• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A method of identifying geological features, such as faults, of performing edge detection to locate a set of discontinuities in a seismic data set. The method involves classifying, using a neural network, each of the set of discontinuities as a seismic fault or non-seismic fault. The method also includes determining the positions of discontinuities classified as seismic faults.
    公开号:FR3070208A1
    申请号:FR1856617
    申请日:2018-07-17
    公开日:2019-02-22
    发明作者:Youli Mao；Bhaskar Mandapaka；Ashwani Dev；Satyam Priyadarshy
    申请人:Landmark Graphics Corp；
    IPC主号:
    专利说明:

    DETECTION OF UNDERGROUND FAULTS ON THE BASIS OF SEISMIC DATA INTERPRETATION
    CONTEXT
    The invention generally relates to the field of underground fault detection, and more particularly the detection of underground faults based on the interpretation of seismic data.
    Interpreting seismic data improves understanding of underground geological features (e.g. faults, fractures, etc.). These seismic interpretations can provide the position and shape of these underground geological features. In turn, knowing the position and shape of these underground geological features can increase the production of hydrocarbons during drilling. For example, the location of the drilling, various drilling parameters, production parameters, characterization and classification of the drilling project, etc. can be determined based on knowledge of the position and shape of these underground geological features. By increasing the accuracy and speed of seismic interpretation through the use of fault interpretation algorithms, significant gains in terms of efficiency, economy and security can be realized.
    The complexity of the seismic data can lead to fault interpretation workflows which include several operations involving significant manual effort and / or a large number of parameters. In addition, each operation can involve significant human intervention, such as testing many different parameters in these algorithms to determine their effects, classifying several types of features detected, and verifying that an algorithm is accurate during post-processing. These factors can increase the time and IT costs of the seismic interpretation operation and reduce the accuracy of the resulting interpretations.
    BRIEF DESCRIPTION OF THE FIGURES
    The embodiments of the invention can be better understood by consulting the diagrams attached.
    2017-IPM-101591-U1-EN
    FIG. 1 is a schematic diagram of an elevation view of a typical marine seismic survey which can be used to provide seismic data, according to some embodiments.
    FIG. 2 shows a schematic diagram of an earthquake earthquake survey environment, according to certain embodiments.
    FIG. 3 represents a set of seismic data and a set of seismic data detected by the edges, according to certain embodiments.
    FIG. 4 represents a 5 × 5 subsample of a set of seismic data detected by the edges and the corresponding probability of fault, according to certain embodiments.
    FIG. 5 represents a neural network applied to a seismic data set, according to certain embodiments.
    FIG. 6 represents a convolutional neural network applied to a seismic data set, according to certain embodiments.
    FIG. 7 represents an automated flaw interpretation workflow, according to certain embodiments.
    FIG. 8 represents a comparison between a dataset marked by an expert and a dataset marked automatically, according to certain embodiments.
    FIG. 9 shows an example of a drilling system near a fault, according to certain embodiments.
    FIG. 10 shows an example of a wellbore system near a fault, according to certain embodiments.
    FIG. 11 shows an example of a computer system, according to certain embodiments.
    DESCRIPTION
    The following description includes examples of systems, methods, techniques, and program streams that represent embodiments of the invention. However, it is understood that this invention can be practiced without these specific details. Through
    2017-IPM-101591-U1-EN example, this invention refers to convolutional neural networks. Aspects of this invention can also be applied to other machine learning methods, such as traditional neural networks, backpropagation neural networks, and recurrent neural networks. In other cases, well-known instructional instances, protocols, structures and techniques have not been shown in detail so as not to obscure the description.
    Various embodiments relate to an automated fault interpretation tool which provides seismic interpretations. The automated fault interpretation tool can be based on a deep learning fault interpretation process which can include a classification algorithm based on deep learning to identify geological features such as faults in seismic volumes. based on a set of data sets detected by the edges. The methods allow automated processing of ever-increasing volumes of seismic data with greater efficiency and accuracy.
