http://www.sciencedirect.com/science/article/pii/S1674987112000254

The applications of intelligent techniques have increased exponentially in recent days to study most of the non-linear parameters.In particular,the behavior of earth resembles the non-linearity applications.An efficient tool is needed for the interpretation of geophysical parameters to study the subsurface of the earth.Artificial Neural Networks(ANN) perform certain tasks if the structure of the network is modified accordingly for the purpose it has been used.The three most robust networks were taken and comparatively analyzed for their performance to choose the appropriate network.The single-layer feed-forward neural network with the back propagation algorithm is chosen as one of the well-suited networks after comparing the results.Initially,certain synthetic data sets of all three-layer curves have been taken for training the network,and the network is validated by the Held datasets collected from Tuticorin Coastal Region(78°7′30″E and 8°48′45″N),Tamil Nadu.India.The interpretation has been done successfully using the corresponding learning algorithm in the present study.With proper training of back propagation networks,it tends to give the resistivity and thickness of the subsurface layer model of the field resistivity data concerning the synthetic data trained earlier in the appropriate network.The network is trained with more Vertical Electrical Sounding(VES) data,and this trained network is demonstrated by the field data.Groundwater table depth also has been modeled.     

Inversion of quasi-3D DC resistivity imaging data using artificial neural networks

The objective of this paper is to investigate the applicability of artificial neural networks in inverting quasi-3D DC resistivity imaging data. An electrical resistivity imaging survey was carried out along seven parallel lines using a dipole-dipole array to confirm the validation of the results of an inversion using an artificial neural network technique. The model used to produce synthetic data to train the artificial neural network was a homogeneous medium of 100Ωm resistivity with an embedded anomalous body of 1000Ωm resistivity. The network was trained using 21 datasets (comprising 12159 data points) and tested on another 11 synthetic datasets (comprising 6369 data points) and on real field data. Another 24 test datasets (comprising 13896 data points) consisting of different resistivities for the background and the anomalous bodies were used in order to test the interpolation and extrapolation of network properties. Different learning paradigms were tried in the training process of the neural network, with the resilient propagation paradigm being the most efficient. The number of nodes, hidden layers, and efficient values for learning rate and momentum coefficient have been studied. Although a significant correlation between results of the neural network and the conventional robust inversion technique was found, the ANN results show more details of the subsurface structure, and the RMS misfits for the results of the neural network are less than seen with conventional methods. The interpreted results show that the trained network was able to invert quasi-3D electrical resistivity imaging data obtained by dipole-dipole configuration both rapidly and accurately.     

A direct inversion scheme for deep resistivity sounding data using artificial neural networks

Initialization of model parameters is crucial in the conventional 1D inversion of DC electrical data, since a poor guess may result in undesired parameter estimations. In the present work, we investigate the performance of neural networks in the direct inversion of DC sounding data, without the need ofa priori information. We introduce a two-step network approach where the first network identifies the curve type, followed by the model parameter estimation using the second network. This approach provides the flexibility to accommodate all the characteristic sounding curve types with a wide range of resistivity and thickness. Here we realize a three layer feed-forward neural network with fast back propagation learning algorithms performing well. The basic data sets for training and testing were simulated on the basis of available deep resistivity sounding (DRS) data from the crystalline terrains of south India. The optimum network parameters and performance were decided as a function of the testing error convergence with respect to the network training error. On adequate training, the final weights simulate faithfully to recover resistivity and thickness on new data. The small discrepancies noticed, however, are well within the resolvability of resistivity sounding curve interpretations.     

A hybrid Bayesian back‐propagation neural network approach to multivariate modelling

There is growing interest in the use of back‐propagation neural networks to model non‐linear multivariate problems in geotehnical engineering. To overcome the shortcomings of the conventional back‐propagation neural network, such as overfitting, where the neural network learns the spurious details and noise in the training examples, a hybrid back‐propagation algorithm has been developed. The method utilizes the genetic algorithms search technique and the Bayesian neural network methodology. The genetic algorithms enhance the stochastic search to locate the global minima for the neural network model. The Bayesian inference procedures essentially provide better generalization and a statistical approach to deal with data uncertainty in comparison with the conventional back‐propagation. The uncertainty of data can be indicated using error bars. Two examples are presented to demonstrate the convergence and generalization capabilities of this hybrid algorithm. Copyright © 2003 John Wiley & Sons, Ltd.     

