专利汇可以提供Neural network interpretation of aeromagnetic data专利检索,专利查询,专利分析的服务。并且A method for determining depth to basement from aeromagnetic data utilizes neural networks (10) to automate the laborious process of profile interpretation. The neural networks provide consistency, accuracy and overall quality without bias of interpretation. A network is provided having input (12), intermediate (14) and output (16) layers. A two-dimensional profile magnetic basement model is formed, reprocessed aeromagnetic survey profiles are windowed for input to the network, the network is trained by weight adjustment using profiles from known basement structures, the trained network is tested and then applied to aeromagnetic profiles over an unknown earth formation.,下面是Neural network interpretation of aeromagnetic data专利的具体信息内容。
The present invention relates to a method for locating magnetic structural bodies within the earth and in particular to a method using neural networks for determining the subsurface location, geometry and depth to basement of these bodies from aeromagnetic data.
Determination of the location and the depth to basement of faults and structures in the earth is one of the most important reconnaissance exploration tools in the petroleum exploration industry. This is because many hydrocarbon reservoirs are associated with uplifted basement fault blocks and accurate mapping of subsurface structures greatly improves exploration well success rates and lowers the cost of finding hydrocarbons in sedimentary basins.
Subsurface geologic structure is commonly deduced from the mapping of surface structures and features, well log correlations, and by seismic reflection and refraction profiling. However, surface features do not always reflect deep structure when masked by surficial alluvium and moderately lithified shallow sediments. Also, the high cost of drilling deep exploratory well holes and collecting reflection seismic data often preclude their economic usefulness in delineating deep structures. Thus surveys which measure the magnetic field at or above the earth's surface, particularly from an airplane (aeromagnetics), can be an economic, environmentally attractive alternative to these other methods in unexplored or underexplored sedimentary basins.
Currently, aeromagnetic data are interpreted using combinations of simple computational and empirical techniques. The following Table I lists several of the manual and computer techniques currently used to predict location and depth of magnetic basement structural features.
However, these techniques are limited in their usefulness and accuracy largely due to the difficulty in differentiating between magnetic anomalies related to compositional contrasts and anomalies related to structural relief. Furthermore, the manual techniques are largely qualitative and yield approximate lateral locations and can not accurately determine depth or vertical relief of the deep structures. Because of these problems in magnetic interpretation and analysis, aeromagnetic data has been of limited use in the petroleum exploration industry today.
Deep basement structures can be accurately located and their shapes mapped from aeromagnetic data using neural network technology. Neural networks are pattern recognition tools that make use of parallel connections of simple non-linear functions. The simplifying assumptions and limitations required by current interpretational art is no longer necessary because the neural network learns the relationship between observed magnetic fields and deep basement structures. Additionally, once the network learns this relationship, it can accurately determine structure throughout the geologic province.
The invention concerns a method of accurately determining the depth, geometry, and location of basement structures from aeromagnetic data using neural networks. It comprises forward modeling of a given basement response on a computer and inputting this computed response into a specially designed backpropagation neural network for the training phase of the invention. What this means is that the neural network "learns" the appropriate magnetic response to a given basement configuration. Once trained, the neural network is then applied to the remaining aeromagnetic data in the area to produce the basement structure for interpretation.
The present invention will now be described, by way of example, with reference to the accompanying drawings, in which:
Heretofore the known method for aeromagnetic processing and interpretation for basement structures has involved a number of qualitative and quantitative procedures which are highly correlated and not particularly accurate. These procedures would typically involve some forward modeling on a computer to calculate the magnetic response of a desired basement structure. This response would then be qualitatively correlated to the aeromagnetic data to see if a similar response exists on the observed data. This procedure would be accompanied by some inverse modeling to determine if an interpreted magnetic anomaly could be caused by a basement structure. These procedures would be iterated until a decision is reached, in concert with the above, and the mapped aeromagnetic data would be run through a series of mathematical filters (i. e. 2nd vertical derivative, horizontal derivative, high pass, structural inversion, etc.). Also, to accentuate structural relief, the map procedures would then be correlated to the profile procedures and the best guess would be made as to the basement configuration. Clearly, this is a time-consuming, laborious process and only yields a qualitative approximation, at best. As a result, aeromagnetic data has been considered to be of limited value to the petroleum exploration industry.
