DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
(Submitted on 12/8/2025)
In regard to 103 rejections
The applicant has amended the claims 1, 9, 14 and 15 and has CANCELED the claim 27. The applicant has added a new claim 31.
- The applicant on Pages 10-14 argues on reference Tschan.
Examiner’s Response:
The applicant states that the examiner is confused with regard to Tschan teachings on Page 12 indicating that Tschan does not teach detecting a single horizon in each layer. The examiner submits that within the context of Chronostratigraphic Surfaces, the integration of multiple attributes allows for the extraction of a high-density "horizon cube"—a set of hundreds of closely spaced, parallel horizons that represent relative geologic time. Examiner disagrees and submits that perhaps the applicant is “confused” about the interpretation of the limitation. It not about determining a single horizon. It is about determining a predicted horizon for groups of subsurface image data involves tracking, interpreting, and modeling geological surfaces (e.g., formation tops, sequence boundaries) from seismic, ground-penetrating radar (GPR), or other imaging data. This process relies on identifying continuous reflection events, often using automated methods like machine learning (CNNs), structure-tensor-based trackers, or relative geologic time (RGT) volumes to handle varying data qualities and structural complexities. Further, the examiner submits that within the context of advanced seismic horizon prediction—particularly in modern AI/deep learning workflows—the output hypercube is a Horizon Cube. Without conceding the arguments of the applicant, the arguments are Moot as a result of Applicant’s arguments with respect to claims 1, 9 and 15 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. The examiner uses two new reference “Ram” and “Dorn” that teaches the amendments. The examiner submits that primary reference “Ram” has been used to teach claims 26, and 29-31 for relevancy made in the amendment of clams 1, 9 and 15. The examiner notice that the applicant has brought the claim limitations of claim 27 to the dependent claims 1, 9 and 15.
In Conclusion, independent claims 1, 9 and 15 and all dependent claims on all dependent claims 2, 5-7, 9-10, 13-14, 16, 20-21, 23-24, 26, and 28-31 are rejected under 35 USC § 103 MOVES the application as FINAL REJECTION .
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 9, 15-16, 26 and 29- 31 are rejected under 35 U.S.C. 103 as being unpatentable over
Mats Ramfjord et.al (hereinafter Ram) US 2020/0301036 A1,
In view of Geoffrey Dorn (hereinafter Dorn ) US 5894417 A
In regard to claim 1. (Currently Amended)
Ram discloses:
- select groupings of input subsurface image data from the received seismic data ,
In [0003]:
A method can include receiving seismic image data; processing the received seismic image data to generate stratigraphic information using a trained convolution neural network that includes channels subjected to convolution, activation and pooling
(BRI: stratigraphic information includes horizon information)
In [0044]:
As an example, reservoir simulation, petroleum systems modeling, etc. may be applied to characterize various types of subsurface environments, including environments such as those of FIG. 1. One or more operations may be performed in an environment based at least in part on such characterization of a subsurface environment or environments (e.g., via acquired data, simulation, modeling, etc.).
PNG
media_image1.png
306
496
media_image1.png
Greyscale
In [0047]:
In basin and petroleum systems modeling, quantities such as temperature, pressure and porosity distributions within the sediments may be modeled
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: in modern subsurface modeling, geological events are often interpreted, modeled, and organized as specific groupings, sequences, or surfaces derived from subsurface image data (such as 3D seismic volumes)
- wherein the selected groupings each have a shared geometric shape;
In [0225]:
seismic image data acquired via the seismic survey can be tiled where a tile that overlaps with a well can be labeled with at least a portion of the well log data. In such an example, the tile can be tagged for augmentation such that a training data set includes multiple instances of information in that tile, which may be deemed to be more accurate as it includes well log data.
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D.
In [0092]:
As an example, a drilling operation can include directional drilling where, for example, at least a portion of a well includes a curved axis. For example, consider a radius that defines curvature where an inclination with regard to the vertical may vary until reaching an angle between about 30 degrees and about 60 degrees or, for example, an angle to about 90 degrees or possibly greater than about 90 degrees.
In [0091] :
FIG. 4 also shows some examples of types of holes that may be drilled. For example, consider a slant hole 472, an S-shaped hole 474, a deep inclined hole 476 and a horizontal hole 478.
In [0093] :
a directional well can include several shapes where each of the shapes may aim to meet particular operational demands
(BRI: A selected grouping where each member shares a geometric shape is generally referred to as classification by geometric attribute or geometric shape sorting. In these groupings, objects are collected based on shared characteristics such as the number of sides, corners, or specific geometric forms)
- select a [[a]] point within the shared geometric shape;
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D.
In [0225]:
seismic image data acquired via the seismic survey can be tiled where a tile that overlaps with a well can be labeled with at least a portion of the well log data. In such an example, the tile can be tagged for augmentation such that a training data set includes multiple instances of information in that tile, which may be deemed to be more accurate as it includes well log data.
In [0226]:
well log data may be utilized to create fiducials (e.g., markers). Such fiducials may be utilized as constraints in training a machine model and/or in generating a model such as a MEM or an implicit function model.
In [0226]:
a layer boundary in a MEM or an implicit function model may be tied to a well log data fiducial (e.g., marker). For example, the model 1510 can include one or more layer boundaries (e.g., as may be represented by corresponding implicit function values) that correspond to depths in well log data.
(BRI: seismic image data acquired via a survey can be tiled, and tiles that overlap with well locations can be labeled with corresponding well log data for machine learning, including data augmentation. This technique enhances training datasets by providing high-fidelity labels (well logs) for specific seismic locations, allowing models to learn more robust features.
- an interpretation model coupled to the tracking and data preprocessing module,
in [0026]:
In the example of FIG. 1, the management components 110 include a seismic data component 112, an additional information component 114 (e.g., well/logging data), a processing component 116, a simulation component 120, an attribute component 130, an analysis/visualization component 142 and a workflow component 144. In operation, seismic data and other information provided per the components 112 and 114 may be input to the simulation component 120.
In [0033]:
FIG. 1 also shows an example of a framework 170 that includes a model simulation layer 180 along with a framework services layer 190, a framework core layer 195 and a modules layer 175.