Artificial neural networks: A new method for mineral prospectivity mapping

A multilayer feed‐forward neural network, trained with a gradient descent, back‐propagation algorithm, is used to estimate the favourability for gold deposits using a raster GIS database for the Tenterfield 1:100 000 sheet area, New South Wales. The database consists of solid geology, regional faults, airborne magnetic and gamma‐ray survey data (U, Th, K and total count channels), and 63 deposit and occurrence locations. Input to the neural network consists of feature vectors formed by combining the values from co‐registered grid cells in each GIS thematic layer. The network was trained using binary target values to indicate the presence or absence of deposits. Although the neural network was trained as a binary classifier, output values for the trained network are in the range [0.1, 0.9] and are interpreted to indicate the degree of similarity of each input vector to a composite of all the deposit vectors used in training. These values are rescaled to produce a multiclass prospectivity map. To validate and assess the effectiveness of the neural‐network method, mineral‐prospectivity maps are also prepared using the empirical weights of evidence and the conceptual fuzzy‐logic methods. The neural‐network method produces a geologically plausible mineral‐prospectivity map similar, but superior, to the fuzzy logic and weights of evidence maps. The results of this study indicate that the use of neural networks for the integration of large multisource datasets used in regional mineral exploration, and for prediction of mineral prospectivity, offers several advantages over existing methods. These include the ability of neural networks to: (i) respond to critical combinations of parameters rather than increase the estimated prospectivity in response to each individual favourable parameter; (ii) combine datasets without the loss of information inherent in existing methods; and (iii) produce results that are relatively unaffected by redundant data, spurious data and data containing multiple populations. Statistical measures of map quality indicate that the neural‐network method performs as well as, or better than, existing methods while using approximately one‐third less data than the weights of evidence method.     

Modeling soil collapse by artificial neural networks

The feasibility of using neural networks to model the complex relationship between soil parameters, loading conditions, and the collapse potential is investigated in this paper. A back propagation neural network process was used in this study. The neural network was trained using experimental data. The experimental program involved the assessment of the collapse potential using the one-dimensional oedometer apparatus. To cover the broadest possible scope of data, a total of eight types of soils were selected covering a wide range of gradation. Various conditions of water content, unit weights and applied pressures were imposed on the soils. For each placement condition, three samples were prepared and tested with the measured collapse potential values averaged to obtain a representative data point. This resulted in 414 collapse tests with 138 average test values, which were divided into two groups. Group I, consisting of 82 data points, was used to train the neural networks for a specific paradigm. Training was carried out until the mean sum squared error (MSSE) was minimized. The model consisting of eight hidden nodes and six variables was the most successful. These variables were: soil coefficient of uniformity, initial water content, compaction unit weight, applied pressure at wetting, percent sand and percent clay. Once the neural networks have been deemed fully trained its accuracy in predicting collapse potential was tested using group II of the experimental data. The model was further validated using information available in the literature. The data used in both the testing and validation phases were not included in the training phase. The results proved that neural networks are very efficient in assessing the complex behavior of collapsible soils using minimal processing of data. This revised version was published online in July 2006 with corrections to the Cover Date.     

An alternative data driven approach for prediction of thermal discharge initial dilution using tee diffusers

Mixing of heated water discharged from outfalls is an efficient and effective method of waste disposal in coastal areas. Discharging the heated water with large quantities of mass flux generally requires multi-port diffusers. In recent years, using numerical models to predict the plume behavior has received attention from many researchers, who are interested in design of outfalls. This study reports the development and application of an artificial neural network model for prediction of initial dilution of multi-port tee diffusers. Several networks with different structures were trained and tested using error back propagation algorithm. Statistical error measures showed that a three layer network with 9 neurons in the hidden layer is skillful in prediction of initial dilution and the outputs are in good agreement (R = 0.97) with experimental results. Furthermore, the sensitivity analyses showed that the width of the equivalent slot of the diffuser is the most important parameter in the estimation of initial dilution.     

Seismic human loss estimation for an earthquake disaster using neural network

In Iran, earthquakes cause enormous damage to the people and economy. If there is a proper estimation of human losses in an earthquake disaster, it could be appropriately responded and its impacts and losses will be decreased. Neural networks can be trained to solve problems involving imprecise and highly complex nonlinear data. Based on the different earthquake scenarios and diverse kind of constructions, it is difficult to estimate the number of injured people. With respect to neural network’s capabilities, this paper describes a back propagation neural network method for modeling and estimating the severity and distribution of human loss as a function of building damage in the earthquake disaster. Bam earthquake data in 2003 were used to train this neural network. The final results demonstrate that this neural network model can reveal much more accurate estimation of fatalities and injuries for different earthquakes in Iran and it can provide the necessary information required to develop realistic mitigation policies, especially in rescue operation.     

