基于神经网络的随机地震反演方法

针对随机地震反演中存在的两个主要问题,随机实现含有噪声和难以从大量随机实现中挖掘有效信息,提出了一种基于神经网络的随机地震反演方法.通过对多组随机实现及其正演地震数据的计算,构建了基于序贯高斯模拟的训练集.这也为应用神经网络求解地球物理反问题,提供了一种有效建立训练集的方法.较之传统的神经网络反演,这种训练集不仅保证了学习样本具有多样性,同时还引入了空间相关性.数值模拟结果表明,该方法只需要通过单层前馈神经网络,就可以比较有效的解决一个500个阻抗参数的反演问题.     

Generalized linear joint PP–PS inversion based on two constraints

Conventional joint PP—PS inversion is based on approximations of the Zoeppritz equations and assumes constant VP/VS; therefore, the inversion precision and stability cannot satisfy current exploration requirements. We propose a joint PP—PS inversion method based on the exact Zoeppritz equations that combines Bayesian statistics and generalized linear inversion. A forward model based on the exact Zoeppritz equations is built to minimize the error of the approximations in the large-angle data, the prior distribution of the model parameters is added as a regularization item to decrease the ill-posed nature of the inversion, low-frequency constraints are introduced to stabilize the low-frequency data and improve robustness, and a fast algorithm is used to solve the objective function while minimizing the computational load. The proposed method has superior antinoising properties and well reproduces real data.     

基于MATLAB神经网络方法的多层砖房震害预测

提出利用MATLAB人工神经网络工具箱建立基于贝叶斯正则算法的BP神经网络模型,以地震区多层砖房震害调查数据为因子的震害预测方法.神经网络模型输入震害因子包括建筑的层数、施工质量、房屋整体性等,输出值为建筑物在地震作用下的破坏程度.结果表明,本方法可以对多层砖房的震害样本进行预测并达到较理想的效果.     

基于迭代残差网络的双能CT图像材料分解研究

双能计算机断层成像技术(DECT)由于其材料分解能力,在高级成像应用中发挥着重要作用.图像域分解直接对CT图像进行线性矩阵反演,但分解后的材料图像会受到噪声和伪影的严重影响.虽然各种正则化方法被提出来解决这个问题,但它们仍然面临着两个挑战:繁琐的参数调整和过度平滑导致的图像细节损失.为此,本文提出一种基于迭代残差网络的...     

基于主成分-正则化极限学习机的超高密度电法非线性反演

超高密度电法是一种新的地球物理探测技术,它通过多通道数据采集和多装置数据联合反演,极大地提高了电法勘探的成像精度.本文提出一种主成分-正则化极限学习机(PC-RELM)非线性反演方法,该方法针对超高密度电法所获取的高维勘探数据进行反演建模,通过随机设定隐层参数来简化模型的学习过程,通过主成分分析方法来进行高维数据降维,最后引入正则化因子提高反演模型的泛化能力.论文给出了超高密度电法的原理、样本构造方法和非线性反演流程,使用交叉验证方法获得了优化的隐节点数目和正则化参数,构造了优化的反演模型.通过两个经典的超高密度模型的反演结果表明,该方法能够较好地解决超高密度电法反演的高维数据非线性建模问题,能够弥补单一装置数据反演的不足,同时相较其他的非线性反演方法(ELM,BPNN和GRNN)具有更加准确的反演结果.     

基于汉南-奎因信息准则的电阻率层析成像径向基神经网络反演

径向基神经网络（RBFNN）具有结构简单、学习速度快、不易陷入局部极小等优点,能够有效地提高电阻率层析成像反演的收敛速度和求解质量.本文针对电阻率层析成像反演的非线性特征,提出了一种基于汉南-奎因信息准则（HQC）的正交最小二乘法（OLS）学习算法（HQOLS）.该算法通过计算HQC的最优值来自动选择RBFNN的网络结构,避免了传统OLS学习算法中阈值参数的设定,保证了网络的泛化性能.通过比较聚类法、梯度法、OLS和HQOLS等学习算法的反演性能,构建了基于RBFNN的电阻率层析成像反演模型.数值仿真和模型反演的结果表明,该方法实现简单,在准确性上优于BP反演,成像质量优于传统最小二乘法反演.     

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.     

广义Stein无偏风险估计与地球物理反问题正则化参数求取

地球物理反演是获取地球信息的重要手段,其求解具有严重的不适定性.为获得稳定的反问题结果,通常需要在目标泛函中加入正则化约束项.正确地估计正则化参数一直是地球物理反问题中的难点.目前存在的选取方法需要根据大量的试验来确定正则化参数,工作量十分巨大,并且存在很大的经验性,很难得到最优的正则化参数.针对这个问题,本文提出了一种基于广义Stein无偏风险估计的正则化参数求取方法.该方法的具体思路是通过求解模型参数均方误差的广义Stein无偏风险估计函数,在反问题求解过程中自动求取正则化参数.本文模型测试结果表明,相比于目前常用的方法,通过该方法得到的正则化参数是最优的.     

