Using machine learning techniques to interpret earthquake faults

Abstract

A method for interpreting seismic data includes receiving seismic data representing subsurface volume and generating longitudinal and lateral line probability values ​​using a first machine learning technique. The first machine learning technique is trained to identify one or more vertical fault lines in the seismic body based on the seismic data. The method includes: generating a merged dataset by combining the longitudinal and lateral line probability values; training a second machine learning technique based on a subset of labeled horizontal planes from the merged dataset, the second machine learning technique being trained to identify horizontal fault lines from the seismic body; and generating a representation of the seismic body based on the second machine learning technique, the representation including indications of three-dimensional fault structures within the seismic body.

Description

[0001] Cross-references to related applications

[0002] This application claims priority to U.S. Provisional Patent Application Serial No. 62 / 853,681, filed on May 28, 2019, the entire contents of which are incorporated herein by reference. Background Technology

[0003] Interpreting geological structures in seismic datasets can enable the exploration, development, and production of resources such as oil. In some examples, interpreting seismic datasets may include interpreting or identifying fault structures within geological formations. Due to the geological complexity of subsurface areas and limitations associated with seismic imaging resolution, the interpretation of seismic datasets can be inefficient, inaccurate, and require a significant amount of repetitive work. For instance, manually interpreting seismic data may be impractical because it is labor-intensive and time-consuming, especially given the increasing volume of expensive seismic datasets.

[0004] In some examples, machine learning techniques can be used to automatically interpret faults from seismic data. Machine learning techniques can detect faults along vertical seismic profiles, such as inline or crossline seismic profiles. In some embodiments, labeled seismic data can be detected as user input. The labeled seismic data can be used to train machine learning techniques, such as deep neural networks. The trained machine learning techniques can be used to detect faults on a given two-dimensional (2D) seismic image corresponding to the inline or crossline seismic data. In some embodiments, the machine learning techniques can then detect predicted faults in subsequent seismic data or 2D seismic images. In some examples, machine learning techniques can aggregate predicted faults from 2D seismic images to form predicted faults in a three-dimensional (3D) seismic body. However, when using 3D seismic images to compare with 2D seismic images, the results include a large amount of false positive noise and often lack 3D consistency. For example, machine learning techniques can produce a large number of false positives when the seismic data is noisy. Summary of the Invention

[0005] Embodiments of this disclosure provide a method for interpreting seismic data. The method includes receiving seismic data representing subsurface volume and generating one or more longitudinal line probability values ​​and one or more transverse line probability values ​​using a first machine learning technique. The first machine learning technique is trained to identify one or more vertical fault lines in a seismic body based on the seismic data. The method further includes generating a merged dataset by combining one or more longitudinal line probability values ​​and one or more transverse line probability values; training a second machine learning technique based on a subset of labeled horizontal planes from the merged dataset, the second machine learning technique being trained to identify one or more horizontal fault lines from the seismic body; and generating a representation of the seismic body based on the second machine learning technique, the representation including an indication of three-dimensional fault structures within the seismic body.

[0006] Embodiments of this disclosure also provide a computational system for interpreting seismic data. The computational system includes one or more processors and a memory system including one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computational system to perform operations. These operations include receiving seismic data representing subsurface volumes and generating one or more longitudinal line probability values ​​and one or more transverse line probability values ​​using a first machine learning technique. The first machine learning technique is trained to identify one or more vertical fault lines in the seismic body based on the seismic data. These operations also include generating a merged dataset by combining one or more longitudinal line probability values ​​and one or more transverse line probability values; training a second machine learning technique based on a subset of labeled horizontal planes from the merged dataset; the second machine learning technique being trained to identify one or more horizontal fault lines from the seismic body; and generating a representation of the seismic body based on the second machine learning technique, the representation including indications of three-dimensional fault structures within the seismic body.

[0007] Embodiments of this disclosure also provide at least one non-transitory computer-readable medium for interpreting seismic data, the at least one computer-readable medium comprising a plurality of computer-executable instructions that, in response to execution by a processor, cause the processor to receive seismic data representing subsurface volume and generate one or more longitudinal line probability values ​​and one or more transverse line probability values ​​using a first machine learning technique. The first machine learning technique is trained to identify one or more vertical fault lines in the seismic body based on the seismic data. The instructions also cause the processor to generate a merged dataset by combining one or more longitudinal line probability values ​​and one or more transverse line probability values, train a second machine learning technique based on a subset of labeled horizontal planes from the merged dataset, the second machine learning technique being trained to identify one or more horizontal fault lines in the seismic body, and generate a representation of the seismic body based on the second machine learning technique, the representation including indications of three-dimensional fault structures within the seismic body.

[0008] Therefore, the computational systems and methods disclosed herein represent more efficient approaches for processing collected data, which may correspond, for example, to surface and subsurface areas. These computational systems and methods improve the effectiveness, efficiency, and accuracy of data processing. These methods and computational systems can complement or replace traditional methods for processing collected data. This overview is provided to introduce some concepts that will be further described in the detailed description below. This overview is not intended to identify key or essential features of the claimed subject matter, nor is it intended to help limit the scope of the claimed subject matter. Attached Figure Description

[0009] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the teachings and, together with the description, serve to explain the principles of the teachings. In the drawings:

[0010] Figure 1A , 1B Figures 1C, 1D, 2, 3A, and 3B show simplified schematic diagrams of an oil field and its operation according to one embodiment.

[0011] Figure 4 The illustration shows a process flowchart of a method for interpreting seismic data according to one embodiment.

[0012] Figure 5 The diagram illustrates a block diagram for interpreting 3D seismic data using machine learning techniques, according to one embodiment.

[0013] Figure 6A This is an example of a data horizontal plane from analyzed 3D seismic data according to one embodiment.

[0014] Figure 6B This is an example of interpreting fault lines in the horizontal plane of data from analyzed 3D seismic data, according to one embodiment.

[0015] Figure 7 This is an example diagram depicting the output from a first machine learning technique according to an embodiment.

[0016] Figure 8 This is an example diagram depicting the output from a second machine learning technique according to one embodiment.

[0017] Figure 9 The figure shows a schematic diagram of a computing system according to one embodiment. Detailed Implementation

[0018] Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings and figures. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without these specific details. In other instances, well-known methods, processes, components, circuits, and networks have not been described in detail to avoid unnecessarily obscuring aspects of the embodiments.

[0019] It should also be understood that although the terms first, second, etc., may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first object may be referred to as a second object, and similarly, a second object may be referred to as a first object, without departing from the scope of the invention. The first object and the second object are two separate objects, but they cannot be considered the same object.