New Prediction Models for Mean Particle Size in Rock Blast Fragmentation

The paper refers the reader to a blast data base developed in a previous study. The data base consists of blast design parameters, explosive parameters, modulus of elasticity and in situ block size. A hierarchical cluster analysis was used to separate the blast data into two different groups of similarity based on the intact rock stiffness. The group memberships were confirmed by the discriminant analysis. A part of this blast data was used to train a single-hidden layer back propagation neural network model to predict mean particle size resulting from blast fragmentation for each of the obtained similarity groups. The mean particle size was considered to be a function of seven independent parameters. An extensive analysis was performed to estimate the optimum value for the number of units for the hidden layer for each of the obtained similarity groups. The blast data that were not used for training were used to validate the trained neural network models. For the same two similarity groups, multivariate regression models were also developed to predict mean particle size. Capability of the developed neural network models as well as multivariate regression models was determined by comparing predictions with measured mean particle size values and predictions based on one of the most applied fragmentation prediction models appearing in the blasting literature. Prediction capability of the trained neural network models as well as multivariate regression models was found to be strong and better than the existing most applied fragmentation prediction model. Diversity of the blasts data used is one of the most important aspects of the developed models.     

Toward real-time three-dimensional mapping of surficial aquifers using a hybrid modeling approach

A hybrid modeling approach is proposed for near real-time three-dimensional (3D) mapping of surficial aquifers. First, airborne frequency-domain electromagnetic (FDEM) measurements are numerically inverted to obtain subsurface resistivities. Second, a machine-learning (ML) algorithm is trained using the FDEM measurements and inverted resistivity profiles, and borehole geophysical and hydrogeologic data. Third, the trained ML algorithm is used together with independent FDEM measurements to map the spatial distribution of the aquifer system. Efficacy of the hybrid approach is demonstrated for mapping a heterogeneous surficial aquifer and confining unit in northwestern Nebraska, USA. For this case, independent performance testing reveals that aquifer mapping is unbiased with a strong correlation (0.94) among numerically inverted and ML-estimated binary (clay-silt or sand-gravel) layer resistivities (5–20 ohm-m or 21–5,000 ohm-m), and an intermediate correlation (0.74) for heterogeneous (clay, silt, sand, gravel) layer resistivities (5–5,000 ohm-m). Reduced correlation for the heterogeneous model is attributed to over-estimating the under-sampled high-resistivity gravels (about 0.5 % of the training data), and when removed the correlation increases (0.87). Independent analysis of the numerically inverted and ML-estimated resistivities finds that the hybrid procedure preserves both univariate and spatial statistics for each layer. Following training, the algorithms can map 3D surficial aquifers as fast as leveled FDEM measurements are presented to the ML network.     

Neural network modeling applications in active slope stability problems

A back propagation artificial neural network approach is applied to three common challenges in engineering geology: (1) characterization of subsurface geometry/position of the slip (or failure surface) of active landslides, (2) assessment of slope displacements based on ground water elevation and climate, and (3) assessment of groundwater elevations based on climate data. Series of neural network models are trained, validated, and applied to a landslide study along Lake Michigan and cases from the literature. The subsurface characterization results are also compared to a limit equilibrium circular failure surface search with specific adopted boundary conditions. It is determined that the neural network models predict slip surfaces better than the limit equilibrium slip surface search using the most conservative criteria. Displacements and groundwater elevations are also predicted fairly well, in real time. The models’ ability to predict displacements and groundwater elevations provides a foundational framework for building future warning systems with additional inputs.     

巷道围岩参数的人工神经网络预测

应用人工智能方法解决地下工程问题,提出了预测巷道围岩参数的人工神经网络预测法,构造了预测围岩参数的神经网络模型。预测结果证明,该模型具有很高的预测精度。提出的方法有一定的实用价值和参考价值。     

An intelligent neural network model for evaluating performance of immobilized cell biofilter treating hydrogen sulphide vapors