Stochastic ground motion models to NGA-West2 and NGA-Sub databases using Bayesian neural network

Ground motion models (GMMs) are traditionally developed from a frequentist approach. The Bayesian framework has received recent attention in developing nonergodic models, measuring uncertainty, or updating the model with additional data. However, no neural networks are developed to date in this framework to predict ground motion parameters or spectra. Hence, the present work develops a probabilistic Bayesian neural network (PBNN) to next-generation attenuation – West2 and Subduction databases using variational inference with mean-field assumption. Network inputs are magnitude, rupture distance, hypocentral depth, shear wave velocity, style of faulting, and region flags; outputs are peak ground values and response spectra. Both models have two hidden layers with seven neurons in each hidden layer. The models are verified for potential overfit, and their performance is validated through the parametric study by varying inputs. The output of a deterministic model is a point estimate. Considering probabilistic layers in hidden and output layers enables the model to capture within-model epistemic uncertainty and aleatory variability. Obtained aleatory standard deviations are consistent with other models. Mean epistemic uncertainty and aleatory variability are in the range 0.07–0.10 and 0.62–0.78 (ln units) for NGA-West2 and 0.09–0.16 and 0.67–0.95 for NGA-Sub models, respectively. The correlation coefficients between recorded and overall mean predictions ranged from 0.94 to 0.97 for NGA-the West2 model and from 0.91 to 0.95 for the NGA-Sub models. Network performance for out-of-training inputs showed increased epistemic deviations with no effect on aleatory deviations.     

Estimation of monthly evaporative loss using relevance vector machine,extreme learning machine and multivariate adaptive regression spline models

The forecasting of evaporative loss (E) is vital for water resource management and understanding of hydrological process for farming practices, ecosystem management and hydrologic engineering. This study has developed three machine learning algorithms, namely the relevance vector machine (RVM), extreme learning machine (ELM) and multivariate adaptive regression spline (MARS) for the prediction of E using five predictor variables, incident solar radiation (S), maximum temperature (T _max), minimum temperature (T _min), atmospheric vapor pressure (VP) and precipitation (P). The RVM model is based on the Bayesian formulation of a linear model with appropriate prior that results in sparse representations. The ELM model is computationally efficient algorithm based on Single Layer Feedforward Neural Network with hidden neurons that randomly choose input weights and the MARS model is built on flexible regression algorithm that generally divides solution space into intervals of predictor variables and fits splines (basis functions) to each interval. By utilizing random sampling process, the predictor data were partitioned into the training phase (70 % of data) and testing phase (remainder 30 %). The equations for the prediction of monthly E were formulated. The RVM model was devised using the radial basis function, while the ELM model comprised of 5 inputs and 10 hidden neurons and used the radial basis activation function, and the MARS model utilized 15 basis functions. The decomposition of variance among the predictor dataset of the MARS model yielded the largest magnitude of the Generalized Cross Validation statistic (≈0.03) when the T _max was used as an input, followed by the relatively lower value (≈0.028, 0.019) for inputs defined by the S and VP. This confirmed that the prediction of E utilized the largest contributions of the predictive features from the T _max, verified emphatically by sensitivity analysis test. The model performance statistics yielded correlation coefficients of 0.979 (RVM), 0.977 (ELM) and 0.974 (MARS), Root-Mean-Square-Errors of 9.306, 9.714 and 10.457 and Mean-Absolute-Error of 0.034, 0.035 and 0.038. Despite the small differences in the overall prediction skill, the RVM model appeared to be more accurate in prediction of E. It is therefore advocated that the RVM model can be employed as a promising machine learning tool for the prediction of evaporative loss.     

A novel approach for classification of earthquake ground-motion records

This paper presents a new clustering procedure based on K-means and self-organizing map (SOM) network algorithms for classification of earthquake ground-motion records. Six scalar indicators are used in data analysis for describing the frequency content features of earthquake ground motions, named as the average spectral period (T _avg ), the mean period (T _m ), the smoothed spectral predominant period (T ₀), the characteristic period (T _4.3), the predominant period based on velocity spectrum (T _gSv ), and the shape factor (Ω). Different clustering validity indexes were applied to determine the best estimates of the number of clusters on real and synthetic data. Results showed the high performance of proposed procedure to reveal salient features of complex seismic data. The comparison between the results of clustering analyses recommend the smoothed spectral predominant period as an effective indicator to describe ground-motion classes. The results also showed that K-means algorithm has better performance than SOM algorithm in identification and classification procedure of ground-motion records.     