[0020] The terminology used in the description of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in the description of the invention and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or,” as used herein, refers to and covers any possible combination of one or more of the associated listed items. It will be further understood that, when used in this specification, the terms “comprising,” “including,” and / or “comprises” designate the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. Furthermore, as used herein, depending on the context, the term “if” can be interpreted as “when” or “at…” or “in response to determination” or “in response to detection.”

[0021] Now, attention is focused on the processing procedures, methods, techniques, and workflows according to some embodiments. Some operations in the processing procedures, methods, techniques, and workflows disclosed herein can be combined and / or the order of some operations can be changed.

[0022] Figure 1A-1D A simplified schematic diagram of an oil field 100, which has a formation 102 containing a reservoir 104, is shown according to various techniques and methods described herein. Figure 1A The illustration depicts a survey operation performed by a surveying tool such as the Seismic Truck 106.1 to measure the characteristics of subsurface strata. The survey operation is a seismic survey operation used to generate acoustic vibrations. Figure 1AIn this process, a sound vibration, such as sound vibration 112 generated by source 110, is reflected away from layer 114 in stratum 116. A sensor located on the Earth's surface, such as a seismograph receiver 118, receives a set of sound vibrations. The received data 120 is provided as input data to computer 122.1 of the seismic truck 106.1, and in response to the input data, computer 122.1 generates seismic data output 124. This seismic data output can be stored, transmitted, or further processed as needed, for example, through data reduction.

[0023] Figure 1B The diagram illustrates a drilling operation performed by drilling tool 106.2, which is suspended by drilling rig 128 and advanced into subsurface formation 102 to form a wellbore 136. A mud pit 130 is used to draw drilling mud into the drilling tool via a flow line 132, circulating the mud downwards through the tool, then upwards into the wellbore 136, and back to the surface. The drilling mud is typically filtered and returned to the mud pit. A circulation system can be used to store, control, or filter the flowing drilling mud. The drilling tool is advanced into subsurface formation 102 to reach reservoir 104. Each well may target one or more reservoirs. The drilling tool is suitable for measuring downhole characteristics using logging-while-drilling (LoWWD) tools. LoWWD tools can also be used to obtain core samples 133, as shown in the figure.

[0024] The computer facilities may be located at different locations (e.g., surface unit 134) and / or remote locations around oilfield 100. Surface unit 134 may be used to communicate with drilling tools and / or field operations, as well as other surface or downhole sensors. Surface unit 134 is capable of communicating with drilling tools to send commands to and receive data from them. Surface unit 134 may also collect data generated during drilling operations and generate data output 135, which may then be stored or transmitted.

[0025] Sensors (S), such as gauges, may be located around oilfield 100 to collect data related to the various oilfield operations described above. As shown, sensors (S) are positioned at one or more locations on the drill string and / or in the drilling rig 128 to measure drilling parameters such as drill bit weight, drill bit torque, pressure, temperature, flow rate, composition, rotational speed, and / or other parameters of field operations. Sensors (S) may also be located at one or more locations within the circulation system.

[0026] Drill string 106.2 may include a bottom hole assembly (BHA) (not shown), typically located near the drill bit (e.g., within a few drill collar lengths of the drill bit). The bottom hole assembly includes the ability to measure, process, and store information, as well as communicate with surface unit 134. The bottom hole assembly also includes drill collars for performing various other measurement functions.

[0027] The bottomhole assembly may include a communication sub-assembly for communicating with the surface unit 134. The communication sub-assembly is adapted to transmit and receive signals to and from the surface using a communication channel (e.g., mud pulse telemetry, electromagnetic telemetry, or wireline drill pipe communication). The communication sub-assembly may include, for example, a transmitter that generates signals representing measured drilling parameters, such as acoustic or electromagnetic signals. Those skilled in the art will understand that various telemetry systems can be employed, such as wireline drill pipe, electromagnetic, or other known telemetry systems.

[0028] Typically, drilling is carried out according to a drilling plan established before drilling begins. The drilling plan usually specifies the equipment, pressure, trajectory, and / or other parameters that define the drilling process at the well site. Drilling operations can then be performed according to the drilling plan. However, as information is gathered, drilling operations may need to deviate from the drilling plan. Furthermore, subsurface conditions may change as drilling or other operations proceed. As new information is collected, the geodetic model may also need to be adjusted.

[0029] Data collected by the sensors (S) can be collected by ground unit 134 and / or other data collection sources for analysis or other processing. Data collected by the sensors (S) can be used alone or in combination with other data. Data can be collected in one or more databases and / or transmitted in the field or remotely. Data can be historical data, real-time data, or a combination thereof. Real-time data can be used immediately or stored for later use. Data can also be combined with historical data or other inputs for further analysis. Data can be stored in separate databases or merged into a single database.

[0030] Surface unit 134 may include transceiver 137 to allow communication between surface unit 134 and various parts or other locations of oilfield 100. Surface unit 134 may also be configured or functionally connected to one or more controllers (not shown) for driving mechanisms at oilfield 100. Surface unit 134 may then send command signals to oilfield 100 in response to received data. Surface unit 134 may receive commands via transceiver 137 or execute commands to the controller itself. A processor may be provided to analyze data (locally or remotely), make decisions, and / or activate the controller. In this way, oilfield 100 can be selectively adjusted based on collected data. This technology can be used to optimize (or improve) certain aspects of field operations, such as controlling drilling, bit weight, pumping speed, or other parameters. These adjustments may be made automatically based on computer protocols and / or manually by the operator. In some cases, well schedules may be adjusted to select optimal (or improved) operating conditions or to avoid problems.

[0031] Figure 1C It shows the suspension and entry of the drill rig 128. Figure 1B Cable operations are performed by cable tool 106.3 in wellbore 136. Cable tool 106.3 is adapted to be deployed in wellbore 136 for generating logging records, performing downhole tests, and / or collecting samples. Cable tool 106.3 can be used to provide another method and apparatus for performing seismic exploration operations. Cable tool 106.3 may, for example, have explosive, radioactive, electrical, or acoustic energy source 144 that sends and / or receives electrical signals to the surrounding subsurface formation 102 and fluids therein.

[0032] The cable tool 106.3 can be operably connected to, for example... Figure 1A The seismic truck 106.1 includes a seismic detector 118 and a computer 122.1. A cable tool 106.3 can also provide data to a surface unit 134. The surface unit 134 can collect data generated during cable operations and can produce data output 135 that can be stored or transmitted. The cable tool 106.3 can be located at different depths in the wellbore 136 to provide surveying or other information related to the subsurface formation 102.

[0033] Sensors (S), such as gauges, may be located around oilfield 100 to collect data related to the various field operations described above. As shown, sensor S is located in cable tool 106.3 to measure downhole parameters related to, for example, porosity, permeability, fluid composition, and / or other parameters of field operations.

