With this the network learns the physical correlations present in the training set, incorporating this information into the inference process, which allows it to make inferences on properties to which the seismic data are most uncertain. Additionally, we demonstrate the importance of applying regularization techniques such as adding noise to the synthetic data for training and show a possibility of estimating uncertainties in the inversion results by training multiple networks.     

基于不等式约束的最小二乘法三维电阻率反演及其算法优化

基于光滑约束的最小二乘法是三维电阻率反演的主要方法,但该方法在某些情况下存在着多解性较强的问题,且普遍耗时较长,严重制约了三维反演方法的推广与发展.为改善上述问题,将表征模型参数变化范围的不等式约束作为先验信息引入最小二乘线性反演方法中,有效地改善了反演结果的精度,降低了反演的多解性问题.为了解决耗时较长的问题,基于预条件共轭梯度(PCG)算法和Cholesky分解法的特点提出了一套优化三维电阻率反演计算效率的计算方案.在该方案中,Cholesky分解法被用来求解敏感度矩阵计算中的多个点源场的正演问题,Cholesky分解法只需对总体系数矩阵进行一次分解,然后对不同的右端向量进行回代即可.将预条件共轭梯度法引入到三维电阻率反演方程的求解中,将雅可比迭代中的对角阵作为预处理矩阵,其具有求逆方便、无需内存空间的特点,有效地加快了收敛速度.对合成数据以及实测数据的反演算例表明,借助不等式约束和反演效率优化方案,最小二乘反演方法可得到较为精确的反演结果,有效地提高了反演计算效率,具有良好的推广前景.     

Assessment of the reliability of 2D inversion of apparent resistivity data

The reliability of inversion of apparent resistivity pseudosection data to determine accurately the true resistivity distribution over 2D structures has been investigated, using a common inversion scheme based on a smoothness‐constrained non‐linear least‐squares optimization, for the Wenner array. This involved calculation of synthetic apparent resistivity pseudosection data, which were then inverted and the model estimated from the inversion was compared with the original 2D model. The models examined include (i) horizontal layering, (ii) a vertical fault, (iii) a low‐resistivity fill within a high‐resistivity basement, and (iv) an upfaulted basement block beneath a conductive overburden. Over vertical structures, the resistivity models obtained from inversion are usually much sharper than the measured data. However, the inverted resistivities can be smaller than the lowest, or greater than the highest, true model resistivity. The substantial reduction generally recorded in the data misfit during the least‐squares inversion of 2D apparent resistivity data is not always accompanied by any noticeable reduction in the model misfit. Conversely, the model misfit may, for all practical purposes, remain invariant for successive iterations. It can also increase with the iteration number, especially where the resistivity contrast at the bedrock interface exceeds a factor of about 10; in such instances, the optimum model estimated from inversion is attained at a very low iteration number. The largest model misfit is encountered in the zone adjacent to a contact where there is a large change in the resistivity contrast. It is concluded that smooth inversion can provide only an approximate guide to the true geometry and true formation resistivity.     

三维地震神经网络反演在探明煤层富水灾害层的应用

在常规测井约束反演的的基础上,开展神经网络特征参数反演,将波阻抗等地震属性转化为与含水性更为密切的孔隙度、视电阻率数据体,使地震反演的地质属性与测井上的地质属性达到最优的相关性,从而实现应用三维地震对煤层顶板富水进行评价的目的。由于煤层顶板富水区的特殊性质,它同样也是地震后的易破坏层,因而对它的探明从抗震角度以及震害预测角度都是有价值的。以淮北某采区为例,通过孔隙度及电阻率的神经网络反演对研究区10#煤层顶板的富水性进行预测。反演结果表明采区北部发育一个强富水陷落柱,与钻孔揭示结果吻合。采区西部10#煤层顶板与第四系含水层呈不整合接触关系,神经网络反演结果预测为强富水区,同样与井下工程揭示富水特征吻合。利用多属性融合的神经网络反演可有效预测煤层顶板的富水特征,为煤矿安全生产以及抗震提供重要保障。     

二维MT快速曲线对比反演方法的可行性研究

最近开发了一种针对二维大地电磁野外数据进行处理解释的新反演方法。该方法以加入阻抗相位信息的一维大地电磁连续介质曲线对比法为基础 ,把一维反演得到的电阻率和相位的数据集作为二维反演的初始模型 ,使用二维有限单元法做正演模拟。在程序的后继迭代中 ,深度方向上用一维反演修改模型的电阻率和深度值 ,沿测线方向由二维有限元作修改 ,反演结束可得到一个接近真实电性分布的电阻率数据集 ,并绘制成电阻率断面图。对模型的反演实验结果显示 ,该反演方法能够较真实地反映地下电性分布 ,而且避免了偏导数矩阵的计算 ,其原理简单 ,计算速度快 ,表明该反演方案是可行的