    In some embodiments, the method of interpreting faults by deep learning includes an edge detection method and a neural network method. After converting original seismic datasets to a spatially digitized dataset, such as a pixel-based dataset, the spatially scanned dataset is processed with an edge detection method to generate a data set detected by the edges. The edge detection method can capture discontinuities such as faults and fractures from the spatially digitized data set and incorporate these discontinuities into the edge detected data set. In some embodiments, these may be the target features of interest. In addition to faults and fractures, non-target features such as signal reflectors with strong discontinuity signals can also be captured by the edge detection process and incorporated into the edge detected dataset. A neural network and / or a deep neural network (for example a neural network with several layers between an input and an output) can be applied to the data set detected by the edges to classify the characteristics of the data set detected by the edges to eliminate non-target features and keep target geological features, such as faults and fractures, to generate a filtered set of data detected by the edges. The deep learning fault interpretation method can be applied to any number of seismic datasets to provide seismic interpretation results that efficiently and accurately reveal geological features such as fractures and faults . In some
    2017-IPM-101591-U1-EN embodiments, these seismic interpretations can be performed in real time or briefly after the acquisition of a seismic data set (for example, within ten minutes of the measurement or acquisition a seismic data set to be interpreted).
    In some embodiments, the seismic interpretation results can be used to plan drilling direction or well stimulation treatments. For example, fault locations can be used to determine a drill plan to ensure that the borehole does not drill into a fault. In some embodiments, the location of the fault can be incorporated into a drilling control system to automatically prevent a borehole from drilling near a limit near the fault line. In addition, the location of faults can determine the parameters of a stimulation treatment so that the stimulation does not damage or puncture the geological environments near a fault.
    Example of a seismic data acquisition system
    FIG. 1 is a schematic diagram of an elevation view of a typical marine seismic survey which can be used to provide seismic data, according to some embodiments. A body of water 101 above the earth 102 is delimited on the surface of the water 103 by a water-air interface and at the bottom of the water 104 by a water-earth interface. Under the bottom of water 104, earth 102 contains underground formations of interest. A seismic vessel 105 moves on the surface of the water 103 and contains seismic acquisition control equipment, generally designated by 106. The seismic acquisition control equipment 106 comprises a navigation control, a source control seismic, seismic sensor control and recording equipment.
    The seismic acquisition control equipment 106 causes a seismic source 107 towed into the body of water 101 by the seismic vessel 105 to be actuated at selected times. The seismic streamers 108 contain sensors for detecting the reflected wave fields triggered by the seismic source 107 and reflected from interfaces in the environment. The seismic streamers 108 may contain pressure sensors such as hydrophones 109 and / or motion sensors of water particles such as geophones 110. Hydrophones 109 and geophones 110 are typically collocated in pairs or pairs of networks of sensors at regular intervals along the seismic streamers 108.
    The seismic source 107 is activated at periodic intervals to emit acoustic waves in the vicinity of the seismic streamers 108 with its sensors 109 and 110.
    2017-IPM-101591-U1-EN
    Each time the seismic source 107 is actuated, an acoustic wave field moves up or down in spherical expanding wave fronts. The moving wave fields will be illustrated by ray paths perpendicular to the expanding wave fronts. The wave field moving down from the seismic source 107, in the ray path 113, will be reflected on the earth-water interface at the bottom of the water 104, then will move up , as in ray path 114, where the wave field can be detected by hydrophones 109 and geophones 110. Such reflection at the bottom of water 104, as in ray path 114, contains information on the bottom of water 104 and can therefore be saved for further processing. In addition, the downward moving wave field, as in ray path 113, can cross the bottom of water 104 as in ray path 118, reflect a layer boundary, such as 116, and then move upward, as in ray path 117. The upwardly moving wave field, ray path 117, can then be detected by hydrophones 109 and geophones 110. Such reflection from a layer boundary 116 may contain useful information on underground formations of interest which can be used to generate seismic data.
    FIG. 2 shows a schematic diagram of an earthquake earthquake survey environment, according to certain embodiments. The seismic receivers 202 are in a spaced arrangement within a borehole 203 for detecting seismic waves. As shown, the receivers 202 can be fixed in place by anchors 204 to facilitate the detection of seismic waves. In different embodiments, the receivers 202 can be part of a chain of well logging tools (LWD) or a chain of cable log tools. Furthermore, the receivers 202 communicate wirelessly or by cable with a data acquisition unit 206 at the surface 205, where the data acquisition unit 206 receives, processes and stores seismic signal data collected by the receivers. 202. To generate seismic signal data, the surveyors trigger a seismic energy source 208 at one or more positions to generate seismic energy waves which propagate through the earth 210. Such waves are reflect from the acoustic impedance discontinuities to reach the receivers 202. Illustrative discontinuities include faults, the boundaries between the formation beds and the boundaries between the formation fluids. The discontinuities can appear as luminous points in the underground representation of the structure derived from the seismic signal data. The collected seismic signal data can be used to generate a seismic data set.