[0034] Figure 1D The diagram illustrates a production operation performed by production tool 106.4, deployed from the production unit or tree 129 into a completed wellbore 136 to pump fluid from the downhole reservoir into the surface facility 142. Fluid flows from the reservoir 104 through perforations in the casing (not shown) into the production tool 106.4 in the wellbore 136 and reaches the surface facility 142 via a collection network 146.

[0035] Sensors (S), such as gauges, may be located around oilfield 100 to collect data related to the various field operations described above. As shown, sensors (S) may be located in production tools 106.4 or related equipment, such as treehouse 129, collection network 146, surface facility 142, and / or production facility, to measure fluid parameters, such as fluid composition, flow rate, pressure, temperature, and / or other parameters of production operations.

[0036] Production may also include injection wells to increase recovery. One or more collection facilities may be operatively connected to one or more well sites for selectively collecting downhole fluids from the well sites.

[0037] Although Figure 1B-1DTools for measuring oilfield properties are illustrated; however, it should be understood that these tools can be used for non-oilfield operations, such as gas fields, mines, aquifers, storage facilities, or other underground installations. Furthermore, while some data acquisition tools are described, it should be understood that various measurement tools are capable of sensing parameters of subsurface formations, such as seismic two-way propagation time, density, resistivity, productivity, etc., and / or can utilize their geological formations. Various sensors (S) can be located at different locations along the wellbore and / or monitoring tools to collect and / or monitor desired data. Additional data sources can also be provided from off-site locations.

[0038] Figure 1A-1D The oilfield configuration described is intended to provide a brief illustration of examples of oilfields that can be used as a framework for oilfield applications. Part or all of oilfield 100 may be on land, on water, and / or offshore. Furthermore, while a single oilfield is depicted at a single location, the oilfield application can be used in any combination of one or more oilfields, one or more processing facilities, and one or more well sites.

[0039] Figure 2 A partial cross-sectional schematic diagram of oil field 200 is shown, which has data acquisition tools 202.1, 202.2, 202.3, and 202.4 located at various locations along oil field 200 for collecting data on subsurface formation 204 according to various implementations of the technologies and processes described herein. Data acquisition tools 202.1-202.4 can be respectively connected to… Figure 1A-1D The data acquisition tools 106.1-106.4 or other tools not shown are the same. As shown in the figure, data acquisition tools 202.1-202.4 generate data plots or measurements 208.1-208.4, respectively. These data plots are drawn along oilfield 200 to illustrate the data generated by various operations.

[0040] Data plots 208.1-208.3 are examples of static data plots that can be generated separately by data acquisition tools 202.1-202.3; however, it should be understood that data plots 208.1-208.3 can also be real-time updated data plots. These measurements can be analyzed to better define the properties of the formation and / or determine the accuracy of the measurements and / or check for errors. Plots for each corresponding measurement can be aligned and scaled for comparison and verification of characteristics.

[0041] Static data figure 208.1 shows the two-way seismic response over a period of time. Static figure 208.2 shows core sample data measured from a core sample in formation 204. Core samples can be used to provide data such as graphs of density, porosity, permeability, or other physical properties along the core length. Density and viscosity tests can be performed on the fluids in the core at different pressures and temperatures. Static data figure 208.3 shows logging trajectories that typically provide formation resistivity or other measurements at different depths.

[0042] The production decline curve, or chart 208.4, is a dynamic data graph showing how fluid flow rate changes over time. Production decline curves typically provide productivity as a function of time. Fluid characteristics, such as flow rate, pressure, and composition, are measured as fluid flows through the wellbore.

[0043] Other data can also be collected, such as historical data, user input, economic information, and / or other measurement data and other parameters of interest. As described below, static and dynamic measurements can be analyzed and used to generate models of subsurface strata to determine their characteristics. Similar measurements can also be used to measure changes in strata over time.

[0044] Subsurface structure 204 has multiple geological formations 206.1-206.4. As shown in the figure, this structure has several strata or layers, including shale layer 206.1, carbonate layer 206.2, shale layer 206.3, and sand layer 206.4. Fault 207 extends through shale layer 206.1 and carbonate layer 206.2. Static data acquisition tools are suitable for measuring and detecting the characteristics of the strata.

[0045] Although specific subsurface strata with particular geological structures are depicted, it should be understood that Oilfield 200 may contain a wide variety of geological structures and / or strata, sometimes exhibiting extreme complexity. In some locations, typically below the waterline, fluids may occupy the pore space of the strata. Each measuring instrument can be used to measure the properties of the strata and / or their geological characteristics. While each acquisition tool is shown as being located at a specific location within Oilfield 200, it should be understood that one or more types of measurements may be performed at one or more locations across one or more oilfields or at other locations for comparison and / or analysis.

[0046] Then it is possible to process and / or evaluate data from various sources (e.g.) Figure 2Data is collected using data acquisition tools. Geophysicists typically use the seismic data shown in Figure 208.1 (static data from data acquisition tool 202.1) to determine the characteristics of subsurface strata and features. Geologists typically use core data shown in static figure 208.2 and / or well logging data from well logging 208.3 to determine various characteristics of subsurface strata. Reservoir engineers typically use production data from figure 208.4 to determine fluid flow reservoir characteristics. The data analyzed by geologists, geophysicists, and reservoir engineers can be analyzed using modeling techniques.

[0047] Figure 3A An oil field 300 is shown operating according to various technologies and processes described herein. As shown, the oil field has multiple well sites 302 operatively connected to a central processing facility 354. Figure 3A The oilfield configuration is not intended to limit the scope of the oilfield application system. Some or all of the oilfield may be onshore and / or offshore. Furthermore, while a single oilfield with a single processing facility and multiple well sites is depicted, any combination of one or more oilfields, one or more processing facilities, and one or more well sites may exist.

[0048] Each well site 302 has equipment for forming a wellbore 336 underground. The wellbore extends through underground formations 306, including reservoirs 304. These reservoirs 304 contain fluids, such as hydrocarbons. The well site extracts fluids from the reservoirs and transports them to a processing facility via a surface network 344. The surface network 344 has piping and control mechanisms for controlling the flow of fluids from the well site to the processing facility 354.

[0049] Now let's turn our attention to... Figure 3B This illustrates a side view of a marine-based survey 360 of a subsurface surface 362 according to one or more embodiments of the various techniques described herein. The subsurface surface 362 includes a seabed surface 364. A seismic source 366 may include a marine source such as a controlled seismic source or an air gun, which can propagate seismic waves 368 (e.g., an energy signal) into the Earth over a prolonged period of time or as a near-instantaneous energy provided by a pulsed source. The seismic waves can be propagated by the marine source as a frequency-sweeping signal. For example, a marine seismic source of the controlled seismic source type may initially emit low-frequency (e.g., 5 Hz) seismic waves and increase the seismic waves to higher frequencies (e.g., 80-90 Hz) over time.

