Building 2-D stratigraphic and structure models from well log data and control horizons

An algorithm to build a gridded 2-D seismic velocity (or any other physical property) model from well log data and control horizons is developed. Interpolation of well log data onto a 2-D grid uses inverse distance weighting or linear interpolation, guided by the shape of the control horizons that are predefined from seismic or other 2-D constraints. A key feature of the models is that they may contain layers that are truncated by unconformities or at faults, or that lap out smoothly at their tops and bottoms. Abrupt or smooth terminations are controlled by user flags. Applications are illustrated using resistivity and acoustic impedance log data from the Blake Ridge (offshore Carolina) and using a complex structure produced by overthrust tectonics in western Canada. Geologically reasonable models can be produced only if there are sufficient wells to sample every salient element in the model and sufficient control horizons to define the lateral character of the structures at the required level of detail.     

免疫遗传算法及其在波阻抗反演中的应用*

提出一种改进的免疫遗传算法（IGA）并用于波阻抗反演。IGA设计了基于抗体激励度和抗体浓度的免疫选择算子及加速全局收敛的非一致性变异算子,提出了抗体规模自适应调整机制,IGA能够改善传统非线性反演方法易早熟和局部收敛等弊端。通过理论模型试算表明,IGA具有较高的反演精度和反演效率;阳泉二矿实际资料反演所得剖面的纵向分辨率明显高于实际地震剖面,弱反射波的连续性和可检测性明显提高,表明改进的免疫遗传算法适用于煤炭地震波阻抗反演。     

Seismic monitoring of hydrocarbon fluid flow

Time-lapse 3-D seismic monitoring of subsurface rock property changes incurred during reservoir fluid-flow processes is an emerging new diagnostic technology for optimizing hydrocarbon production. I discuss the physical theory relevant for three-phase fluid flow in a producing oil reservoir, and rock physics transformations of fluid-flow pressure, temperature and pore-fluid saturation values to seismic P-wave and S-wave velocity. I link fluid-flow physical parameters to seismic reflection data amplitudes and traveltimes through elastic wave equation modeling and imaging theory. I demonstrate in a simulated data example that changes in fluid-flow can be monitored and imaged from repeated seismic surveys acquired at varying production calendar times.I would like to thank Amos Nur, Jack Dvorkin and James Packwood of the Stanford Rock Physics Laboratory for their help with the rock physics component of the synthetic data example. Sverre Strandenes and Norsk Hydro were very helpful in providing the reservoir geology information and fluid-flow simulation data. This work was supported by the Sponsors of the Stanford Exploration Project, under the Directorship of Prof. Jon Claerbout.     

Inversion of a velocity model using artificial neural networks

We present a velocity model inversion approach using artificial neural networks (NN). We selected four aftershocks from the 2000 Tottori, Japan, earthquake located around station SMNH01 in order to determine a 1D nearby underground velocity model. An NN was trained independently for each earthquake-station profile. We generated many velocity models and computed their corresponding synthetic waveforms. The waveforms were presented to NN as input. Training consisted in associating each waveform to the corresponding velocity model. Once trained, the actual observed records of the four events were presented to the network to predict their velocity models. In that way, four 1D profiles were obtained individually for each of the events. Each model was tested by computing the synthetic waveforms for other events recorded at SMNH01 and at two other nearby stations: TTR007 and TTR009.     

Petrophysical data prediction from seismic attributes using committee fuzzy inference system

This study presents an intelligent model based on fuzzy systems for making a quantitative formulation between seismic attributes and petrophysical data. The proposed methodology comprises two major steps. Firstly, the petrophysical data, including water saturation (S_w) and porosity, are predicted from seismic attributes using various fuzzy inference systems (FISs), including Sugeno (SFIS), Mamdani (MFIS) and Larsen (LFIS). Secondly, a committee fuzzy inference system (CFIS) is constructed using a hybrid genetic algorithms-pattern search (GA-PS) technique. The inputs of the CFIS model are the outputs and averages of the FIS petrophysical data. The methodology is illustrated using 3D seismic and petrophysical data of 11 wells of an Iranian offshore oil field in the Persian Gulf. The performance of the CFIS model is compared with a probabilistic neural network (PNN). The results show that the CFIS method performed better than neural network, the best individual fuzzy model and a simple averaging method.     