Biofiltration has shown to be a promising technique for handling malodours arising from process industries. The present investigation pertains to the removal of hydrogen sulphide in a lab scale biofilter packed with biomedia, encapsulated by sodium alginate and poly vinyl alcohol. The experimental data obtained under both steady state and shock loaded conditions were modelled using the basic principles of artificial neural networks. Artificial neural networks are powerful data driven modelling tools which has the potential to approximate and interpret complex input/output relationships based on the given sets of data matrix. A predictive computerised approach has been proposed to predict the performance parameters namely, removal efficiency and elimination capacity using inlet concentration, loading rate, flow rate and pressure drop as the input parameters to the artificial neural network model. Earlier, experiments from continuous operation in the biofilter showed removal efficiencies from 50 to 100 % at inlet loading rates varying up to 13 g H₂S/m³h. The internal network parameter of the artificial neural network model during simulation was selected using the 2^k factorial design and the best network topology for the model was thus estimated. The results showed that a multilayer network (4-4-2) with a back propagation algorithm was able to predict biofilter performance effectively with R² values of 0.9157 and 0.9965 for removal efficiency and elimination capacity in the test data. The proposed artificial neural network model for biofilter operation could be used as a potential alternative for knowledge based models through proper training and testing of the state variables.     

Seismic random noise attenuation using artificial neural network and wavelet packet analysis

In this paper, we present a method for attenuating background random noise and enhancing resolution of seismic data, which takes advantage of semi-automatic training of feed forward back propagation (FFBP) artificial neural network (ANN) in a multiscale domain obtained from wavelet packet analysis (WPA). The images of approximations and details of the input seismic sections are calculated and utilized to train neural network to model coherent events by an automatic algorithm. After the modeling of coherent events, the remainder data are assumed to be related to background random noise. The proposed method is applied on both synthetic and real seismic data. The results are compared with that of the adaptive Wiener filter (AWF) in synthetic shot gather and real common midpoint gather and also with that of band-pass filtering on real common offset gather. The comparison indicates substantially higher efficiency of the proposed method in attenuating random noise and enhancing seismic signals.     

基于WorldView-02高分影像的BP和RBF神经网络遥感水深反演

遥感水深反演是水深测量的一种重要技术和手段.以美济礁水深反演为例,选择WorldView-02高分影像为数据源,在辐射定标和大气校正的基础上,构建BP（Back Propagation）和RBF（Radial Basis Function）人工神经网络水深反演模型,以遥感影像8个波段为输入层,通过tansig、logsig、高斯函数和purelin函数变换实现从输入层到隐含层、隐含层到输出层的转换,以便反演水深.最后对反演水深与实测水深采用回归分析,求解决定系数（coefficient of determination,R2）、平均决定误差（Mean Absolute Error,MAE）、均方根误差（Root Mean Square Error,RMSE）等进行比较,评价2种模型的精度.结果表明,RBF神经网络模型结构更简单,对样本要求更低,反演精度达到0.995,更适合遥感水深反演.      

基于改进BP网络算法的隧洞围岩分类

围岩分类对指导地下工程的设计和施工具有非常重要的意义.引入人工神经网络的方法, 进行隧洞围岩分类, 在传统BP算法的基础上, 通过改进学习算法、优化传递函数和网络结构进行神经网络方法优化.采用附加动量法和学习速率自适应调整的策略改进学习算法, 使得当误差大于上临界值时, 则降低学习率, 当误差小于下临界值时, 则适当提高学习率, 这样可加快网络的训练速度, 确保网络的稳定性; 通过引入调整学习率参数, 使得传递过程更加敏感, 加快了传递函数的收敛速度, 提高了训练函数的计算精度; 通过给定隐含层节点模型的取值范围, 对网络结构进行优化, 提高了泛化精度.将改进的BP网络模型应用于广东省东深供水改造工程的隧洞围岩分类中, 分类结果与根据《水工隧洞设计规范(SL279-2002) 》的分类结果完全一致, 表明该方法具有良好的工程实用性.      

利用共轭梯度方法的激发极化三维快速反演

利用共轭梯度方法实现了激发极化(IP)三维快速反演。首先,利用共轭梯度方法反演电位数据,得到地下的三维电阻率模型,(由于避免了直接求偏导数矩阵,因此反演计算速度很快。)然后,以此电阻率模型为地下电导率分布,再反演IP数据得到三维极化率分布理论模型。试算结果表明其效果较好。      

地地震多参数BP神经网络预测煤层厚度

依据煤层反射波运动学和动力学特征，提取出了波峰波谷振幅A1、平均频率Fa、主频带能量Qf1、低频带宽能量Qf和峰值频率Fmain等5个地地震特征参数。选取8组学习样本，利用4层BP（Back Propagation)人工神经网络模型，采用动量法和自适应调整的改进算法，训练BP网络，用训练好的BP网络预测煤层厚度。经实例验证，地震多参数BP网络预测煤层厚度精度高，是一种有效的煤厚预测方法。