(BRI: model simulation layer often includes modules that provide an interpretation model coupled to a tracking module)
- wherein the interpretation module includes a processor, the processor configured to:
In [0034] :
The OMEGA® framework provides features that can be implemented for processing of seismic data, for example, through prestack seismic interpretation and seismic inversion. A framework may be scalable such that it enables processing and imaging on a single workstation, on a massive compute cluster, etc.
- to receive the selected groupings of input subsurface image data from the tracking and data preprocessing module; and
In [0033]:
FIG. 1 also shows an example of a framework 170 that includes a model simulation layer 180 along with a framework services layer 190, a framework core layer 195 and a modules layer 175. The framework 170 may include the OCEAN® framework where the model simulation layer 180 is the PETREL® model-centric software package that hosts OCEAN® framework applications. In an example embodiment, the PETREL® software may be considered a data-driven application. The PETREL® software can include a framework for model building and visualization.
In [0034]:
A framework may be scalable such that it enables processing and imaging on a single workstation, on a massive compute cluster, etc.
(BRI: a software framework 170 that includes a modular layer frequently may provide dedicated tracking and data processing module)
- predict a horizon for each grouping of input subsurface image data received from the tracking and data preprocessing module,
In [0220]:
FIG. 15 shows an example of pre-processed seismic data 1510, which can be described as an implicit function model of a region of the Earth, and an example of a method 1550. The pre-processed seismic data 1510 is volumetric (e.g., from a seismic cube and optionally other data) and includes layers such as the layers labeled with 1521 , 1522, 1523, and 1525 where such layers can be repeating with respect to an implicit function approach. An implicit function can be calculated for a geologic region where the values of the implicit function can represent iso-surfaces such as horizons (e.g., horizons of an implicit function model). While the values may repeat for use of an implicit function, the actual layers are distinct layers that can define one or more stratigraphic units. As an example, an implicit function can be solved for values using picked points in seismic data and/or points from well log data, which can constrain the solution. As mentioned, output from a solver can be implicit function values where the values correspond to iso-surfaces that approximate the locations of materials in the geologic region (e.g., beds, etc.). Such iso-surfaces and/or implicit function values may be utilized as or for generating labels for purposes of training a CNN to generate a trained CNN.
In [0151]:
a small portion of a single seismic cube (volumetric seismic data) to generate a trained NNS and where the trained NNS is utilized for prediction on one or more other portions of the single seismic cube.
( BRI: training a Neural Network System (NNS)—typically a Convolutional Neural Network (CNN)—on a small, labeled subset of a single 3D seismic volume (cube) and applying it to predict horizons across the rest of that same cube is a recognized. The NNS is trained on this subset where the horizons are already known (labeled/interpreted).
- wherein predicting the horizon includes applying a trained algorithmic model to each grouping of input subsurface image data received from the tracking and data preprocessing module
in [0026]:
In the example of FIG. 1, the management components 110 include a seismic data component 112, an additional information component 114 (e.g., well/logging data), a processing component 116, a simulation component 120, an attribute component 130, an analysis/visualization component 142 and a workflow component 144. In operation, seismic data and other information provided per the components 112 and 114 may be input to the simulation component 120.
In [0027]:
In an example embodiment, the simulation component 120 may rely on entities 122. Entities 122 may include earth entities or geological objects such as wells, surfaces, bodies, reservoirs, etc.
In [0027]:
The entities 122 may include entities based on data acquired via sensing, observation, etc. (e.g., the seismic data 112 and other information 114).
(BRI: the seismic data component 112 is a tracking module)
(BRI: a simulation component that relies on entities such as geological objects (which include surfaces) derived from seismic data represents a structured interpretation of the subsurface image data from a tracking module)
[0220]:
FIG. 15 shows an example of pre-processed seismic data 1510, which can be described as an implicit function model of a region of the Earth, and an example of a method 1550. The pre-processed seismic data 1510 is volumetric (e.g., from a seismic cube and optionally other data) and includes layers such as the layers labeled with 1521 , 1522, 1523, and 1525 where such layers can be repeating with respect to an implicit function approach. An implicit function can be calculated for a geologic region where the values of the implicit function can represent iso-surfaces such as horizons (e.g., horizons of an implicit function model). While the values may repeat for use of an implicit function, the actual layers are distinct layers that can define one or more stratigraphic units. As an example, an implicit function can be solved for values using picked points in seismic data and/or points from well log data, which can constrain the solution. As mentioned, output from a solver can be implicit function values where the values correspond to iso-surfaces that approximate the locations of materials in the geologic region (e.g., beds, etc.). Such iso-surfaces and/or implicit function values may be utilized as or for generating labels for purposes of training a CNN to generate a trained CNN.
In [0047]:
In basin and petroleum systems modeling, quantities such as temperature, pressure and porosity distributions within the sediments may be modeled
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: in modern subsurface modeling, geological events are often interpreted, modeled, and organized as specific groupings, sequences, or surfaces derived from subsurface image data (such as 3D seismic volumes)
- to [[predict ]] determine the predicted the horizon for each respective grouping of input subsurface image data, the predicted horizon passing through the selected point within each respective grouping of input subsurface image
In [0220]:
FIG. 15 shows an example of pre-processed seismic data 1510, which can be described as an implicit function model of a region of the Earth, and an example of a method 1550. The pre-processed seismic data 1510 is volumetric (e.g., from a seismic cube and optionally other data) and includes layers such as the layers labeled with 1521 , 1522, 1523, and 1525 where such layers can be repeating with respect to an implicit function approach. An implicit function can be calculated for a geologic region where the values of the implicit function can represent iso-surfaces such as horizons (e.g., horizons of an implicit function model).
In [0220]:
an implicit function can be solved for values using picked points in seismic data and/or points from well log data, which can constrain the solution. As mentioned, output from a solver can be implicit function values where the values correspond to iso-surfaces that approximate the locations of materials in the geologic region (e.g., beds, etc.). Such iso-surfaces and/or implicit function values may be utilized as or for generating labels for purposes of training
In [0151]:
a small portion of a single seismic cube (volumetric seismic data) to generate a trained NNS and where the trained NNS is utilized for prediction on one or more other portions of the single seismic cube.