Application of a radial basis function artificial neural network to seismic data inversion

We investigate here the performance and the application of a radial basis function artificial neural network (RBF-ANN) type, in the inversion of seismic data. The proposed structure has the advantage of being easily trained by means of a back-propagation algorithm without getting stuck in local minima. The effects of network architectures, i.e. the number of neurons in the hidden layer, the rate of convergence and prediction accuracy of ANN models are examined. The optimum network parameters and performance were decided as a function of testing error convergence with respect to the network training error. An adequate cross-validation test is run to ensure the performance of the network on new data sets. The application of such a network to synthetic and real data shows that the inverted acoustic impedance section was efficient.     

Permeability prediction and construction of 3D geological model: application of neural networks and stochastic approaches in an Iranian gas reservoir

Determination of petrophysical parameters by using available data has a specific importance in exploration and production studies for oil and gas industries. Modeling of corrected permeability as a petrophysical parameter can help in decision making processes. The objective of this study is to construct a comprehensive and quantitative characterization of a carbonate gas reservoir in marine gas field. Artificial neural network is applied for prediction of permeability in accordance with other petrophysical parameters at well location. Correlation coefficient for this method is 84 %. In the study, the geological reservoir model is developed in two steps: First, the structure skeleton of the field is constructed, and then, reservoir property is distributed within it by applying new stochastic methods. Permeability is modeled by three techniques: kriging, sequential Gaussian simulation (SGS) and collocated co-simulation using modeled effective porosity as 3D secondary variable. This paper enhances the characterization of the reservoir by improving the modeling of permeability through a new algorithm called collocated co-simulation. Kriging is very simple in modeling the reservoir permeability, and also, original distribution of the data changes considerably in this model. In addition, the SGS model is noisy and heterogeneous, but it retains the original distribution of the data. However, the addition of a 3D secondary variable in third method resulted in a much more reliable model of permeability.     

Mapping of complex bedrock structure using the high-resolution reflection seismic technique

In May of 1996, four reflection seismic profiles and one VSP (vertical seismic profile) were acquired near Colonsay, Saskatchewan. This survey was conducted as part of a site-specific study of the high-resolution reflection seismic profiling, and shallow open-borehole VSP methods. The reflection seismic/VSP survey was designed with four specific objectives in mind. The sponsor wanted to determine: (1) whether high-resolution reflection seismic data could provide an essentially continuous structural image of the shallow subsurface (Quaternary alluvium and Cretaceous shale to a depth of ca 200 m); (2) if the Cretaceous bedrock and overlying Quaternary strata in the study area are extensively faulted; (3) whether useful VSP data could be acquired from shallow open-boreholes in the study area; and (4) optimum high-resolution reflection seismic acquisition parameters (given the target zone and site conditions) for future production-oriented work in the area. From a technical perspective the survey was successful. The acquired high-resolution reflection seismic data effectively image the shallow subsurface, and support the thesis that bedrock and overlying Quaternary strata are extensively faulted in the study area. The VSP data establish that sonic log velocities recorded above the water table in a fluid-filled borehole are anomalously high (relative to the seismic velocity of an equivalent thickness of unsaturated sediment). On the basis of the VSP control, the sonic log data were modified such that the corresponding synthetic seismograms correlate well with the reflection seismic profiles. Additionally, on the basis of field testing, optimum site/target specific field acquisition parameters were determined. These will be of use in designing any subsequent seismic reflection and VSP surveys in the study area.     