[0050] Components of seismic wave 368 can be reflected and converted by the seabed surface 364 (i.e., a reflector), and the reflected seismic wave 370 can be received by multiple seismic receivers 372. The seismic receivers 372 can be mounted on multiple towed cables (i.e., a towed cable array 374). The seismic receivers 372 can generate electrical signals representing the received reflected seismic waves 370. Information about the subsurface 362 can be embedded in these electrical signals, and they are captured as a record of seismic data.

[0051] In one embodiment, each towline may include towline maneuvering devices, such as a bird, deflector, tail buoy, etc., which are not shown in this application. According to the techniques described herein, the towline maneuvering devices can be used to control the position of the towline.

[0052] In one implementation, the seismic wave reflection 370 can propagate upwards and reach the water / air interface at the water surface 376. A portion of the reflection 370 can then be reflected downwards again (i.e., a surface ghost wave 378) and received by multiple seismic receivers 372. The surface ghost wave 378 can be referred to as a surface multiple wave. The point on the water surface 376 where the wave is reflected downwards is generally referred to as the downward reflection point.

[0053] Electrical signals can be transmitted to vessel 380 via transmission cables, wireless communication, etc. Vessel 380 can then transmit the electrical signals to a data processing center. Alternatively, vessel 380 may include an onboard computer capable of processing electrical signals (i.e., seismic data). Those skilled in the art who benefit from this disclosure will understand that this illustration is highly idealized. For example, the survey may be of strata deep below the surface. Strata may typically include multiple reflectors, some of which may include dipping events and may generate multiple reflections (including wave conversion) for reception by seismic receiver 372. In one embodiment, seismic data can be processed to generate a seismic image of the subsurface 362.

[0054] The marine seismic acquisition system tows each cable in an array of 374 towed cables at the same depth (e.g., 5-10 m). However, ocean-based surveys 360 can tow each cable in the array of 374 towed cables at different depths, allowing seismic data to be acquired and processed in a manner that avoids destructive interference caused by surface ghost waves. For example, Figure 3B The ocean-based survey 360 shows eight towlines being towed by vessel 380 at eight different depths. The depth of each towline can be controlled and maintained using sensors mounted on each towline.

[0055] Figure 4A flowchart of a method 400 for interpreting seismic data according to an embodiment is shown. Specifically, the illustrated method 400 can be used to process 3D seismic data using various machine learning techniques. In some embodiments, method 400 can be implemented using any suitable computing device, such as... Figure 9 The computing system 901A will be described in more detail below.

[0056] At box 402, method 400 can acquire seismic data. Seismic data can be acquired by generating and recording seismic waves that propagate through subsurface regions (e.g., subsurface volumes) or are reflected from reflectors (e.g., interfaces between different types of strata). As described above, recording (or tracing) can be accomplished using recording devices such as seismic detectors, hydrophones, etc.

[0057] In one example, seismic data can be received from receiver 372, such as... Figure 3B As shown. When dragged, receiver 372 can continuously record seismic data received from reflector 370. The seismic data received by each receiver 372 can be characterized as a function of time and space. Time can represent the time when the seismic data is received or acquired, and space can indicate the geographical location or place where the seismic data is received or acquired.

[0058] In another example, seismic data can be received from receiver 118, such as Figure 1A As shown. In this example, acoustic vibrations 112 generated by source 110 are reflected away from layer 114 in stratum 116. A sensor located on the Earth's surface, such as a seismograph receiver 118, can receive a set of acoustic vibrations. The seismic data received by receiver 118 can be represented as a function of time and space. Time can represent the time when the seismic data is received or acquired, and space can indicate the geographic location or place where the seismic data is received or acquired. In one example, whether in a marine or terrestrial environment, the seismic data can be pre-recorded and retrieved from a storage device.

[0059] In some embodiments, seismic data may be aggregated into three-dimensional (3D) seismic volumes, which indicate the stratigraphic characteristics of subsurface strata within various geographic regions. For example, a 3D seismic volume may include seismic data indicating discontinuities, which may be interpreted as horizontal and / or vertical fault lines. In some embodiments, any suitable number of machine learning techniques may be used to interpret the 3D seismic volume, as described in more detail below with respect to boxes 404-408.

[0060] At box 404, method 400 may further include combining longitudinal line probability values ​​and lateral line probability values ​​generated by a first machine learning technique to form a merged dataset. The longitudinal line probability value may indicate the probability of a fault existing in a plane perpendicular to the seismic data. The vertical plane is parallel to the direction in which the seismic data was acquired. The lateral line probability value may indicate the probability that a fault is located in a plane perpendicular to the seismic data. The vertical plane is perpendicular to the direction in which the seismic data was acquired. In some embodiments, the longitudinal line probability value is represented as a longitudinal line prediction probability cube, where each data point indicates the probability that the data point is associated with a fault line. In some examples, each data point in the longitudinal line prediction probability cube may correspond to a different region within a three-dimensional region of the subsurface area. For example, the longitudinal line prediction probability cube may include probability values ​​indicating the probability that a vertical cross-section of the three-dimensional seismic data includes a fault line. The longitudinal line prediction probability cube may correspond to a fault line parallel to the direction in which the seismic data was acquired.

[0061] In some embodiments, the lateral line probability value is represented as a lateral line prediction probability cube, where each data point indicates the probability that the data point is associated with a fault line. In some embodiments, the lateral line prediction probability cube corresponds to a fault line perpendicular to the direction in which the seismic data is acquired.

[0062] In some examples, the merged dataset is computed by applying any suitable function to the longitudinal and lateral line probability values. For instance, the merged dataset can be computed using a maximization function, etc., based on a comparison of corresponding data points of the longitudinal and lateral line probability values. For example, the maximization function can be applied to data points from longitudinal and lateral line prediction probability cubes corresponding to the same location in the subsurface area. The maximization function can generate combined or merged probability cubes, which will be discussed below. Figure 7 Further detailed discussion is needed.

[0063] In some examples, at box 404, a first machine learning technique can be trained to identify one or more vertical fault lines from a 3D seismic body. For example, labeled vertical planes or slices of 3D seismic data can be provided to the first machine learning technique. The labeled vertical planes can indicate data points corresponding to fault lines and false positive data points, etc. In some embodiments, the first machine learning technique can be a neural network, a classification technique, a regression-based technique, a support vector machine, etc. In some examples, the first neural network can include any suitable number of interconnected neuron layers in each layer. For example, the first neural network can include any number of fully connected neuron layers organized with the seismic data provided as input. Organized data allows visualization of the probability of fault lines within the vertical planes of the seismic data, as combined below. Figure 7 More detailed description.