    2017-IPM-101591-U1-EN
    Examples of operations
    FIG. 3 represents a set of seismic data and a set of seismic data detected by the edges, according to certain embodiments. The seismic data set 302 can be processed with a phase congruence algorithm. The data stored in the seismic data set 302 can be converted from an alternative data format such as a SEGY file format by preprocessing. For example, the data stored in the SEGY file format can be preprocessed in a spatially digitized data set such as the pixel-based data format represented in the seismic data set 302. In this conversion, a bijective modeling is constructed between the magnitude of each of a set of seismic cross sections (for example line cross sections and crossed cross sections) and a pixel value in an image. Other embodiments of the seismic data set 302 may include other spatially digitized data sets such as voxel-based data sets.
    In some embodiments, discontinuities can be better captured by signals at a particular frequency. For example, a discontinuity in a seismic data set can be captured more precisely by high frequency signals. A set of bandpass filters can be applied to collected seismic signals to isolate or capture more details on discontinuities such as fractures and faults. As an illustrative example, a low bandpass with a frequency of 5-10 Hz and a high bandpass with a frequency of 60 Hz can be applied to a seismic data set to better capture larger or larger discontinuities small.
    The seismic dataset 302 can be processed by an edge detection algorithm to produce a seismic dataset detected by the edges 304. For example, a pixel-based phase congruence algorithm can be used to identify corners and edges from pixel images to capture discontinuous features such as unmarked fracture 306 and unmarked fault 312. In addition, the phase congruence algorithm can also capture the unmarked reflector 310, which may have a sharp edge. In some embodiments, continuous events such as structural and stratigraphic features in the seismic data 302 can be eliminated gradually by the edge detection algorithm. In some embodiments, the phase congruence algorithm can capture discontinuities that could be missed by other discontinuity detection methods such as discontinuity detection algorithms based on machine learning.
    2017-IPM-101591-U1-EN
    FIG. 4 represents a 5 × 5 subsample of a set of seismic data detected by the edges and the corresponding probability of fault, according to certain embodiments. For example, with reference to FIG. 3, the set of seismic data detected by the edges 304 may have dimensions in pixels of 1,301 pixels by 1,889 pixels and a corresponding marking network of equal size. The formation and processing of a large pixel image can have significant digital costs and pose classification problems. One of these challenges is that each fault segment can be unique and that it can be difficult to assign fault segments to a certain class. For calculation efficiency, a marked seismic data set can be subsampled into 5x5 pixel images with their corresponding markings.
    In addition to subsampling, a binary classification scheme can be applied to increase efficiency in a neural network filter. For example, a marking for a whole subsampled volume or image can be assigned on the basis of all the markings of its constituent elements. For example, with reference to FIG. 3, the seismic data set detected by edges 304 can be divided into a series of subsamples to more effectively form a neural network to perform tagging based on the probability of fault. The sub-sample can be the image of 5x5 pixels 402 and can have a corresponding pixel in the network for marking sub-samples 5x5 404, in which each element of the network for marking sub-samples 5x5 404 has a value of 0 to 1 to represent the probability of a discontinuity for its corresponding pixel. The 5x5404 sub-sample marking network can receive a subset marking for the entire 5x5404 sub-sample marking network. The subset marking scheme can be a quantitative scheme, a scheme for multiclass classification (three or more possible classifications), or a binary classification scheme (only two possible classifications). For example, a binary classification scheme can be used with a threshold value of 0.5 and compared with the arithmetic mean of the 5x5,404 subsample marking network to determine whether the subsample should be marked as a target characteristic. (for example a geological fault) or not. If the arithmetic mean is greater than 0.5, the 5x5 pixel 402 image would be classified as a geological fault. Otherwise, the 5x5 pixel 402 image would not be classified as a geological fault. In some embodiments, a neural network can be formed and validated with a binary classification scheme with greater efficiency and accuracy than a quantitative or multiclass classification scheme.