A different approach is employed in accordance with the present invention. For the present invention, highly accurate and more rapid reduction of aeromagnetic data for delineating basement structure is achieved through the use of neural networks. The neural network 10 (Fig. 1) is designed with three layers, an input layer 12 which contains the aeromagnetic data, an intermediate or hidden layer 14, and an output layer 16 which contains the information to be learned. All of the layers are fully connected to one another. The network preferably contains nine input elements, five hidden elements, and nine output elements to distinguish between compositional bodies and structural bodies on the aeromagnetic data. It should be here noted that one skilled in the art may come up with many variations on the network design in order to achieve comparable results without departing from the spirit or essential characteristics of the present invention. For this example of the invention, the input variables included total field intensity, lagged total field intensity and various transforms from both real and synthetically generated data. What this means is that the total magnetic intensity samples were shifted up and down around a magnetic sample creating a window of data around a central point so that the neural network could "sense" both the amplitude of the anomaly as well as the frequency.
The computer generated or synthetic magnetic profile responses were used to "train" the neural network to recognize a variety of possible structural features anticipated to be found in the real data. During training, the neural network was given the synthetic data, asked to analyze it and predict the structure. This predicted structure was then compared with the structure used to generate the synthetic data, also lagged in the same manner as the total magnetic intensities, and the connection weights were adjusted to minimize the difference between the predicted and actual model.
In backpropogation, the responsibility for reducing output error is shared among all of the connection weights. In this invention, the well known Delta Rule is used for weight adjustment during learning. The global error function to be minimized is defined as
where the subscript k refers to the kth output node, Dk is the desired output, and Ok the actual output from the kth output node.
The global error is then redistributed throughout the network according to
where Ej(s) is the error assigned to the jth node in the sth layer. The connection weights are then adjusted according to
where 0<lcoef<1 is the learning coefficient.
It is the connection weight values at the end of training that determine the quality of the network for basement relief mapping.
This was done until the difference between the predicted and actual structure reached an acceptable tolerance, usually around 8,000 passes of the data. Once training of the neural network was completed, the network was then rigorously tested against known structural anomalies on aeromagnetic data, as well as other training data, to insure the accuracy of the results.
The procedure for deriving subsurface structure of the magnetic basement from recorded aeromagnetic profile data requires six steps.
This six step procedure of network training using computer generated synthetic data from models characterizing the basin and network application to the analysis of field recorded aeromagnetic data from the same basin is demonstrably quicker and more accurate than is possible with the heretofore known practices.
Figure 3 is a typical magnetic profile model across a known oil field. Figure 4 is a theoretical structure model including six small basement structures and two intrabasement compositional bodies and used to train the subject neural network. Figure 5 is the result of applying the network to the training model. Figure 6 is the result of applying the network to an actual aeromagnetic profile across the test field producing basement relief.
The neural network system has proven to be a fast, accurate, and objective method for recognition of magnetic structural anomalies even in the presence of noise and intrabasement signal.
The present embodiment is intended in all respects to be illustrative and not restrictive as to the scope of the present invention as defined by the appended claims.
标题 | 发布/更新时间 | 阅读量 |
---|---|---|
基于自适应深度学习的新能源锂电池表面缺陷检测方法 | 2020-05-08 | 358 |
一种生成带深度信息的全景图的方法、装置及存储介质 | 2020-05-08 | 174 |
基于深度神经网络的风力涡轮机齿轮箱故障诊断方法 | 2020-05-11 | 28 |
一种书法字体类型与文字内容同步识别方法 | 2020-05-08 | 727 |
一种基于知识注意力的细粒度图像分类方法 | 2020-05-08 | 939 |
一种基于卷积神经网络的管道缺陷深度的反演方法 | 2020-05-11 | 671 |
一种视频业务播放卡顿的识别方法和装置 | 2020-05-08 | 647 |
一种基于振动机理与深度学习技术的整周期无键相监测方法 | 2020-05-08 | 954 |
一种基于概率分布自适应的实体关系分类方法 | 2020-05-08 | 748 |
基于深度学习的纵膈淋巴结超声弹性图像分割分类方法 | 2020-05-11 | 704 |
高效检索全球专利专利汇是专利免费检索,专利查询,专利分析-国家发明专利查询检索分析平台,是提供专利分析,专利查询,专利检索等数据服务功能的知识产权数据服务商。
我们的产品包含105个国家的1.26亿组数据,免费查、免费专利分析。
专利汇分析报告产品可以对行业情报数据进行梳理分析,涉及维度包括行业专利基本状况分析、地域分析、技术分析、发明人分析、申请人分析、专利权人分析、失效分析、核心专利分析、法律分析、研发重点分析、企业专利处境分析、技术处境分析、专利寿命分析、企业定位分析、引证分析等超过60个分析角度,系统通过AI智能系统对图表进行解读,只需1分钟,一键生成行业专利分析报告。