Sparse Bayesian linearized amplitude-versus-angle inversion

The technique of seismic amplitude-versus-angle inversion has been widely used to estimate lithology and fluid properties in seismic exploration. The amplitude-versus-angle inversion problem is intrinsically ill-posed and generally stabilized by the use of L2-norm regularization methods but with drawback of smoothing important boundaries between adjacent layers. In this study, we propose a sparse Bayesian linearized solution for amplitude-versus-angle inversion problem to preserve the sharp geological interfaces. In this regard, a priori constraint term with two regularization functions is presented: the sparse constraint regularization and the low-frequency model information. In addition, to obtain high-resolution reflectivity estimation, the model parameters decorrelation technique combined with dipole decomposition method is employed. We validate the applicability of the presented method by both synthetic and real seismic data from the Gulf of Mexico. The accuracy improvement of the presented method is also confirmed by comparing the results with the commonly used Bayesian linearized amplitude-versus-angle inversion.     

An Amplitude-normalized Pseudo Well-log Construction Method and its Application on AVO Inversion in a Well-absent Marine Area

AVO inversion is hard to be efficiently applied in unexploited fields due to the insufficiency of well information. For the sake of AVO inversion in a well-absent area, the most conventional method is to construct pseudo well-logs by defining seismic processing velocity as the P-velocity and computing S-velocity and density using empirical formulas, yet the resolution of the corresponding earth models and final inverted results could be extremely low, and a rough formula could destroy the inversion thoroughly. To overcome this problem, an amplitudenormalized pseudo well-log construction method that reconstructs pseudo well-logs in accordance with computed P-wave reflection amplitudes and nearby drilling data is proposed in this paper. It enhances the inversion resolution efficiently with respect to the real elastic parameter relationships, so that the corresponding AVO inversion results are reasonably improved. In summary, the proposed method is successfully applied in the AVO inversion of a well-absent marine area, and could be valuable in the early phase, particularly of the offshore hydrocarbon exploration.     

人工神经网络系统在接收函数反演中的应用

为使接收函数的反演更为简便,本文提出了一种基于人工神经网络误差反传(BP)算法的接收函数反演新方法,该方法采用人工神经网络反演系统,避免了接收函数反演过程中复杂的地震响应计算及耗时的雅可比矩阵计算,只需经过学习训练就能够解决复杂的实际问题,而且具有记忆功能,这使接收函数的反演工作具有延续性和可继承性.理论数据的反演计算结果表明,该方法是切实可行的.     

Comparison of Bayesian estimates of peak ground acceleration (<Emphasis Type="Italic">A</Emphasis>max) with PSHA in Iran

Bayesian probability theory is an appropriate and useful method for estimating parameters in seismic hazard analysis. The analysis in Bayesian approaches is based on a posterior belief, also their special ability is to take into account the uncertainty of parameters in probabilistic relations and a priori knowledge. In this study, we benefited the Bayesian approach in order to estimate maximum values of peak ground acceleration (A_max) also quantiles of the relevant probabilistic distributions are figured out in a desired future interval time in Iran. The main assumptions are Poissonian character of the seismic events flow and properties of the Gutenberg-Richter distribution law. The map of maximum possible values of A_max and also map of 90% quantile of distribution of maximum values of A_max on a future interval time 100 years is presented. According to the results, the maximum value of the A_max is estimated for Bandar Abbas as 0.3g and the minimum one is attributed to Esfahan as 0.03g. Finally, the estimated values in Bayesian approach are compared with what was presented applying probabilistic seismic hazard (PSH) methods based on the conventional Cornel (1968) method. The distribution function of A_max for future time intervals of 100 and 475 years are calculated for confidence limit of probability level of 90%.     

Dimensionality reduction for efficient Bayesian estimation of groundwater flow in strongly heterogeneous aquifers