    2017-IPM-101591-U1-EN
    The sub-sample marking networks can be combined into a collective sub-sample image and a collective sub-sample marking network. In some embodiments, the collective subsample image can be used in a neural network such as a convolutional neural network. In some embodiments, for a convolutional neural network, a data set can be subsampled before the application of a convolutional filter or after the application of a convolutional filter.
    In other embodiments, the dimensions of the marking array can have any arbitrary dimensions, such as 2 pixels by 4 pixels, 10 pixels by 10 pixels, or 30 pixels by 20 pixels. In alternative three-dimensional embodiments, a marking network may also have three dimensions. For example, the marking array may have dimensions of arbitrary size smaller than the size of a complete set of seismic data, such as 2x2x2 pixels, 3x3x2 pixels or 10x20x25 pixels. In other embodiments, instead of comparing the average of the values of a sub-sample marking network to a threshold, other methods of reducing the size of a data set can be used, for example example by determining a weighted average, a median value or a random value. Other embodiments may also determine that a value below a threshold is a criterion for classifying a characteristic as a target characteristic. Alternative embodiments may also include a reflector or other geological feature other than faults as a target feature, and fractures / faults as non-target features.
    FIG. 5 represents a neural network applied to a seismic data set, according to certain embodiments. FIG. 6 shows an example of a convolutional neural network process 600. A seismic dataset such as the seismic dataset 602 can be partitioned into networks of subsets 504. Each of the networks of subjobs 504 can be quantified or classified with a value . In some embodiments, the value is based on whether an average or a weighted average of a fault likelihood estimate for each element (for example a pixel) in the subjacent arrays 504 is greater than or less than a threshold. In other embodiments, a value for each element in the subset networks 504 can be used directly as a training or validation set and processed with the neural network 506.
    The neural network 506 can generate a set of interpreted seismic data 508, in which target discontinuities such as fractures and faults
    2017-IPM-101591-U1-FR can be identified. In some embodiments, the seismic dataset 502 can be processed with an edge detection algorithm and in some embodiments, the seismic dataset and its corresponding seismic dataset detected by the edges can both be used as inputs to the neural network 506 to generate the interpreted seismic data set 508.
    FIG. 6 represents a convolutional neural network applied to a seismic data set, according to certain embodiments. FIG. 6 shows an example of a convolutional neural network process 600. A seismic dataset such as the seismic dataset 602 can be processed in the grouped convoluted datasets 608. In some embodiments, the seismic dataset 602 can be processed by an edge detection algorithm before being processed by the convolutional neural network. Each dataset of the pooled convoluted datasets 608 can be generated by a separate convolution filter, in which each convoluted dataset can be based on convoluted samples. For example, a network of subsets 604 can be convoluted by the convolution filter 605 in the convolution dataset 606, in which the convolution dataset 606 is one of a series of datasets that include a dataset of the pooled convoluted datasets 608.
    The pooled convolved data sets 608 can then be subsampled in the subsampled convolutional data set 614 and assigned to an appropriate subsampling marking network using the methods described above. For example, a subsample 610 can be reduced by the subsampling filter 611 in the subsampled convolutional sample 612.
    Once the pooled subsampled convolutional datasets 614 have been generated, the subsampled convolutional datasets 614 can be processed by one or more layers of activation function units such as rectified linear units (ReLU) 616. The output of ReLU 316 units can then be processed by a layer of weakly maximal units 618 to produce a reduced dimensional vector which can be used to classify various characteristics. For example, with reference to FIG. 3, after appropriate training, the reduced dimensional vector can be used to classify the characteristics in the seismic data set detected by the edges 304 and mark the unmarked reflector 310 as "not a fault" and mark the unmarked fault 312 as " fault ".
    2017-IPM-101591-U1-EN
    FIG. 7 represents an automated flaw interpretation workflow, according to certain embodiments. After first acquiring seismic data such as a set of two-dimensional seismic images which can be stacked to form a three-dimensional seismic volume 702, an edge detection algorithm such as a phase congruence algorithm can be used to process seismic data to identify discontinuities. For example, the three-dimensional seismic volume 702 can be processed by the edge detection algorithm 704 to generate a volume extracted by the edges 706.