Improvement of the Umkehr ozone profile by the neural network method: analysis of the Belsk (51.80°N, 20.80°E) Umkehr data

Vertical profiles of atmospheric ozone by the neural network (NN) method are compared with those obtained by the standard Umkehr inversion algorithm – UMK92. Both methods used the same input, the so-called N values, derived from Umkehr measurements at Belsk (51.80°N, 20.80°E), Poland, by the Dobson spectrophotometer No 84. The vertical profiles of ozone from satellite observations, Microwave Limb Sounder (MLS) overpasses for the period 2004–2009, and from ozonesoundings performed at the nearby aerological station, Legionowo (52.4° N, 21.0° E), for the period 2000–2009 provide a reference data set for the NN model building. The NN methodology appears to be a promising tool for extracting information about the vertical ozone profile from ground-based Umkehr measurements, despite some limitations of the NN method itself, such as the results being limited to the analysed station, sensitivity to errors in the reference data sets, and lack of possibility to determine the actual retrieval errors. Accuracy of the NN ozone profiles is better for all Umkehr layers than that by the standard Umkehr inversion algorithm when NN and UMK92 profiles are compared with the reference profiles. It is especially pronounced for comparisons with the ozonesonde profiles for layers 4 and 1, where the absolute error changes from 10.6 Dobson units (DU) (UMK92) to 4.4 DU (NN) and from 6.6 DU (UMK92) to 3.5 DU (NN), respectively (1 Dobson unit is equal to 2.69 × 10²⁰ molecules/m²). The mean (over all Umkehr layers) correlation coefficient between NN-MLS, and NN-ozonesonde profiles is 0.75 and 0.85, respectively. The corresponding correlation coefficients for the comparison with UMK92 profiles are lower, i.e. 0.61 and 0.64, respectively.     

Wavelet extractor: A Bayesian well-tie and wavelet extraction program

We introduce a new open-source toolkit for the well-tie or wavelet extraction problem of estimating seismic wavelets from seismic data, time-to-depth information, and well-log suites. The wavelet extraction model is formulated as a Bayesian inverse problem, and the software will simultaneously estimate wavelet coefficients, other parameters associated with uncertainty in the time-to-depth mapping, positioning errors in the seismic imaging, and useful amplitude-variation-with-offset (AVO) related parameters in multi-stack extractions. It is capable of multi-well, multi-stack extractions, and uses continuous seismic data-cube interpolation to cope with the problem of arbitrary well paths. Velocity constraints in the form of checkshot data, interpreted markers, and sonic logs are integrated in a natural way.The Bayesian formulation allows computation of full posterior uncertainties of the model parameters, and the important problem of the uncertain wavelet span is addressed uses a multi-model posterior developed from Bayesian model selection theory.The wavelet extraction tool is distributed as part of the Delivery seismic inversion toolkit. A simple log and seismic viewing tool is included in the distribution. The code is written in Java, and thus platform independent, but the Seismic Unix (SU) data model makes the inversion particularly suited to Unix/Linux environments. It is a natural companion piece of software to Delivery, having the capacity to produce maximum likelihood wavelet and noise estimates, but will also be of significant utility to practitioners wanting to produce wavelet estimates for other inversion codes or purposes. The generation of full parameter uncertainties is a crucial function for workers wishing to investigate questions of wavelet stability before proceeding to more advanced inversion studies.     

The 3-D gravity inversion package GROWTH2.0 and its application to Tenerife Island, Spain

We present the gravity inversion software GROWTH2.0 and its application to recently obtained gravity data from the volcanic island of Tenerife (Canary Islands, Spain) to inform on its subsurface density structure. GROWTH2.0 is an inversion tool which enables the user to obtain, in a nearly automatic and nonsubjective mode, a 3D model of the subsurface density anomalies based on observed gravity anomaly data. The package is composed of three parts: (a) GRID3D to generate a 3D partition of the subsurface volume into parallelepiped elements, (b) GROWTH to perform the inversion routine and to obtain a 3D anomalous density model, and (c) VIEW for visual representation of the input data, the inversion model, and modeling residuals. The current version of the tool has been developed from an earlier code (Camacho et al., 2002) and now incorporates several novelties: (1) a Graphical User Interface (GUI), (2) an optional automated routine for determination of parameter λ, which controls the balance between model fitness and smoothness, (3) optional determination of values for minimum density contrast, (4) a robust handling of outlier data, and (5) improved automated data reduction for terrain effects based on anticorrelation with topographic data. The new capabilities and applicability of GROWTH2.0 for 3-D gravity inversion are demonstrated by a case example using new gravity data from the volcanic island of Tenerife. In a nearly automatic approach, the software provides a 3-D model informing on the location and shape of the main structural building blocks of the island. Our model results allow us to shed light on the low-density structure of the islands dominant Pico Viejo-Pico Teide (PV-PT) volcanic complex and the identification of an intrusive structure (the east bulge volcano) embedded in Teide's east flank. A low-density body located at around 5.8 km depth beneath PT's summit may represent a current magma or hybrid reservoir.     