- wherein the trained algorithmic model is trained by fitting a training set to an algorithmic model,
In [Abstract]:
A method can include receiving seismic image data; processing the received seismic image data to generate stratigraphic information using a trained convolution neural network that includes channels subjected to convolution, activation and pooling that reduce spatial resolution and subjected to deconvolution and concatenation that increase spatial resolution
In [0148]:
In one or more embodiments, a computational imaging framework uses deep convolutional neural networks (CNN) to detect stratigraphic units in images of seismic sections. Such an approach can allow users to gain new insight from seismic data by quickly getting an indication of which stratigraphic units are present in an area of interest. The domain knowledge of seismic interpretation experts can be implicitly captured by a neural network when it is properly trained. In other words, once the neural network has been trained, it can inherit the domain knowledge that has been put into an interpretation by an interpretation expert or experts.
In 0114]:
an inversion process can commence with forward modeling, for example, to provide a model of layers with estimated formation depths, thicknesses, densities and velocities, which may, for example, be based at least in part on information such as well log information.
In [0115]:
Inversion can aim to generate a “best-fit” model by, for example, iterating between forward modeling and inversion while seeking to minimize differences between a synthetic trace or traces and actual seismic data.
- wherein the training set includes a plurality of training set groupings of subsurface image data,
In [0167]:
a seismic image dataset, which can include associated seismic information (e.g., coordinates, amplitude, time, etc.). As an example, a series of selections can be associated with a particular structural feature of a subsurface region. Thus, for example, interpretation of a bed (e.g., a reflector) via picking can generate a series of points that are believed to be associated with the bed.
(BRI: in modern subsurface modeling, geological events are often interpreted, modeled, and organized as specific groupings, sequences, or surfaces derived from subsurface image data (such as 3D seismic volumes. The grouping are based on a seismic image dataset that includes associated, structured information—such as coordinates (inline/crossline), amplitude, time, and metadata (e.g., station/well data) are grouping that is a standard and foundational representation of a training set for subsurface image data in machine learning)
- wherein each training set grouping of subsurface image data includes a horizon extending through a common point on [[the]] each respective training set grouping,
In [0045]:
n FIG. 2, the sedimentary basin 210, which is a geologic environment, includes horizons, faults, one or more geobodies and facies formed over some period of geologic time. These features are distributed in two or three dimensions in space, for example, with respect to a Cartesian coordinate system (e.g., x, y and z) or other coordinate system (e.g., cylindrical, spherical, etc.). As shown, the model building method 220 includes a data acquisition block 224 and a model geometry block 228,
in [0162]:
As shown in FIG. 10, the seismic image 1000 can be rendered using seismic image data that can be in the form of seismic traces, illustrated approximately in a graphic that includes waveforms of amplitude with respect to depth where traces are acquired with respect to time using seismic acquisition equipment
In [0169]:
As to depth, a scale is shown in FIG. 10 ranging from z meters to z+Δz meters. In such an example, z meters can be based on a reference location, which may be, for example, the surface of the Earth. As mentioned, time may be a proxy for depth (e.g., traveltime, etc.). In the example of FIG. 10, various windows (e.g., tiles) may be depth referenced with respect to a common reference location.
PNG
media_image2.png
682
543
media_image2.png
Greyscale
(BRI: the method leverages the fact that a horizon is a, physically continuous boundary; hence, training sets often contain sections where the horizon, if it extends through a specific location(commo reference location) , must be tracked, facilitating the identification of that horizon )
- each horizon determined based on the geological age at the common point within each respective training set grouping of subsurface image data,
In [0045]:
Furthermore, data may include depth and thickness maps stemming from facies variations (e.g., due to seismic unconformities) assumed to following geological events (“iso” times) and data may include lateral facies variations (e.g., due to lateral variation in sedimentation characteristics).
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: geological events are fundamental to determining geologic age, providing both relative sequences and absolute (numeric) ages. Events like sediment deposition, volcanic eruptions, and faulting are recorded in rock layers (strata), allowing scientists to use the law of superposition and radiometric dating to determine when events occurred in Earth's history)
- wherein each training set grouping of subsurface image data is labelled with information regarding the horizon for the respective training set grouping of subsurface image data.
In [0167] ]:
In picking, selecting can occur by positioning a cursor on a rendered seismic image at a particular location that is interpreted to be a point of interest (e.g., a seismic event, etc.).
In [0143]:
Seismic data includes information as to reflectors. A reflector can be an interface between layers of contrasting acoustic properties. Seismic waves can be reflected at such an interface. In seismic data, a reflector might represent a change in lithology, a fault or an unconformity. A reflector can be expressed as a reflection in seismic data. As an example, a seismic survey can have an associated acquisition geometry and acquisition parameters that can determine resolution. Where samples of seismic energy as acquired by one or more seismic energy sensors (e.g., receivers) provide for a depth spacing of about 10 m, a reflector may be interpreted to have a position as to depth that is accurate to within approximately 10 m
In [0167]:
As an example, a series of selections can be associated with a particular structural feature of a subsurface region. Thus, for example, interpretation of a bed (e.g., a reflector) via picking can generate a series of points that are believed to be associated with the bed.
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D. The seismic image 1000 is rendered using seismic image data as a pixel image to a display using a computerized device or system, for example, by accessing seismic image data from a data storage device and processing the seismic image data to be pixels of a desired resolution (e.g., resolution of the display, etc.), which may be adjustable based on resolution of the seismic image date. As an example, for interpretation, selection of training data, etc., the seismic image 1000 may be zoomed in, zoomed out, etc.
(BRI: in machine learning for seismic interpretation, a seismic image (or window/patch of seismic data) is typically labeled with information regarding specific, pre-interpreted horizons (reflectors) to train models)
Ram does not explicitly disclose:
- An apparatus for analyzing seismic data, the apparatus comprising:
- a tracking and data preprocessing module that includes a processor, the processor configured to:
- receive seismic data representing a geological formation
However, Dorn discloses:
- An apparatus for analyzing seismic data, the apparatus comprising:
In [Abstract]:
A computer system for interpreting seismic signals to identify reflective horizons is disclosed. The seismic signals are retrieved and conventional corrections are applied.
- a tracking and data preprocessing module that includes a processor, the processor configured to:
In [Col 2, lines 24-28]:
A conventional semi-automated approach to horizon interpretation is commonly referred to as autotracking, or volume autotracking, or volume autotracking. According to this technique, slices are again made in the seismic survey volume, and displayed by the computer system.