    The volume extracted by the edges 706 can be processed by a deep neural network filter such as the convolutional neural network filters 708 to reduce the volume extracted by the edges 706 to a denoised volume 710. In certain embodiments, the neural network deep can use a supervised learning approach based on learning datasets. In some embodiments, the training datasets can be based on the manual interpretation of seismic data with fractures / faults marked by experts in the human domain. In some embodiments, the training data sets can be generated by software and based on fault probability algorithms such as a pretend-based algorithm. In some embodiments, the marked data sets may include data assigned to pixels of a seismic data set, in which the data contains binary markings such as "flaw" or "not a flaw" for each pixel of the seismic dataset.
    The deep neural network can be used for various classification purposes. In some embodiments, deep neural network filters can mark imaging artifacts such as reflectors and outliers. In addition, deep neural network filters can classify fractures / faults in the denoised volume. Deep neural network filters can assign a “fracture / fault” or “no fracture / fault” marking to each element of a seismic dataset, for example each pixel of a two-dimensional dataset, and remove it of all elements not assigned as fracture / fault.
    Based on the values of the denoised volume, a data modeling / migration method 712 can be applied to combine the fractures / faults identified in the denoised volume 710 to generate an interpreted seismic volume 720. In certain embodiments, the modeling of data may involve the superimposition of the output of the denoised volume on the interpreted seismic volume. In some embodiments, intermediate processing may take place before or as part of the modeling of
    2017-IPM-101591-U1-FR data to shrink, connect, extend or otherwise clarify the fault geometry interpreted in the denoised volume.
    FIG. 8 represents a comparison between a dataset marked by an expert and a dataset marked automatically, according to certain embodiments. After the formation of the convolutional neural network of an automated fault interpretation system with a data set marked by an 802 expert, the automatically marked data set for the automatic interpretation of fault 850 can provide results similar to the set of data marked by an 802 expert, in which similarity can be defined as a 10% pixel difference between the two images.
    Example of a drilling system
    FIG. 9 shows an example of a drilling system near a fault, according to certain embodiments. FIG. 9 shows a drilling system 900. The drilling system 900 includes a drilling installation 901 situated on the surface 902 of a borehole 903. The initial location of the borehole 903 and various operational parameters (for example speed of drilling, drill bit weight, drilling fluid flow, direction of drilling, composition of drilling fluid) for drilling can be selected based on the results of operations using seismic interpretation (as described above). For example, the location of borehole 903 can be selected to avoid the faults identified in the operations described above. Drill stand 904 can be operated to drill borehole 903 through underground formation 932 with the downhole assembly (BHA).
    The BHA includes a drill bit 930 at the downhole end of the drill string 904. The drill bit is in the vicinity of fault 975, the position of fault 975 being determined by seismic interpretation. The BHA and the drill bit 930 can be coupled to the computer system 950, which can operate the drill bit 930 as well as data received based on the sensors attached to the BHA. The drill bit 930 can be operated to create the borehole 903 by entering the surface 902 and into the subterranean formation 932. In some embodiments, a drilling plan may require that the drill bit 930 stop drilling when '' it is in a range of fault 975. By increasing the accuracy of seismic interpretation, drill bit 930 can more easily and safely avoid entering through fault 975. For example, sensors on the BHA can transmit a signal to the 950 computer system that the drill bit is close to a fault, and the computer system can stop the 930 drill bit.
    2017-IPM-101591-U1-EN
    Example of a wellbore system
    FIG. 10 shows an example of a wellbore system near a fault, according to certain embodiments. A wellbore system 1000 shown in Figure 10 includes a wellbore 1004 entering at least part of an underground formation 1002. The wellbore 1004 includes one or more injection points 1014 in which one or more fluids may be injected from wellbore 1004 into the underground formation 1002. The underground formation 1002 may include pores initially saturated with reservoir fluids (eg, oil, gas and / or water). In some embodiments, the wellbore system 1000 can be treated by injecting a fracturing fluid, an acid, or a proppant at one or more injection points 1014 into the well. of drilling 1004. In certain embodiments, the injection point (s) 1014 may correspond to the injection points 1014 in an envelope of the drilling well 1004. When the fluid enters the underground formation 1002 at the level of the injection 1014, one or more fractures 1018 can be opened. In some embodiments, a diverting agent can enter injection point 1014 and restrict the flow of another fluid. In some embodiments, the fracturing fluid may include a deflector.