[0064] The first neural network can employ any suitable convolutional neural network technique, encoding technique, or clustering technique, such as k-means clustering, hierarchical clustering, etc. In some embodiments, the convolutional neural network may include any suitable number of local or global pooling layers, which can reduce the dimensionality of seismic data by combining the outputs of a cluster of neurons from one layer into a single neuron in subsequent layers. In some examples, the convolutional neural network may include pooling the seismic data while calculating a maximum or average value. For example, the maximum value of a cluster of neurons in a previous layer can be selected and used in subsequent layers. In some examples, the average value from a cluster of neurons can be selected and used in subsequent layers.

[0065] In some embodiments, each seismic data point may initially be placed in a first cluster, and the first cluster may be merged with additional clusters using a bottom-up approach. In another embodiment, seismic data points may be included in a first cluster, and the first cluster may be segmented using recursive techniques as neurons of the neural network move in a top-down manner. Seismic data points may be grouped or segmented based on dissimilarity values ​​between them. For example, dissimilarity values ​​may indicate the distance between two seismic data points or two sets of seismic data points. This distance may be calculated using any suitable technique, such as Euclidean distance, squared Euclidean distance, Manhattan distance, maximum distance, or Mahalanobis distance, etc. In some embodiments, the first neural network may also include a link criterion that specifies the dissimilarity of the seismic dataset as a function of pairwise distances between data points in the seismic dataset. For example, the link criterion may be calculated using maximum or fully linked clustering, minimum or single linked clustering, unweighted average linked clustering, weighted average linked clustering, centroid linked clustering, or minimum energy clustering, etc.

[0066] At box 406, method 400 may include training a second machine learning technique based on a subset of labeled horizontal planes from a merged dataset. In some examples, the second machine learning technique may be trained to identify one or more horizontal fault lines from a 3D seismic body. The labeled horizontal planes used to train the second machine learning technique may indicate data points corresponding to horizontal fault lines and data points with false positives. In some examples, false positives may not correspond to fault lines, even though the probability values ​​are high. In some embodiments, the second machine learning technique may be initiated or trained using a combined or merged probability cube or a subset of the horizontal planes of the merged dataset.

[0067] In some embodiments, the second machine learning technique may be a neural network, a classification technique, a regression-based technique, a support vector machine, or any other machine learning technique. In some examples, the second machine learning technique may be smaller than the first machine learning technique and include fewer neuron layers and / or fewer neurons per layer. In some embodiments, the neurons in the second machine learning technique may be fully connected or partially connected.

[0068] In some examples, the trained second machine learning technique can process the remaining horizontal planes of the merged dataset. Furthermore, in some examples, the second machine learning technique can be trained using horizontal planes from seismic data from multiple seismic bodies. In some embodiments, the initialized or trained second machine learning technique can analyze the merged dataset without detecting the labeled horizontal planes of the seismic data of the seismic body being analyzed. For example, the trained second machine learning technique can process the merged dataset of seismic bodies based on weights assigned to neurons in previous training iterations of the machine learning technique. In some embodiments, the output of the second machine learning technique can include a modified merged dataset or a modified probability cube reflecting the probability that each point in the vertical or horizontal plane of the seismic data corresponds to a fault line. In some examples, the output of the second machine learning technique can identify and remove false positive data values ​​previously identified as fault lines. The following is combined with... Figure 8 Describe an example of the output of the second machine learning technique.

[0069] At box 408, method 400 may include generating a three-dimensional representation of the seismic body based on a second machine learning technique. In some embodiments, the three-dimensional representation may include an indication of a three-dimensional fault structure within the 3D seismic body. As discussed herein, the three-dimensional fault structure may include any suitable number of vertical faults, horizontal faults, or combinations thereof. In some examples, the three-dimensional representation of the seismic body may remove false positives unrelated to fault lines and connect horizontal and vertical fault lines. In some examples, the three-dimensional representation of the seismic body is capable of selecting drilling plans corresponding to reservoirs in the seismic data. For example, the three-dimensional representation may indicate areas of subsurface regions to be avoided in conventional drilling due to possible fault lines. In some examples, the three-dimensional representation may indicate areas of subsurface regions to be used in unconventional drilling. For example, identified faults can remove additional oil and / or gas from the reservoir by allowing oil and / or gas to pass through the fault and be trapped within it.

[0070] In some embodiments, a three-dimensional representation of the seismic body may be stored on a local computing device or transferred to an external computing device for storage. In some examples, alarms may be generated based on the three-dimensional representation of the seismic body and transmitted to devices controlling the drill string. For example, the alarm may indicate horizontal and / or vertical changes in the drilling direction based on identified horizontal and / or vertical fault lines.

[0071] Figure 4 The process flowchart is not intended to indicate that the operations of method 400 will be performed in any particular order, or that all operations of method 400 are included in every case. Furthermore, method 400 may include any suitable number of additional operations. For example, method 400 may include creating a three-dimensional subsurface model of the subsurface rock associated with the oil and gas reservoir. Method 400 may also include selecting an oil and gas production plan based on the three-dimensional subsurface model. For example, the three-dimensional subsurface model can be used to indicate the location and size of potential reservoirs. Additionally, method 400 may include transmitting oil and gas production plans to equipment to induce resource retrieval from oil and gas reservoirs.

[0072] Figure 5 A block diagram illustrating the use of machine learning techniques to interpret 3D seismic data according to embodiments herein is shown. In some examples, seismic volume 502 may include any suitable number of three-dimensional seismic datasets. In some embodiments, the three-dimensional seismic dataset may include any number of data points corresponding to transverse and longitudinal data values.

[0073] In some embodiments, vertical fault lines 504 may be labeled within the seismic body 502. For example, any suitable subset of the vertical planes from the data of the seismic body 502 may be labeled as vertical fault lines, false positives, etc. In some embodiments, the labeled vertical fault lines 504 may be provided to a first neural network (NN#1) 506. As described above, the first neural network 506 may analyze the remaining vertical planes of the seismic body 502 and generate an output including longitudinal prediction values ​​508 and lateral prediction values ​​510. The longitudinal prediction value 508 may include any suitable representation of data points indicating the probability or likelihood of fault lines being interrupted in the longitudinal vertical plane of the seismic body 502. The longitudinal vertical plane may include a two-dimensional vertical plane of the seismic body 502, wherein the longitudinal vertical plane is parallel to the direction in which the seismic body 502 was acquired. The lateral prediction value 510 may include any suitable representation of data points indicating the probability or likelihood of fault lines being interrupted in the lateral vertical plane of the seismic body 502. The lateral line vertical plane may include a two-dimensional vertical plane of the seismic body 502, wherein the lateral line vertical plane is perpendicular to the direction in which the seismic body 502 is acquired. In some examples, the longitudinal line prediction value 508 and the lateral line prediction value 510 may be represented as a three-dimensional probability cube, wherein each data point of the 3D probability cube corresponds to the probability of the fault line at that data point.