In [Col 3, lines 21-24]:
provide an automated system and method for performing horizon interpretation which can be efficiently applied to complex geological survey regions
In [Col 1, lines 25-32]:
After recording and storage of the detected seismic signals, conventional seismic signal processing techniques process and spatially arrange the data into a survey of the subsurface geology. Conventional techniques such as normal moveout, migration to correct for dip and diffraction effects, and noise filtering, are first applied to the seismic signals to remove known sources of error,
In [Col 4, lines 57-63]:
Referring now to FIG. 1, a computer system into which the preferred embodiment of the invention may be implemented will be described. This system includes system computer 30, which may be implemented as any conventional personal computer or workstation implemented either in standalone fashion or as part of a network arrangement.
In [Col 2, lines 31-34]:
computer system begins to extend horizon surfaces from the seed points, based on a selected algorithm and according to the seismic signals at neighboring locations, resulting in a connected surface extrapolation of the horizon
( BRI: automated systems for horizon interpretation (particularly in seismic data, medical imaging, or data analysis) typically include a preprocessing step. Preprocessing is critical in these systems to enhance data quality, remove noise, and prepare data for machine learning algorithms)
- receive seismic data representing a geological formation
In [Col 4, lines 42-52]:
seismic surveys may be of the two-dimensional (2-D) type, in which the seismic source is incrementally moved along a line that is parallel from a line of receivers, to obtain survey signals corresponding to a line of midpoints. Three-dimensional (3-D) surveys are also well known in the art, and are obtained through use of at least one array of receivers arranged in multiple parallel lines, with source energy imparted at varying locations and varying offsets from the arrays of receivers. The present invention, while applicable to 2-D surveys, is also applicable to, and is especially beneficial in connection with, 3-D seismic surveys. It is contemplated that those of ordinary skill in the art are familiar with conventional techniques of data acquisition in seismic surveys of the 2-D- and 3-D type, in both the marine and land environments.
In [Col 1, lines 43-56]:
After the seismic survey has been acquired, processed, and generated, the survey must be interpreted in order to fully understand the geology represented by the signal data. The digitization of horizon surfaces can thus be considered to convert an arrangement of time-domain seismic signals into a graphic representation of the subsurface geology, in two or three dimensions. The depth, size, and locations of interfaces between geological formations can be deduced from such a representation
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, and Dorn.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
One of ordinary skill would have motivation to combine Ram, and Dorn that provide improved resolution of the seismic trace by compressing the selected trace using system computer ((Dorn [Col 8, lines 3-49]).
In regard to claim 9. (Currently Amended)
Ram discloses:
- select groupings of input subsurface image data from the received seismic data;
In [0003]:
A method can include receiving seismic image data; processing the received seismic image data to generate stratigraphic information using a trained convolution neural network that includes channels subjected to convolution, activation and pooling
(BRI: stratigraphic information includes horizon information)
In [0044]:
As an example, reservoir simulation, petroleum systems modeling, etc. may be applied to characterize various types of subsurface environments, including environments such as those of FIG. 1. One or more operations may be performed in an environment based at least in part on such characterization of a subsurface environment or environments (e.g., via acquired data, simulation, modeling, etc.).
PNG
media_image1.png
306
496
media_image1.png
Greyscale
In [0047]:
In basin and petroleum systems modeling, quantities such as temperature, pressure and porosity distributions within the sediments may be modeled
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: in modern subsurface modeling, geological events are often interpreted, modeled, and organized as specific groupings, sequences, or surfaces derived from subsurface image data (such as 3D seismic volumes)
- wherein the selected groupings each have a shared geometric shape;
In [0225]:
seismic image data acquired via the seismic survey can be tiled where a tile that overlaps with a well can be labeled with at least a portion of the well log data. In such an example, the tile can be tagged for augmentation such that a training data set includes multiple instances of information in that tile, which may be deemed to be more accurate as it includes well log data.
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D.
In [0092]:
As an example, a drilling operation can include directional drilling where, for example, at least a portion of a well includes a curved axis. For example, consider a radius that defines curvature where an inclination with regard to the vertical may vary until reaching an angle between about 30 degrees and about 60 degrees or, for example, an angle to about 90 degrees or possibly greater than about 90 degrees.
In [0091] :
FIG. 4 also shows some examples of types of holes that may be drilled. For example, consider a slant hole 472, an S-shaped hole 474, a deep inclined hole 476 and a horizontal hole 478.
In [0093] :
a directional well can include several shapes where each of the shapes may aim to meet particular operational demands
(BRI: A selected grouping where each member shares a geometric shape is generally referred to as classification by geometric attribute or geometric shape sorting. In these groupings, objects are collected based on shared characteristics such as the number of sides, corners, or specific geometric forms)
- select a [[a]] point within the shared geometric shape;
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D.
In [0225]:
seismic image data acquired via the seismic survey can be tiled where a tile that overlaps with a well can be labeled with at least a portion of the well log data. In such an example, the tile can be tagged for augmentation such that a training data set includes multiple instances of information in that tile, which may be deemed to be more accurate as it includes well log data.
In [0226]:
well log data may be utilized to create fiducials (e.g., markers). Such fiducials may be utilized as constraints in training a machine model and/or in generating a model such as a MEM or an implicit function model.
In [0226]:
a layer boundary in a MEM or an implicit function model may be tied to a well log data fiducial (e.g., marker). For example, the model 1510 can include one or more layer boundaries (e.g., as may be represented by corresponding implicit function values) that correspond to depths in well log data.
(BRI: seismic image data acquired via a survey can be tiled, and tiles that overlap with well locations can be labeled with corresponding well log data for machine learning, including data augmentation. This technique enhances training datasets by providing high-fidelity labels (well logs) for specific seismic locations, allowing models to learn more robust features.
- and an interpretation module coupled to the tracking and data preprocessing module,
in [0026]:
In the example of FIG. 1, the management components 110 include a seismic data component 112, an additional information component 114 (e.g., well/logging data), a processing component 116, a simulation component 120, an attribute component 130, an analysis/visualization component 142 and a workflow component 144. In operation, seismic data and other information provided per the components 112 and 114 may be input to the simulation component 120.
In [0033]:
FIG. 1 also shows an example of a framework 170 that includes a model simulation layer 180 along with a framework services layer 190, a framework core layer 195 and a modules layer 175.