    As shown in FIG. 10, the underground formation 1002 comprises at least one fracture network 1008 connected to the wellbore 1004. The fracture network 1008 represented in FIG. 10 contains a number of junctions and fractures 1018. The number of Junctions and fractures can vary drastically and / or unpredictably depending on the specific characteristics of the underground formation 1002. For example, the network of fractures 1008 can include on the order of thousands of fractures 1018 to tens of thousands of fractures 1018. In certain embodiments, these fractures can be in the range of a fault 1075, the position, the orientation and / or the shape of the fault 1075 being determined in the operations described above.
    In some embodiments, an operational parameter may include one or more wellbore processing regulations and / or wellbore production regulations. These operational parameters can be selected to avoid the faults identified in the operations described above. In some embodiments, the wellbore processing commands may characterize a processing operation for a wellbore 1004 penetrating at least a portion of an underground formation 1002. In some embodiments, the operational parameters may include , without there
    2017-IPM-101591-U1-EN limit, an amount of acid, fracturing fluid or deflector pumped into the wellbore system 1000, a concentration of proppant pumped into the wellbore system 1000 , a size of support agent during pumping in the wellbore system 1000, a wellbore pressure at the injection points 1014, a fluid or deflector flow rate at the inlet of the wellbore 1010, the pressure at the inlet of the wellbore 1010, a duration of an acidification / stimulation treatment, a particle diameter of the deflector and any combination thereof. In some embodiments, in response to calculations determining that a fracturing or acidification operation damages or punctures fault 1075, an operational parameter can be changed to prevent damage / puncture from occurring. For example, a computer system can determine that a set of operational parameters will damage the 1075 fault and, in response, reduce a flow of fluid to the 1010 surface.
    In some embodiments, the operational parameter (s) can be changed in response to real-time measurements. In some embodiments, the real-time measurements include pressure measurements, flow measurements and seismic measurements. In some embodiments, real-time measurements can be obtained from one or more sources or sensors of well location data in acoustic communication with the underground formation 1002. Data sources of well locations may include, but are not limited to, flow sensors, pressure sensors, thermocouples, and any other suitable measuring device. In some embodiments, well location data sources may be positioned on the surface, on a downhole tool, in wellbore 1004 or in fractures 1018. Pressure measurements may, for example , be obtained from a pressure sensor on a surface of wellbore 1004.
    Example of computer system
    FIG. 11 shows an example of a computer system, according to certain embodiments. A computer device 1100 includes a processor 1101 (possibly including several processors, several cores, several nodes, and / or implementing multi-thread processing, etc.). The computing device 1100 includes a memory 1107. The memory 1107 can be a system memory (for example, one or more of a cache memory, SRAM, DRAM, RAM without capacitor, double transistor RAM, eDRAM, EDO RAM, DDR RAM , EEPROM, NRAM, RRAM, SONOS, PRAM, etc.) or any one or more of the possible embodiments already described above of machine-readable media. The
    2017-IPM-101591-U1-FR IT device 1100 also includes a 1103 bus (for example, PCI, ISA, PCI-Express, HyperTransport® bus, InfiniBand® bus, NuBus, etc.) and a 1105 network interface (for example , a Fiber Channel interface, an Ethernet interface, a small computer system interface, a SONET interface, a wireless interface, etc.).
    In some embodiments, the computing device 1100 includes an edge detector 1111. The edge detector 1111 can perform one or more operations to detect discontinuities in the seismic data set, including operations to apply a phase congruence algorithm (as described above). Neural network processor 1112 can perform one or more operations to classify and filter a seismic dataset, including operations to apply a convolutional neural network to classify discontinuities as faults / fractures and remove discontinuities that are not faults / fractures (as described above). The operational parameter regulator 1113 can perform one or more operations for controlling a drilling system or a wellbore system, including controlling a bit rate of a drill bit or of a fluid pump. Any of the previously described functionalities can be partially (or entirely) implemented in the hardware and / or on the processor 1101. For example, the functionality can be implemented with an application-specific integrated circuit, in a logic implemented in the processor 1101, in a coprocessor on a peripheral or a card, etc. In addition, the embodiments may include fewer or additional components not shown in Figure 11 (for example, video cards, audio cards, additional network interfaces, peripherals, etc.). The processor 1101 and the network interface 1105 are coupled to the bus 1103. Although illustrated as being coupled to the bus 1103, the memory 1107 can be coupled to the processor 1101. The computing device 1100 can be integrated into the component (s) ( s) of the borehole and / or be a separate device on the surface communicatively coupled to the bottom of the BHA hole for controlling and processing signals (as described here).