[0074] In some embodiments, a merged prediction 512 is generated by combining the longitudinal profile prediction 508 and the transverse profile prediction 510. The merged prediction 512 can be generated using a maximum value function or any other suitable function. For example, the longitudinal profile prediction 508 and the transverse profile prediction 510 can be merged by creating a merged three-dimensional probability cube, where each data point is the maximum data point based on the corresponding data point in the longitudinal profile prediction 508 and the transverse profile prediction 510. In some examples, the longitudinal profile prediction 508 and the transverse profile prediction 510 can be merged with any other suitable function, such as a minimum function, an average function, a mean function, etc.

[0075] In some embodiments, a subset of the horizontal planes from the data 514 of the merged prediction 512 can be extracted, labeled, and used to train a second neural network (NN#2) 516. In some examples, the number of horizontal planes of the labeled data 514 used to train the second neural network 516 may be less than the number of vertical planes of the labeled data used to train the first neural network 506. For example, the second neural network 516 can be trained with labeled data 514 from as few as one or two horizontal planes of the merged prediction 512. In some embodiments, the second neural network 516 can generate an output, such as a final prediction 518. In some embodiments, the final prediction 518 indicates a horizontal fault line within the merged prediction 512. In some embodiments, the second neural network 516 can identify and remove false positive data values ​​that do not correspond to fault lines. Therefore, the final prediction 518 may include fewer false positive data points, which can reduce the noise in the final prediction 518 to more clearly indicate vertical and / or horizontal fault lines.

[0076] In some examples, the final prediction 518 can be any suitable three-dimensional representation of data points indicating the probability of horizontal and / or vertical fault lines. The final prediction 518 can be displayed within a user interface, transmitted to an external device, and / or stored within any suitable computing device. In some embodiments, the final prediction 518 can be used to modify drilling plans. The final prediction 518 is described below in conjunction with... Figure 8 Describe it.

[0077] Figure 6A This is an example of a data horizontal plane from analyzed 3D seismic data. In some embodiments, the analyzed 3D seismic data corresponds to a merged probability cube based on a longitudinal prediction probability cube and a transverse prediction probability cube. In some examples, a first machine learning technique can generate longitudinal and transverse prediction probability cubes, which can be combined based on any suitable function (e.g., a maximization function, etc.) to form the merged probability cube. Figure 6AThe horizontal plane of data point 600 represents data points along the X and Y axes. Data points 602 and 604 may represent high-probability data values ​​that could correspond to a horizontal fault line. A horizontal fault line is represented as a discontinuity pattern that may exist in the horizontal, vertical, or any combination thereof. In some examples, data points 606 and 608 may represent high-probability data values, but because data points 606 and 608 are isolated, these high-probability data values ​​may be false positives for the horizontal fault line.

[0078] Figure 6B This is an example of interpreting fault lines within the horizontal plane of analyzed 3D seismic data. In some embodiments, the analyzed 3D seismic data corresponds to a combined probability cube based on a longitudinal prediction probability cube and a transverse prediction probability cube. In some examples, a first machine learning technique can generate longitudinal and transverse prediction probability cubes, which can be combined based on any suitable function (e.g., a maximization function) to form the combined probability cube. In some embodiments, data points 610 and 612 correspond to marked horizontal fault lines. In some examples, marked horizontal fault lines are identified due to their interconnectivity with surrounding high-probability data points. For example, data points 610 and 612 belong to discontinuities present within the 3D seismic data. In some embodiments, fault lines can be predicted if a discontinuity pattern exceeding a predetermined threshold exists within a region of the 3D seismic data. In some embodiments, data points 614 and 616 are considered false positives not interconnected with adjacent high-probability data points. Therefore, in some examples, data points 614 and 616 are not identified as horizontal fault lines and can be discarded or ignored. In some embodiments, the probability values ​​of data points 614 and 616 can be adjusted to the lower probability values ​​of adjacent data points.

[0079] Figure 7 This is an example graph depicting the output from the first machine learning technique. In some embodiments, output 700 represents a combined representation of the 3D longitudinal line prediction probability cube and the 3D transverse line prediction probability cube. As described above, output 700 from the first neural network can indicate the probability of a vertical fault line in the seismic body. However, regions 702, 704, and 706, as well as other regions, may be noisy and include a higher percentage of false positives. In some examples, regions 702, 704, and 706 may also lack continuity and consistency. In some embodiments, output 700 may be difficult to interpret due to the representation of a higher concentration of false positive data values ​​and the lack of continuity.

[0080] Figure 8This is an example diagram depicting the output from a second machine learning technique. In some embodiments, output 800 represents the probability of vertical and horizontal fault lines in the seismic body. The second neural network can remove false positive data points from the horizontal plane, which can remove noise from output 800. The second neural network can also improve the continuity and consistency of data values ​​corresponding to fault lines. In some embodiments, regions 802, 804, and 806 can include clear indications of fault structures in the seismic body. In some embodiments, output 800 is capable of detecting three-dimensional fault structures in the seismic body by generating points within regions 802, 804, and 806 that indicate a high probability of horizontal or vertical fault lines at specific locations in the subsurface region.

[0081] In one or more embodiments, the described functionality can be implemented in hardware, software, firmware, or any combination thereof. For software implementation, the techniques described herein can be implemented using modules (e.g., procedures, functions, subroutines, programs, routines, subroutines, modules, software packages, classes, etc.) that perform the functions described herein. A module can be coupled to another module or hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc., can be passed, forwarded, or transmitted using any suitable means, including memory sharing, messaging, token passing, network transmission, etc. Software code can be stored in memory units and executed by a processor. Memory units can be implemented inside or outside the processor, in which case they can be communicatively coupled to the processor via various means known in the art.

[0082] In some embodiments, any of the methods disclosed herein may be performed by a computing system. Figure 9An example of such a computing system 900 according to some embodiments is shown. The computing system 900 may include a computer or computer system 901A, which may be a standalone computer system 901A or an arrangement of distributed computer systems. The computer system 901A includes one or more analysis modules 902 configured to perform various tasks according to some embodiments, such as one or more methods disclosed herein. To perform these different tasks, the analysis modules 902 execute independently or in conjunction with one or more processors 904 connected to one or more storage media 906. The processor 904 is also connected to a network interface 907 to allow the computer system 901A to communicate with one or more additional computer systems and / or computing systems, such as 901B, 901C and / or 901D, via a data network 909 (note that computer systems 901B, 901C and / or 901D may or may not share the same architecture as computer system 901A and may be located in different physical locations; for example, computer systems 901A and 901B may be located in a processing facility while communicating with one or more computer systems such as 901C and / or 901D located in one or more data centers and / or in different countries on different continents).