(BRI: model simulation layer often includes modules that provide an interpretation model coupled to a tracking module)
- the interpretation module including a processor, the processor configured to:
In [0034] :
The OMEGA® framework provides features that can be implemented for processing of seismic data, for example, through prestack seismic interpretation and seismic inversion. A framework may be scalable such that it enables processing and imaging on a single workstation, on a massive compute cluster, etc.
- receive the selected groupings of input subsurface image data from the tracking and data preprocessing module;
In [0033]:
FIG. 1 also shows an example of a framework 170 that includes a model simulation layer 180 along with a framework services layer 190, a framework core layer 195 and a modules layer 175. The framework 170 may include the OCEAN® framework where the model simulation layer 180 is the PETREL® model-centric software package that hosts OCEAN® framework applications. In an example embodiment, the PETREL® software may be considered a data-driven application. The PETREL® software can include a framework for model building and visualization.
In [0034]:
A framework may be scalable such that it enables processing and imaging on a single workstation, on a massive compute cluster, etc.
(BRI: a software framework 170 that includes a modular layer frequently may provide dedicated
- and predict a horizon for each grouping of input subsurface image data received from the tracking and data preprocessing module,
In [0220]:
FIG. 15 shows an example of pre-processed seismic data 1510, which can be described as an implicit function model of a region of the Earth, and an example of a method 1550. The pre-processed seismic data 1510 is volumetric (e.g., from a seismic cube and optionally other data) and includes layers such as the layers labeled with 1521 , 1522, 1523, and 1525 where such layers can be repeating with respect to an implicit function approach. An implicit function can be calculated for a geologic region where the values of the implicit function can represent iso-surfaces such as horizons (e.g., horizons of an implicit function model). While the values may repeat for use of an implicit function, the actual layers are distinct layers that can define one or more stratigraphic units. As an example, an implicit function can be solved for values using picked points in seismic data and/or points from well log data, which can constrain the solution. As mentioned, output from a solver can be implicit function values where the values correspond to iso-surfaces that approximate the locations of materials in the geologic region (e.g., beds, etc.). Such iso-surfaces and/or implicit function values may be utilized as or for generating labels for purposes of training a CNN to generate a trained CNN.
In [0151]:
a small portion of a single seismic cube (volumetric seismic data) to generate a trained NNS and where the trained NNS is utilized for prediction on one or more other portions of the single seismic cube.
( BRI: training a Neural Network System (NNS)—typically a Convolutional Neural Network (CNN)—on a small, labeled subset of a single 3D seismic volume (cube) and applying it to predict horizons across the rest of that same cube is a recognized. The NNS is trained on this subset where the horizons are already known (labeled/interpreted).
- wherein predicting the horizon includes applying a trained algorithmic model to each grouping of input subsurface image data received from the tracking and data preprocessing module to predict the horizon for each respective grouping of input subsurface image data, the predicted horizon passing through the selected point within each respective grouping of input subsurface image data
In [0220]:
FIG. 15 shows an example of pre-processed seismic data 1510, which can be described as an implicit function model of a region of the Earth, and an example of a method 1550. The pre-processed seismic data 1510 is volumetric (e.g., from a seismic cube and optionally other data) and includes layers such as the layers labeled with 1521 , 1522, 1523, and 1525 where such layers can be repeating with respect to an implicit function approach. An implicit function can be calculated for a geologic region where the values of the implicit function can represent iso-surfaces such as horizons (e.g., horizons of an implicit function model).
In [0220]:
an implicit function can be solved for values using picked points in seismic data and/or points from well log data, which can constrain the solution. As mentioned, output from a solver can be implicit function values where the values correspond to iso-surfaces that approximate the locations of materials in the geologic region (e.g., beds, etc.). Such iso-surfaces and/or implicit function values may be utilized as or for generating labels for purposes of training
In [0151]:
a small portion of a single seismic cube (volumetric seismic data) to generate a trained NNS and where the trained NNS is utilized for prediction on one or more other portions of the single seismic cube.
- a display module the display module configured to generate a display displaying the horizion predicted for at least one grouping of input subsurface image data,
In [0037]:
In the example of FIG. 1, the model simulation layer 180 may provide domain objects 182, act as a data source 184, provide for rendering 186 and provide for various user interfaces 188. Rendering 186 may provide a graphical environment in which applications can display their data
(BRI: a graphical environment includes a display module)
In [0045]:
In FIG. 2, the sedimentary basin 210, which is a geologic environment, includes horizons, faults, one or more geobodies and facies formed over some period of geologic time.
(BRI: the geological environment includes horizons, both in the context of soil science (pedology) and broader geological studies (stratigraphy).
In [0217]:
real seismic data can be interpreted and formatted as two images, one image being a seismic image of the seismic data and the other image being labels as to stratigraphy, which can be shown as graphical labels when rendered to a display. Such two images can be overlaid such that stratigraphy can be understood visually when the merged image is rendered to a display.
(BRI: within the geological and stratigraphic contexts, graphic labels placed on a cross-section, log, or diagram do represent horizons—specifically, they represent the, surfaces, boundaries, or specific, often synchronous, bedding planes separating different units)
In [0038]:
As an example, the domain objects 182 can include entity objects, property objects and optionally other objects. Entity objects may be used to geometrically represent wells, surfaces, bodies, reservoirs, etc., while property objects may be used to provide property values as well as data versions and display parameters. For example, an entity object may represent a well where a property object provides log information as well as version information and display information (e.g., to display the well as part of a model).
- wherein the trained algorithmic model is generated by fitting a training set to an algorithmic model, wherein the training set includes a plurality of training set groupings of subsurface image data,
In [Abstract]:
A method can include receiving seismic image data; processing the received seismic image data to generate stratigraphic information using a trained convolution neural network that includes channels subjected to convolution, activation and pooling that reduce spatial resolution and subjected to deconvolution and concatenation that increase spatial resolution
In [0148]:
In one or more embodiments, a computational imaging framework uses deep convolutional neural networks (CNN) to detect stratigraphic units in images of seismic sections. Such an approach can allow users to gain new insight from seismic data by quickly getting an indication of which stratigraphic units are present in an area of interest. The domain knowledge of seismic interpretation experts can be implicitly captured by a neural network when it is properly trained. In other words, once the neural network has been trained, it can inherit the domain knowledge that has been put into an interpretation by an interpretation expert or experts.