3D object recognition method with multiple feature extraction from LiDAR point clouds

During autonomous driving, fast and accurate object recognition supports environment perception for local path planning of unmanned ground vehicles. Feature extraction and object recognition from large-scale 3D point clouds incur massive computational and time costs. To implement fast environment perception, this paper proposes a 3D recognition system with multiple feature extraction from light detection and ranging point clouds modified by parallel computing. Effective object feature extraction is a necessary step prior to executing an object recognition procedure. In the proposed system, multiple geometry features of a point cloud that resides in corresponding voxels are computed concurrently. In addition, a scale filter is employed to convert feature vectors from uncertain count voxels to a normalized object feature matrix, which is convenient for object-recognizing classifiers. After generating the object feature matrices of all voxels, an initialized multilayer neural network (NN) model is trained offline through a large number of iterations. Using the trained NN model, real-time object recognition is realized using parallel computing technology to accelerate computation.

     

Neural networks for the recognition and pose estimation of 3D objects from a single 2D perspective view

In this paper we present a neural network (NN) based system for recognition and pose estimation of 3D objects from a single 2D perspective view. We develop an appearance based neural approach for this task. First the object is represented in a feature vector derived by a principal component network. Then a NN classifier trained with Resilient backpropagation (Rprop) algorithm is applied to identify it. Next pose parameters are obtained by four NN estimators trained on the same feature vector. Performance on recognition and pose estimation for real images under occlusions are shown. Comparative studies with two other approaches are carried out.     

三维油藏彩色模型原理和算法

三维油藏彩色模型可真实直观地反三维地质构造的特点、渗透率和孔隙度等物性参数的分布规律。文中讨论了在ＳＵＮ工作站的Ｘ－Ｗｉｎｄｏｗｓ／Ｍｏｔｉｆ平台上建立２５６色三维油藏几何模型的基本方法,研究了如何建立三维网络化数据体,追踪可视面,对模型进行旋转和切割,以及着色、消隐等可视化技术的实现。     

Knowledge-based image analysis for geophysical interpretation

Geophysical seismic interpretation is part of geophysical oil prospecting. It evaluates and analyses seismic reflection data, aiming at the detection of the position of hydrocarbon reservoirs. This paper provides a review of current efforts to automate, at least partially, seismic interpretation. As will be shown, this research area is very active and is a melting pot of various different approaches and techniques: artificial intelligence, pattern recognition, image processing, graphics, fuzzy set theory and, of course, geophysics and geology. Some methods of seismic pattern recognition (e.g. remote correlation, fuzzy seismic modeling, recognition of reservoir boundaries) and of seismic image processing (horizon following, texture analysis) are presented and some applications are shown. Expert systems used in geophysical interpretation (mainly in well log interpretation) are also briefly described. Finally, an automated system for knowledge-based image analysis for geophysical interpretation is dicussed. Its low-level vision techniques, its knowledge representation, and the control strategy for seismic pattern search are described.     

Correlation of ternary liquid--liquid equilibrium data using neural network-based activity coefficient model

Liquid--liquid equilibrium (LLE) data are important in chemical industry for the design of separation equipments, and it is troublesome to determine experimentally. In this paper, a new method for correlation of ternary LLE data is presented. The method is implemented by using a combined structure that uses genetic algorithm (GA)--trained neural network (NN). NN coefficients that satisfy the criterion of equilibrium were obtained by using GA. At the training phase, experimental concentration data and corresponding activity coefficients were used as input and output, respectively. At the test phase, trained NN was used to correlate the whole experimental data by giving only one initial value. Calculated results were compared with the experimental data, and very low root-mean-square deviation error values are obtained between experimental and calculated data. By using this model tie-line and solubility curve data of LLE can be obtained with only a few experimental data.     