[0083] The processor may include a microprocessor, a microcontroller, a processor module or subsystem, a programmable integrated circuit, a programmable gate array, or another control or computing device.

[0084] Storage medium 906 can be implemented as one or more non-transitory computer-readable or non-transitory machine-readable storage media. Note that, although in Figure 9 In the example embodiments, storage medium 906 is depicted as being within computer system 901A; however, in some embodiments, storage medium 906 may be distributed within and / or across multiple internal and / or external enclosures of computing system 901A and / or additional computing systems. Storage medium 906 may include one or more different forms of memory, including semiconductor memory devices such as dynamic or static random access memory (DRAM or SRAM), erasable and programmable read-only memory (EPROM), electrically erasable and programmable read-only memory (EEPROM) and flash memory, magnetic disks (such as fixed disks, floppy disks, and removable disks), other magnetic media (including magnetic tape), optical media such as optical discs (CDs) or digital video discs (DVDs), etc. Disk or other types of optical storage, or other types of storage devices. Note that the instructions discussed above may be provided on a single computer-readable or machine-readable storage medium, or alternatively, on multiple computer-readable or machine-readable storage media distributed across a large system that may have multiple nodes. Such computer-readable or machine-readable storage media or media are considered part of an article (or a manufactured article). An article or manufactured article may refer to any single or multiple manufactured components. One or more storage media may be located in a machine that executes the machine-readable instructions, or at a remote site from which machine-readable instructions can be downloaded via a network for execution.

[0085] In some embodiments, computing system 900 includes one or more interpretation modules 908. In an example of computing system 900, computer system 901A includes interpretation module 908. In some embodiments, a single interpretation module 908 may be used to perform some or all aspects of one or more embodiments of the method. In alternative embodiments, multiple interpretation modules 908 may be used to perform some or all aspects of the techniques described herein.

[0086] It should be understood that computing system 900 is merely an example of a computing system, and computing system 900 may have more or fewer components than shown, and may be combined. Figure 9 Additional components not shown in the exemplary embodiments, and / or the computing system 900 may have Figure 9 The different configurations or arrangements of the components shown in the figure. Figure 9 The various components shown can be implemented in hardware, software, or a combination of hardware and software, including one or more signal processing and / or application-specific integrated circuits.

[0087] Furthermore, the steps in the processing methods described herein can be implemented by running one or more functional modules in an information processing device such as a general-purpose processor or a dedicated chip such as an ASIC, FPGA, PLD, or other suitable device. These modules, combinations of these modules, and / or their combinations with general-purpose hardware are all included within the scope of this invention.

[0088] Geological interpretations, models, and / or other interpretation aids can be refined iteratively; this concept applies to embodiments of the method discussed herein. This can include the use of feedback loops executed on an algorithmic basis, such as in a computing device (e.g., computing system 900). Figure 9 On the surface, and / or through manual control by the user, the user can determine whether a given set of steps, actions, templates, models, or curves has become sufficiently accurate to assess the subsurface 3D geological structures under consideration.

[0089] For purposes of explanation, the foregoing description has been described with reference to specific embodiments. However, the illustrative discussion above is not intended to be exhaustive or to limit the invention to the precise forms disclosed. In view of the foregoing teachings, many modifications and variations are possible. Furthermore, the order in which the elements of the method are shown and described may be rearranged, and / or two or more elements may appear simultaneously. The embodiments were chosen and described in order to best explain the principles of the invention and its practical application, thereby enabling others skilled in the art to best utilize the invention and its various embodiments with various modifications suitable for the intended particular use.

Claims

1. A method for interpreting seismic data, comprising: Receive seismic data representing subsurface volume; One or more longitudinal line probability values ​​and one or more transverse line probability values ​​are generated using a first machine learning technique, wherein the first machine learning technique is trained to identify one or more vertical fault lines in a seismic body based on seismic data. A merged dataset is generated by combining one or more longitudinal line probability values ​​and one or more transverse line probability values. A second machine learning technique is trained based on a subset of labeled horizontal planes from the merged dataset, which is trained to identify one or more horizontal fault lines from the seismic body. A representation of the earthquake body is generated based on a second machine learning technique, which includes an indication of the three-dimensional fault structure within the earthquake body; Generate a longitudinal survey line prediction probability cube based on one or more longitudinal survey line probability values; and A lateral line prediction probability cube is generated based on one or more lateral line prediction values, wherein the lateral line prediction probability cube and the longitudinal line prediction probability cube are generated based on the seismic body provided to the first machine learning technique.

2. The method according to claim 1, wherein, The representation is a three-dimensional digital representation visualized on a display.

3. The method according to claim 1, wherein, Generating the merged dataset involves applying a maximum value function to one or more longitudinal line probability values ​​and one or more transverse line probability values.

4. The method according to claim 1, wherein, The first machine learning technique is the first neural network, and the second machine learning technique is the second neural network.

5. The method according to claim 4, wherein, The second neural network is further trained with a second subset of labeled horizontal planes from previously analyzed seismic bodies, including merged probability cubes generated by the first neural network.

6. The method according to claim 5, wherein, The second neural network is configured to identify horizontal fault lines in subsequent seismic volumes without a subset of labeled horizontal planes from the subsequent seismic volumes.

7. The method of claim 4 further includes creating a three-dimensional underground model of the underground rock.

8. The method of claim 7 further includes selecting oil and gas production plans based on the three-dimensional underground model.

9. The method of claim 8, further comprising transmitting an oil and gas production plan to the equipment to induce resource retrieval from the oil and gas reservoir.

10. A method for interpreting seismic data, comprising: Receive seismic data representing subsurface volume; One or more longitudinal line probability values ​​and one or more transverse line probability values ​​are generated using a first machine learning technique, wherein the first machine learning technique is trained to identify one or more vertical fault lines in a seismic body based on seismic data. A merged dataset is generated by combining one or more longitudinal line probability values ​​and one or more transverse line probability values. A second machine learning technique is trained based on a subset of labeled horizontal planes from the merged dataset, which is trained to identify one or more horizontal fault lines from the seismic body. and A representation of the earthquake body is generated based on a second machine learning technique, which includes an indication of the three-dimensional fault structure within the earthquake body. The first machine learning technique is the first neural network, and the second machine learning technique is the second neural network. The second neural network is further trained with a second subset of the labeled horizontal planes from previously analyzed seismic bodies, which include a merged probability cube generated by the first neural network.