In 0114]:
an inversion process can commence with forward modeling, for example, to provide a model of layers with estimated formation depths, thicknesses, densities and velocities, which may, for example, be based at least in part on information such as well log information.
In [0115]:
Inversion can aim to generate a “best-fit” model by, for example, iterating between forward modeling and inversion while seeking to minimize differences between a synthetic trace or traces and actual seismic data.
- wherein each training set grouping of subsurface image data includes a horizon extending through a common point on [[the]] each respective training set grouping,
In [0045]:
n FIG. 2, the sedimentary basin 210, which is a geologic environment, includes horizons, faults, one or more geobodies and facies formed over some period of geologic time. These features are distributed in two or three dimensions in space, for example, with respect to a Cartesian coordinate system (e.g., x, y and z) or other coordinate system (e.g., cylindrical, spherical, etc.). As shown, the model building method 220 includes a data acquisition block 224 and a model geometry block 228,
in [0162]:
As shown in FIG. 10, the seismic image 1000 can be rendered using seismic image data that can be in the form of seismic traces, illustrated approximately in a graphic that includes waveforms of amplitude with respect to depth where traces are acquired with respect to time using seismic acquisition equipment
In [0169]:
As to depth, a scale is shown in FIG. 10 ranging from z meters to z+Δz meters. In such an example, z meters can be based on a reference location, which may be, for example, the surface of the Earth. As mentioned, time may be a proxy for depth (e.g., traveltime, etc.). In the example of FIG. 10, various windows (e.g., tiles) may be depth referenced with respect to a common reference location.
PNG
media_image2.png
682
543
media_image2.png
Greyscale
(BRI: the method leverages the fact that a horizon is a, physically continuous boundary; hence, training sets often contain sections where the horizon, if it extends through a specific location(commo reference location) , must be tracked, facilitating the identification of that horizon )
- each horizon determined based on the geological age at the common point within each respective training set grouping of subsurface image data,
In [0045]:
Furthermore, data may include depth and thickness maps stemming from facies variations (e.g., due to seismic unconformities) assumed to following geological events (“iso” times) and data may include lateral facies variations (e.g., due to lateral variation in sedimentation characteristics).
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: geological events are fundamental to determining geologic age, providing both relative sequences and absolute (numeric) ages. Events like sediment deposition, volcanic eruptions, and faulting are recorded in rock layers (strata), allowing scientists to use the law of superposition and radiometric dating to determine when events occurred in Earth's history)
- wherein each training set grouping of subsurface image data is labelled with information regarding the horizon for the respective training set grouping of subsurface image data.
In [0167] ]:
In picking, selecting can occur by positioning a cursor on a rendered seismic image at a particular location that is interpreted to be a point of interest (e.g., a seismic event, etc.).
In [0143]:
Seismic data includes information as to reflectors. A reflector can be an interface between layers of contrasting acoustic properties. Seismic waves can be reflected at such an interface. In seismic data, a reflector might represent a change in lithology, a fault or an unconformity. A reflector can be expressed as a reflection in seismic data. As an example, a seismic survey can have an associated acquisition geometry and acquisition parameters that can determine resolution. Where samples of seismic energy as acquired by one or more seismic energy sensors (e.g., receivers) provide for a depth spacing of about 10 m, a reflector may be interpreted to have a position as to depth that is accurate to within approximately 10 m
In [0167]:
As an example, a series of selections can be associated with a particular structural feature of a subsurface region. Thus, for example, interpretation of a bed (e.g., a reflector) via picking can generate a series of points that are believed to be associated with the bed.
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D. The seismic image 1000 is rendered using seismic image data as a pixel image to a display using a computerized device or system, for example, by accessing seismic image data from a data storage device and processing the seismic image data to be pixels of a desired resolution (e.g., resolution of the display, etc.), which may be adjustable based on resolution of the seismic image date. As an example, for interpretation, selection of training data, etc., the seismic image 1000 may be zoomed in, zoomed out, etc.
(BRI: in machine learning for seismic interpretation, a seismic image (or window/patch of seismic data) is typically labeled with information regarding specific, pre-interpreted horizons (reflectors) to train models)
Ram does not explicitly disclose:
- system for processing seismic data, the system comprising: a tracking and data preprocessing module that includes a processor, the processor configured to: receive seismic data representing a geological formation
However, Dorn discloses:
- system for processing seismic data, the system comprising: a tracking and data preprocessing module that includes a processor, the processor configured to: receive
- seismic data representing a geological formation
In [Col 4, lines 42-52]:
seismic surveys may be of the two-dimensional (2-D) type, in which the seismic source is incrementally moved along a line that is parallel from a line of receivers, to obtain survey signals corresponding to a line of midpoints. Three-dimensional (3-D) surveys are also well known in the art, and are obtained through use of at least one array of receivers arranged in multiple parallel lines, with source energy imparted at varying locations and varying offsets from the arrays of receivers. The present invention, while applicable to 2-D surveys, is also applicable to, and is especially beneficial in connection with, 3-D seismic surveys. It is contemplated that those of ordinary skill in the art are familiar with conventional techniques of data acquisition in seismic surveys of the 2-D- and 3-D type, in both the marine and land environments.
In [Col 1, lines 43-56]:
After the seismic survey has been acquired, processed, and generated, the survey must be interpreted in order to fully understand the geology represented by the signal data. The digitization of horizon surfaces can thus be considered to convert an arrangement of time-domain seismic signals into a graphic representation of the subsurface geology, in two or three dimensions. The depth, size, and locations of interfaces between geological formations can be deduced from such a representation
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, and Dorn.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
One of ordinary skill would have motivation to combine Ram, and Dorn that provide improved resolution of the seismic trace by compressing the selected trace using system computer ((Dorn [Col 8, lines 3-49]).
In regard to claim 15. (Currently Amended)
Ram discloses:
- selecting groupings of input subsurface image data from the received seismic data;
In [0003]:
A method can include receiving seismic image data; processing the received seismic image data to generate stratigraphic information using a trained convolution neural network that includes channels subjected to convolution, activation and pooling
(BRI: stratigraphic information includes horizon information)
In [0044]:
As an example, reservoir simulation, petroleum systems modeling, etc. may be applied to characterize various types of subsurface environments, including environments such as those of FIG. 1. One or more operations may be performed in an environment based at least in part on such characterization of a subsurface environment or environments (e.g., via acquired data, simulation, modeling, etc.).