Three-Dimensional Feedforward Neural Networks and Their Realization by Nano-Devices

The three-dimensional (3D) model of a feedforward neural network(NN) based on so called N-hypercube topology isproposed. The N-hypercube is different from theclassical hypercube used in communication theory, and in Booleanalgebra. This new structure has been created based on a novelalgorithm for embedding a binary decision tree and binary decisiondiagram into a N-hypercube. It is shown thatN-hypercube topology is a reasonable solution toimplement NN of threshold gates, in particular, on thesingle-electron devices. The 3D design methodology of feedforwardNN is oriented to technology mapping to nanodevices. Results ofextensive experimental study of feedforward networks consistingof over 3500 N-hypercubes are presented.     

基于国产众核处理器三维地震声波正演模拟

三维地震声波理论与计算方法是地质勘探研究的基础,通过分析不同介质中声波的传播特性,完成三维地震声波正演模拟。针对三维地震声波有限差分交错网格方程正演过程中存在数值计算大、内存消耗大等实际问题,提出了基于神威·太湖之光超级计算机系统中国产异构众核处理器(申威26010)的三维地震声波正演模拟编程模型,完成了基于处理器间的进程级并行基于计算核心间的线程级并行优化策略。研究了DMA(直接内存读取)通信方式,提出2.5D流水线任务划分、通信与计算的相互掩盖的多角度优化策略。实验结果表明,该策略有效缓解了带宽瓶颈,发挥了处理器强大的计算能力,解决了程序在申威26010异构众核处理器处理有限差分问题时,并行效率过低的问题。在大规模测试下,使用266240个计算核心,程序仍能够保持稳定的计算性能,达到5.5 GFlops的场值更新。     

Robust error measure for supervised neural network learning withoutliers

Most supervised neural networks (NNs) are trained by minimizing the mean squared error (MSE) of the training set. In the presence of outliers, the resulting NN model can differ significantly from the underlying system that generates the data. Two different approaches are used to study the mechanism by which outliers affect the resulting models: influence function and maximum likelihood. The mean log squared error (MLSE) is proposed as the error criteria that can be easily adapted by most supervised learning algorithms. Simulation results indicate that the proposed method is robust against outliers.     

Synthetic retrieval of hourly net ecosystem exchange using the neural network model with combined MI and GOCI geostationary sensor datasets and ground-based measurements

Net ecosystem carbon dioxide (CO₂) exchange (NEE) is a key parameter for understanding the terrestrial plant ecosystems, but it is difficult to monitor or predict over large areas at fine temporal resolutions. In this research, we estimated the hourly NEE using a combination of the integrated neural network (NN) model with geostationary satellite imagery to overcome the limitations of existing daily polar orbiting satellite-derived carbon flux products. Two sets of satellite imageries (i.e. the meteorological imager (MI) and geostationary ocean colour imager (GOCI) aboard communication, ocean, and meteorological satellite (COMS)) and CO₂ flux data derived from eddy covariance measurements were used to verify the feasibility of applying hourly geostationary satellite imagery with an NN-based approach for estimating NEE at high temporal resolutions. For the NN model, the optimum neuronal architecture was established using an NN with one hidden layer that was trained using the Levenberg–Marquardt back propagation algorithm. The hourly NEE values estimated in test period from the NN model using the combined COMS MI and GOCI imagery and ground measurements as model inputs were compared with the eddy covariance NEE values from the measurement tower, which yielded reliable statistical agreement. The hourly NEE results from the NN model based on COMS MI and GOCI imagery and ground measurement data had the highest accuracy (RMSE = 2.026 μmol m^?2 s^?2, R = 0.975), while the root mean square error (RMSE) and the regression coefficient (R) generated by the NN model based on satellite imagery as the sole input variable were relatively lower (RMSE = 3.230 μmol m^?2 s^?2, R = 0.952). Although the simulations for the satellite-only NEE were showed as lower accuracy than the NN model that included all input variables, the hourly variations in NEE also appeared to describe its daily growth and development pattern well, indicating the possibility of deriving hourly-based products from the proposed NN model using geostationary satellite data as inputs.