11. The method according to claim 10, wherein, The representation is a three-dimensional digital representation visualized on a display.

12. The method according to claim 10, wherein, Generating the merged dataset involves applying a maximum value function to one or more longitudinal line probability values ​​and one or more transverse line probability values.

13. The method according to claim 10, wherein, The second neural network is configured to identify horizontal fault lines in subsequent seismic volumes without a subset of labeled horizontal planes from the subsequent seismic volumes.

14. The method of claim 10, further comprising creating a three-dimensional underground model of the underground rock.

15. The method of claim 14, further comprising selecting oil and gas production plans based on the three-dimensional underground model.

16. The method of claim 15, further comprising transmitting an oil and gas production plan to the equipment to induce resource retrieval from oil and gas reservoirs.

17. A computational system for interpreting seismic data, comprising: One or more processors; and A memory system comprising one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations including: Receive seismic data representing subsurface volume; One or more longitudinal line probability values ​​and one or more transverse line probability values ​​are generated using a first machine learning technique, wherein the first machine learning technique is trained to identify one or more vertical fault lines in a seismic body based on seismic data. A merged dataset is generated by combining one or more longitudinal line probability values ​​and one or more transverse line probability values. A second machine learning technique is trained based on a subset of labeled horizontal planes from the merged dataset, which is trained to identify one or more horizontal fault lines from the seismic body. A representation of the earthquake body is generated based on a second machine learning technique, which includes an indication of the three-dimensional fault structure within the earthquake body. A longitudinal prediction probability cube is generated based on the one or more longitudinal prediction values, and a lateral prediction probability cube is generated based on the one or more lateral prediction values, wherein the lateral prediction probability cube and the longitudinal prediction probability cube are generated based on the seismic body provided to the first machine learning technique.

18. The computing system according to claim 17, wherein, Generating the merged dataset involves applying a maximum value function to one or more longitudinal line probability values ​​and one or more transverse line probability values.

19. The computing system according to claim 17, wherein, The seismic body comprises a set of data representing the characteristics of one or more underground rock strata.

20. The computing system according to claim 17, wherein, The first machine learning technique is a first neural network, and the second machine learning technique is a second neural network.

21. The computing system according to claim 20, wherein, The second neural network is further trained with a second subset of labeled horizontal planes from previously analyzed seismic bodies, which include merged probability cubes generated by the first neural network.

22. The computing system according to claim 21, wherein, The second neural network is used to identify horizontal fault lines in subsequent seismic bodies without a subset of labeled horizontal planes from the subsequent seismic bodies.

23. The computing system according to claim 20, wherein, The instructions cause one or more processors to create a three-dimensional underground model of the underground rock.

24. The computing system according to claim 23, wherein, The instructions cause the one or more processors to select an oil and gas production plan based on the three-dimensional underground model.

25. A computational system for interpreting seismic data, comprising: One or more processors; and A memory system comprising one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations including: Receive seismic data representing subsurface volume; One or more longitudinal line probability values ​​and one or more transverse line probability values ​​are generated using a first machine learning technique, wherein the first machine learning technique is trained to identify one or more vertical fault lines in a seismic body based on seismic data. A merged dataset is generated by combining one or more longitudinal line probability values ​​and one or more transverse line probability values. A second machine learning technique is trained based on a subset of labeled horizontal planes from the merged dataset, which is trained to identify one or more horizontal fault lines from the seismic body. and A representation of the earthquake body is generated based on a second machine learning technique, which includes an indication of the three-dimensional fault structure within the earthquake body. Wherein, the first machine learning technique is a first neural network, and the second machine learning technique is a second neural network. The second neural network is further trained with a second subset of labeled horizontal planes from previously analyzed seismic bodies, which include merged probability cubes generated by the first neural network.

26. The computing system according to claim 25, wherein, Generating the merged dataset involves applying a maximum value function to one or more longitudinal line probability values ​​and one or more transverse line probability values.

27. The computing system according to claim 25, wherein, The seismic body comprises a set of data representing the characteristics of one or more underground rock strata.

28. The computing system according to claim 25, wherein, The second neural network is used to identify horizontal fault lines in subsequent seismic bodies without a subset of labeled horizontal planes from the subsequent seismic bodies.

29. The computing system according to claim 25, wherein, The instructions cause one or more processors to create a three-dimensional underground model of the underground rock.

30. The computing system according to claim 29, wherein, The instructions cause the one or more processors to select an oil and gas production plan based on the three-dimensional underground model.

31. At least one non-transitory computer-readable medium for interpreting seismic data, the at least one computer-readable medium comprising a plurality of computer-executable instructions, the instructions being responsive to execution by a processor, such that the processor: Receive seismic data representing subsurface volume; One or more longitudinal line probability values ​​and one or more transverse line probability values ​​are generated using a first machine learning technique, wherein the first machine learning technique is trained to identify one or more vertical fault lines in a seismic body based on seismic data. A merged dataset is generated by combining one or more longitudinal line probability values ​​and one or more transverse line probability values. A second machine learning technique is trained based on a subset of labeled horizontal planes from the merged dataset. The second machine learning technique is trained to identify one or more horizontal fault lines from the seismic body. and A representation of the earthquake body is generated based on a second machine learning technique, which includes an indication of the three-dimensional fault structure within the earthquake body. A longitudinal prediction probability cube is generated based on the one or more longitudinal prediction values, and a lateral prediction probability cube is generated based on the one or more lateral prediction values, wherein the lateral prediction probability cube and the longitudinal prediction probability cube are generated based on the seismic body provided to the first machine learning technique.

32. At least one non-transitory computer-readable medium for interpreting seismic data, the at least one computer-readable medium comprising a plurality of computer-executable instructions, the instructions being responsive to execution by a processor, such that the processor: Receive seismic data representing subsurface volume; One or more longitudinal line probability values ​​and one or more transverse line probability values ​​are generated using a first machine learning technique, wherein the first machine learning technique is trained to identify one or more vertical fault lines in a seismic body based on seismic data. A merged dataset is generated by combining one or more longitudinal line probability values ​​and one or more transverse line probability values. A second machine learning technique is trained based on a subset of labeled horizontal planes from the merged dataset. The second machine learning technique is trained to identify one or more horizontal fault lines from the seismic body. and A representation of the earthquake body is generated based on a second machine learning technique, which includes an indication of the three-dimensional fault structure within the earthquake body. Wherein, the first machine learning technique is a first neural network, and the second machine learning technique is a second neural network. The second neural network is further trained with a second subset of labeled horizontal planes from previously analyzed seismic bodies, which include merged probability cubes generated by the first neural network.

Images