PNG
media_image1.png
306
496
media_image1.png
Greyscale
In [0047]:
In basin and petroleum systems modeling, quantities such as temperature, pressure and porosity distributions within the sediments may be modeled
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: in modern subsurface modeling, geological events are often interpreted, modeled, and organized as specific groupings, sequences, or surfaces derived from subsurface image data (such as 3D seismic volumes)
- wherein the selected groupings each have a shared geometric shape;
In [0225]:
seismic image data acquired via the seismic survey can be tiled where a tile that overlaps with a well can be labeled with at least a portion of the well log data. In such an example, the tile can be tagged for augmentation such that a training data set includes multiple instances of information in that tile, which may be deemed to be more accurate as it includes well log data.
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D.
In [0092]:
As an example, a drilling operation can include directional drilling where, for example, at least a portion of a well includes a curved axis. For example, consider a radius that defines curvature where an inclination with regard to the vertical may vary until reaching an angle between about 30 degrees and about 60 degrees or, for example, an angle to about 90 degrees or possibly greater than about 90 degrees.
In [0091] :
FIG. 4 also shows some examples of types of holes that may be drilled. For example, consider a slant hole 472, an S-shaped hole 474, a deep inclined hole 476 and a horizontal hole 478.
In [0093] :
a directional well can include several shapes where each of the shapes may aim to meet particular operational demands
(BRI: A selected grouping where each member shares a geometric shape is generally referred to as classification by geometric attribute or geometric shape sorting. In these groupings, objects are collected based on shared characteristics such as the number of sides, corners, or specific geometric forms)
- selecting a [[a]] point within the shared geometric shape;
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D.
In [0225]:
seismic image data acquired via the seismic survey can be tiled where a tile that overlaps with a well can be labeled with at least a portion of the well log data. In such an example, the tile can be tagged for augmentation such that a training data set includes multiple instances of information in that tile, which may be deemed to be more accurate as it includes well log data.
In [0226]:
well log data may be utilized to create fiducials (e.g., markers). Such fiducials may be utilized as constraints in training a machine model and/or in generating a model such as a MEM or an implicit function model.
In [0226]:
a layer boundary in a MEM or an implicit function model may be tied to a well log data fiducial (e.g., marker). For example, the model 1510 can include one or more layer boundaries (e.g., as may be represented by corresponding implicit function values) that correspond to depths in well log data.
(BRI: seismic image data acquired via a survey can be tiled, and tiles that overlap with well locations can be labeled with corresponding well log data for machine learning, including data augmentation. This technique enhances training datasets by providing high-fidelity labels (well logs) for specific seismic locations, allowing models to learn more robust features)
- predicting a horizon for each grouping of input subsurface image data received from the tracking and data preprocessing module;
in [0026]:
In the example of FIG. 1, the management components 110 include a seismic data component 112, an additional information component 114 (e.g., well/logging data), a processing component 116, a simulation component 120, an attribute component 130, an analysis/visualization component 142 and a workflow component 144. In operation, seismic data and other information provided per the components 112 and 114 may be input to the simulation component 120.
In [0027]:
In an example embodiment, the simulation component 120 may rely on entities 122. Entities 122 may include earth entities or geological objects such as wells, surfaces, bodies, reservoirs, etc.
In [0027]:
The entities 122 may include entities based on data acquired via sensing, observation, etc. (e.g., the seismic data 112 and other information 114).
(BRI: the seismic data component 112 is a tracking module)
(BRI: a simulation component that relies on entities such as geological objects (which include surfaces) derived from seismic data represents a structured interpretation of the subsurface image data from a tracking module)
[0220]:
FIG. 15 shows an example of pre-processed seismic data 1510, which can be described as an implicit function model of a region of the Earth, and an example of a method 1550. The pre-processed seismic data 1510 is volumetric (e.g., from a seismic cube and optionally other data) and includes layers such as the layers labeled with 1521 , 1522, 1523, and 1525 where such layers can be repeating with respect to an implicit function approach. An implicit function can be calculated for a geologic region where the values of the implicit function can represent iso-surfaces such as horizons (e.g., horizons of an implicit function model). While the values may repeat for use of an implicit function, the actual layers are distinct layers that can define one or more stratigraphic units. As an example, an implicit function can be solved for values using picked points in seismic data and/or points from well log data, which can constrain the solution. As mentioned, output from a solver can be implicit function values where the values correspond to iso-surfaces that approximate the locations of materials in the geologic region (e.g., beds, etc.). Such iso-surfaces and/or implicit function values may be utilized as or for generating labels for purposes of training a CNN to generate a trained CNN.
In [0047]:
In basin and petroleum systems modeling, quantities such as temperature, pressure and porosity distributions within the sediments may be modeled
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: in modern subsurface modeling, geological events are often interpreted, modeled, and organized as specific groupings, sequences, or surfaces derived from subsurface image data (such as 3D seismic volumes)
- wherein predicting the horizon includes applying a trained algorithmic model to each grouping of input subsurface image data received from the tracking and data preprocessing module to [[predict]] determine the predicted horizon for each respective grouping of input subsurface image data, the predicted horizon passing through the selected point within each respective grouping of input subsurface image data
In [0220]:
FIG. 15 shows an example of pre-processed seismic data 1510, which can be described as an implicit function model of a region of the Earth, and an example of a method 1550. The pre-processed seismic data 1510 is volumetric (e.g., from a seismic cube and optionally other data) and includes layers such as the layers labeled with 1521 , 1522, 1523, and 1525 where such layers can be repeating with respect to an implicit function approach. An implicit function can be calculated for a geologic region where the values of the implicit function can represent iso-surfaces such as horizons (e.g., horizons of an implicit function model).
In [0220]:
an implicit function can be solved for values using picked points in seismic data and/or points from well log data, which can constrain the solution. As mentioned, output from a solver can be implicit function values where the values correspond to iso-surfaces that approximate the locations of materials in the geologic region (e.g., beds, etc.). Such iso-surfaces and/or implicit function values may be utilized as or for generating labels for purposes of training
In [0151]:
a small portion of a single seismic cube (volumetric seismic data) to generate a trained NNS and where the trained NNS is utilized for prediction on one or more other portions of the single seismic cube.