    As will be understood, aspects of the invention can be realized as a system, method or program code / instruction stored in one or more machine readable media. As a result, the aspects can take the form of hardware, software (including firmware, resident software, microcode, etc.) or a combination of software and hardware aspects which can all generally be referred to here as a "circuit ", A" module "or a" system ". The functionalities presented as individual modules / units in the example of illustrations can be organized differently depending on any of the elements among the platform (operating system
    2017-IPM-101591-U1-FR and / or hardware), the application ecosystem, interfaces, programmer preferences, programming language, administrator preferences, etc.
    Any combination of one or more machine-readable media can be used. The machine readable medium may be a machine readable signal medium or a machine readable storage medium. A machine-readable storage medium can be, for example, but not limited to, a system, apparatus or device which uses any one or a combination of electronic, magnetic (s) technology (s) , optical (s), electromagnetic (s), infrared (s) or semiconductors to store the program code. More specific examples (non-exhaustive list) of machine-readable storage media would include: a laptop floppy disk, a hard drive, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (memory EPROM or flash memory), read only memory on portable compact disc (CD-ROM), optical storage device, magnetic storage device or any suitable combination of the above. In the context of this document, a machine-readable storage medium can be any tangible medium which can contain or store a program to be used by or in connection with a system, apparatus or device for executing instructions . A machine-readable storage medium is not a machine-readable signal medium.
    A machine-readable signal carrier may include a propagated data signal with a machine-readable program code incorporated therein, for example, in baseband or as part of a carrier wave. Such a propagated signal can take any of a variety of forms, including, but not limited to, electromagnetic, optical, or any suitable combination thereof. A machine-readable signal medium can be any machine-readable medium which is not a machine-readable storage medium and which can communicate, propagate or transport a program to be used by or in connection with a system, a instruction execution device or device.
    Fe program code embedded in machine readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wired, fiber optic cable, RF, etc., or any combination appropriate from the above.
    The computer program code used to perform operations for aspects of the invention can be written in any combination of one or more programming languages, including an object oriented programming language such as the programming language. Java®, C ++ or a similar program; a dynamic programming language such as Python; a scripting language such as the language of
    2017-IPM-101591-U1-FR Peri programming or the PowerShell scripting language; and conventional procedural programming languages, such as "C" programming language or similar programming languages. Program code can run entirely on a stand-alone machine, run distributed over multiple machines, and run on machine 5 while providing results and / or accepting input on another machine.
    The program code / instructions can also be stored in a machine-readable medium which can control a machine in a particular way, so that the instructions stored in the machine-readable medium produce an article of manufacture comprising instructions which set out performs the function / act specified in the flowchart 10 and / or block the block (s) of the diagram.
    权利要求:
    Claims (3)
    [1" id="c-fr-0001]
    1. Process consisting of:
    perform edge detection to locate a set of discontinuities in a game
    [2" id="c-fr-0002]
    5 of seismic data;
    classify, using a neural network, each of the sets of discontinuities as a seismic fault or a non-seismic fault; and determine the positions of the discontinuities classified as
    [3" id="c-fr-0003]
    10 seismic faults.