We focus on the Bayesian estimation of strongly heterogeneous transmissivity fields conditional on data sampled at a set of locations in an aquifer. Log-transmissivity, Y, is modeled as a stochastic Gaussian process, parameterized through a truncated Karhunen–Loève (KL) expansion. We consider Y fields characterized by a short correlation scale as compared to the size of the observed domain. These systems are associated with a KL decomposition which still requires a high number of parameters, thus hampering the efficiency of the Bayesian estimation of the underlying stochastic field. The distinctive aim of this work is to present an efficient approach for the stochastic inverse modeling of fully saturated groundwater flow in these types of strongly heterogeneous domains. The methodology is grounded on the construction of an optimal sparse KL decomposition which is achieved by retaining only a limited set of modes in the expansion. Mode selection is driven by model selection criteria and is conditional on available data of hydraulic heads and (optionally) Y. Bayesian inversion of the optimal sparse KLE is then inferred using Markov Chain Monte Carlo (MCMC) samplers. As a test bed, we illustrate our approach by way of a suite of computational examples where noisy head and Y values are sampled from a given randomly generated system. Our findings suggest that the proposed methodology yields a globally satisfactory inversion of the stochastic head and Y fields. Comparison of reference values against the corresponding MCMC predictive distributions suggests that observed values are well reproduced in a probabilistic sense. In a few cases, reference values at some unsampled locations (typically far from measurements) are not captured by the posterior probability distributions. In these cases, the quality of the estimation could be improved, e.g., by increasing the number of measurements and/or the threshold for the selection of KL modes.     

The neural network approximation method for solving multidimensional nonlinear inverse problems of geophysics

The iterative approximation neural network method for solving conditionally well-posed nonlinear inverse problems of geophysics is presented. The method is based on the neural network approximation of the inverse operator. The inverse problem is solved in the class of grid (block) models of the medium on a regularized parameterization grid. The construction principle of this grid relies on using the calculated values of the continuity modulus of the inverse operator and its modifications determining the degree of ambiguity of the solutions. The method provides approximate solutions of inverse problems with the maximal degree of detail given the specified degree of ambiguity with the total number of the sought parameters ~n × 10³ of the medium. The a priori and a posteriori estimates of the degree of ambiguity of the approximated solutions are calculated. The work of the method is illustrated by the example of the three-dimensional (3D) inversion of the synthesized 2D areal geoelectrical (audio magnetotelluric sounding, AMTS) data corresponding to the schematic model of a kimberlite pipe.     

基于构造约束的波阻抗边界刻画

通过边界保护正则化和约束反演,在反演的目标函数中引入各种先验信息约束,以解决波阻抗反演的病态问题和带限问题.为了克服波阻抗模型边界过于平滑,在反演中引入地层和断层等构造信息约束,并且通过调整地层分界面和断层处的正则参数值来实现构造约束.此外,采用各向异性扩散法进行平滑处理,改善反演结果.通过合成数据测试和实际资料反演,证明了本文提出的方法对刻画模型边界是有效的.     

Estuary water-stage forecasting by using radial basis function neural network

The Radial basis function neural network (RBFNN) has been successfully applied to many tasks due to its powerful properties in classification and functional approximation. This paper presents a novel RBFNN for water-stage forecasting in an estuary under high flood and tidal effects. The RBFNN adopts a hybrid two-stage learning scheme, unsupervised and supervised learning. In the first scheme, fuzzy min–max clustering is proposed for choosing best patterns for cluster representation in an efficient and automatic way. The second scheme uses supervised learning, which is a multivariate linear regression method to produce a weighted sum of the output from the hidden layer. Since this network has only one layer using a supervised learning algorithm, its training process is much faster than the error back propagation based multilayer perceptrons. Moreover, only one parameter, θ, must be determined manually. The other parameters used in this model can be adjusted automatically by model training. The water-stage data of the Tanshui River under tidal effect are used to construct a water-stage forecasting model that can also be used during flood. The results show that the RBFNN can be applied successfully and provide high accuracy and reliability of water-stage forecasting in an estuary.     

Post-stack impedance blocky inversion based on analytic solution of viscous acoustic wave equation

The conventional impedance inversion method ignores the attenuation effect, transmission loss and inter-layer multiple waves; the smooth-like regularization approach makes the corresponding impedance solution excessively smooth. Both fundamentally limit the resolution of impedance result and lead to the inadequate ability of boundary characterization. Therefore, a post-stack impedance blocky inversion method based on the analytic solution of viscous acoustic equation is proposed. Based on the derived recursive formula of reflections, the 1D viscous acoustic wave equation is solved analytically to obtain zero-offset full-wave field response. Applying chain rule, the analytical expression of the Fréchet derivative is derived for gradient-descent non-linear inversion. Combined with smooth constraints, the blocky constraints can be introduced into the Bayesian inference framework to obtain stable and well-defined inversion results. According to the above theory, we firstly use model data to analyse the influence of incompleteness of forward method on seismic response, and further verify the effectiveness of the proposed method. Then the Q-value sensitivity analysis of seismic trace is carried out to reduce the difficulty of Q-value estimation. Finally, the real data from Lower Congo Basin in West Africa indicate that the proposed approach provide the high-resolution and well-defined impedance result. As a supplement and development of linear impedance inversion method, the non-linear viscous inversion could recover more realistic and reliable impedance profiles.