    类似技术:
    公开号 | 公开日 | 专利标题
    FR3070208A1|2019-02-22|DETECTION OF UNDERGROUND FAILURES BASED ON THE INTERPRETATION OF SEISMIC DATA
    Wang et al.2018|Successful leveraging of image processing and machine learning in seismic structural interpretation: A review
    Ross et al.2018|P wave arrival picking and first‐motion polarity determination with deep learning
    Pham et al.2019|Automatic channel detection using deep learning
    US9261613B2|2016-02-16|Passive monitoring method for seismic events
    EP1707993A1|2006-10-04|Method and computer program for the determination of discontinuities
    WO1999031530A1|1999-06-24|Method for automatic detection of planar heterogeneities crossing the stratification of an environment
    Adu-Gyamfi et al.2012|Multiresolution information mining for pavement crack image analysis
    BR112021011253A2|2021-08-24|Machine Learning Systems Training for Seismic Interpretation
    Di et al.2020|Seismic stratigraphy interpretation by deep convolutional neural networks: A semisupervised workflow
    Di et al.2019|Improving seismic fault detection by super-attribute-based classification
    FR3060174A1|2018-06-15|WAVELET ESTIMATION FOR A QUADRIDIMENSIONAL CHARACTERIZATION OF SUBSURFACE PROPERTIES BASED ON DYNAMIC SIMULATION
    FR3057963A1|2018-04-27|MULTIVARIABLE ANALYSIS OF SEISMIC DATA, MICROSISMIC DATA, AND PETROPHYSICAL PROPERTIES IN FRACTURE MODELING
    FR3012623A1|2015-05-01|
    Di et al.2020|A comparison of seismic saltbody interpretation via neural networks at sample and pattern levels
    Shafiq et al.2018|The role of visual saliency in the automation of seismic interpretation
    FR3057092A1|2018-04-06|GEOSTATISTICAL ANALYSIS OF MICROSISMIC DATA IN THE FRAMEWORK OF FRACTURE MODELING
    EP3701124A1|2020-09-02|Methods of analyzing cement integrity in annuli of a multiple-cased well using machine learning
    EP3570074A1|2019-11-20|Method for detecting geological objects in an image
    Janowski et al.2021|From acoustics to underwater archaeology: Deep investigation of a shallow lake using high‐resolution hydroacoustics—The case of Lake Lednica, Poland
    Pregesbauer et al.2014|An object oriented approach to automatic classification of archaeological features in magnetic prospection data
    Zhao et al.2020|A Comprehensive Horizon‐Picking Method on Subbottom Profiles by Combining Envelope, Phase Attributes, and Texture Analysis
    WO2021050122A1|2021-03-18|Geological feature detection using generative adversarial neural networks
    FR3070207A1|2019-02-22|DETECTION OF UNDERGROUND GEOLOGICAL STRUCTURES BASED ON THE INTERPRETATION OF SEISMIC DATA
    Jones2021|Tutorial: a geophysical perspective on machine learning
    同族专利:
    公开号 | 公开日
    NO20200020A1|2020-01-06|
    GB202000106D0|2020-02-19|
    AU2018317327A1|2019-11-21|
    WO2019036144A1|2019-02-21|
    US20200064507A1|2020-02-27|
    CA3063929A1|2019-02-21|
    GB2578060A|2020-04-15|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US20130294197A1|2012-05-06|2013-11-07|Schlumberger Technology Corporation|Automatic extraction and characterization of fault and fracture populations|
    GC0000235A|2000-08-09|2006-03-29|Shell Int Research|Processing an image|
    WO2009082545A1|2007-12-21|2009-07-02|Exxonmobil Upstream Research Company|Detection of features in seismic images|
    US8395967B2|2008-06-11|2013-03-12|Baker Hughes Incorporated|Vector migration of virtual source VSP data|
    US9105075B1|2013-02-06|2015-08-11|Ihs Global Inc.|Enhancing seismic features using an optical filter array|US20200292723A1|2019-03-12|2020-09-17|Bp Corporation North America Inc.|Method and Apparatus for Automatically Detecting Faults Using Deep Learning|
    CN111079631A|2019-12-12|2020-04-28|哈尔滨市科佳通用机电股份有限公司|Method and system for identifying falling fault of hook lifting rod of railway wagon|
    GB2592203A|2020-02-18|2021-08-25|Foster Findlay Ass Ltd|A System and method for improved geophysical data interpretation|
    CN111505708B|2020-04-28|2021-04-20|西安交通大学|Deep learning-based strong reflection layer stripping method|
    RU2746691C1|2020-08-26|2021-04-19|Общество с ограниченной ответственностью «Газпромнефть Научно-Технический Центр»|System for intelligent identification of tectonic faults with the help of seismic data on the basis of convolutional neural networks|
    CN112444841B|2020-12-09|2021-10-08|同济大学|Thin-layer-containing lithology earthquake prediction method based on scale-division multi-input convolution network|
    CN112526606A|2021-02-08|2021-03-19|南京云创大数据科技股份有限公司|Seismic source type prediction method and system based on heterogeneous multi-classification model|
    法律状态:
    2019-07-30| PLFP| Fee payment|Year of fee payment: 2 |
    2020-05-15| PLSC| Search report ready|Effective date: 20200515 |
    2021-05-14| RX| Complete rejection|Effective date: 20210402 |
    优先权:
    申请号 | 申请日 | 专利标题
    US201762547648P| true| 2017-08-18|2017-08-18|
    US62547648|2017-08-18|
    [返回顶部]