- wherein the trained algorithmic model is generated by fitting a training set to an algorithmic model, wherein the training set includes a plurality of training set groupings of subsurface image data,
In [Abstract]:
A method can include receiving seismic image data; processing the received seismic image data to generate stratigraphic information using a trained convolution neural network that includes channels subjected to convolution, activation and pooling that reduce spatial resolution and subjected to deconvolution and concatenation that increase spatial resolution
In [0148]:
In one or more embodiments, a computational imaging framework uses deep convolutional neural networks (CNN) to detect stratigraphic units in images of seismic sections. Such an approach can allow users to gain new insight from seismic data by quickly getting an indication of which stratigraphic units are present in an area of interest. The domain knowledge of seismic interpretation experts can be implicitly captured by a neural network when it is properly trained. In other words, once the neural network has been trained, it can inherit the domain knowledge that has been put into an interpretation by an interpretation expert or experts.
In 0114]:
an inversion process can commence with forward modeling, for example, to provide a model of layers with estimated formation depths, thicknesses, densities and velocities, which may, for example, be based at least in part on information such as well log information.
In [0115]:
Inversion can aim to generate a “best-fit” model by, for example, iterating between forward modeling and inversion while seeking to minimize differences between a synthetic trace or traces and actual seismic data.
- wherein each training set grouping of subsurface image data includes a horizon extending through a common point on [[the]] each respective training set grouping,
In [0045]:
n FIG. 2, the sedimentary basin 210, which is a geologic environment, includes horizons, faults, one or more geobodies and facies formed over some period of geologic time. These features are distributed in two or three dimensions in space, for example, with respect to a Cartesian coordinate system (e.g., x, y and z) or other coordinate system (e.g., cylindrical, spherical, etc.). As shown, the model building method 220 includes a data acquisition block 224 and a model geometry block 228,
in [0162]:
As shown in FIG. 10, the seismic image 1000 can be rendered using seismic image data that can be in the form of seismic traces, illustrated approximately in a graphic that includes waveforms of amplitude with respect to depth where traces are acquired with respect to time using seismic acquisition equipment
In [0169]:
As to depth, a scale is shown in FIG. 10 ranging from z meters to z+Δz meters. In such an example, z meters can be based on a reference location, which may be, for example, the surface of the Earth. As mentioned, time may be a proxy for depth (e.g., traveltime, etc.). In the example of FIG. 10, various windows (e.g., tiles) may be depth referenced with respect to a common reference location.
PNG
media_image2.png
682
543
media_image2.png
Greyscale
(BRI: the method leverages the fact that a horizon is a, physically continuous boundary; hence, training sets often contain sections where the horizon, if it extends through a specific location(commo reference location) , must be tracked, facilitating the identification of that horizon )
- each horizon determined based on the geological age at the common point within each respective training set grouping of subsurface image data,
In [0045]:
Furthermore, data may include depth and thickness maps stemming from facies variations (e.g., due to seismic unconformities) assumed to following geological events (“iso” times) and data may include lateral facies variations (e.g., due to lateral variation in sedimentation characteristics).
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: geological events are fundamental to determining geologic age, providing both relative sequences and absolute (numeric) ages. Events like sediment deposition, volcanic eruptions, and faulting are recorded in rock layers (strata), allowing scientists to use the law of superposition and radiometric dating to determine when events occurred in Earth's history)
- wherein each training set grouping of subsurface image data is labelled with information regarding the horizon for the respective training set grouping of subsurface image data.
In [0167] ]:
In picking, selecting can occur by positioning a cursor on a rendered seismic image at a particular location that is interpreted to be a point of interest (e.g., a seismic event, etc.).
In [0143]:
Seismic data includes information as to reflectors. A reflector can be an interface between layers of contrasting acoustic properties. Seismic waves can be reflected at such an interface. In seismic data, a reflector might represent a change in lithology, a fault or an unconformity. A reflector can be expressed as a reflection in seismic data. As an example, a seismic survey can have an associated acquisition geometry and acquisition parameters that can determine resolution. Where samples of seismic energy as acquired by one or more seismic energy sensors (e.g., receivers) provide for a depth spacing of about 10 m, a reflector may be interpreted to have a position as to depth that is accurate to within approximately 10 m
In [0167]:
As an example, a series of selections can be associated with a particular structural feature of a subsurface region. Thus, for example, interpretation of a bed (e.g., a reflector) via picking can generate a series of points that are believed to be associated with the bed.
In [0161]:
FIG. 10 shows an example of a seismic image 1000 and some examples of windows of seismic data (e.g., windows or tiles of a seismic image, etc.), labeled A, B, C and D. The seismic image 1000 is rendered using seismic image data as a pixel image to a display using a computerized device or system, for example, by accessing seismic image data from a data storage device and processing the seismic image data to be pixels of a desired resolution (e.g., resolution of the display, etc.), which may be adjustable based on resolution of the seismic image date. As an example, for interpretation, selection of training data, etc., the seismic image 1000 may be zoomed in, zoomed out, etc.
(BRI: in machine learning for seismic interpretation, a seismic image (or window/patch of seismic data) is typically labeled with information regarding specific, pre-interpreted horizons (reflectors) to train models)
Ram does not explicitly disclose:
- a method comprising: receive seismic data representing a geological formation
However, Dorn discloses:
- a method comprising: receive seismic data representing a geological formation
In [Col 4, lines 42-52]:
seismic surveys may be of the two-dimensional (2-D) type, in which the seismic source is incrementally moved along a line that is parallel from a line of receivers, to obtain survey signals corresponding to a line of midpoints. Three-dimensional (3-D) surveys are also well known in the art, and are obtained through use of at least one array of receivers arranged in multiple parallel lines, with source energy imparted at varying locations and varying offsets from the arrays of receivers. The present invention, while applicable to 2-D surveys, is also applicable to, and is especially beneficial in connection with, 3-D seismic surveys. It is contemplated that those of ordinary skill in the art are familiar with conventional techniques of data acquisition in seismic surveys of the 2-D- and 3-D type, in both the marine and land environments.
In [Col 1, lines 43-56]:
After the seismic survey has been acquired, processed, and generated, the survey must be interpreted in order to fully understand the geology represented by the signal data. The digitization of horizon surfaces can thus be considered to convert an arrangement of time-domain seismic signals into a graphic representation of the subsurface geology, in two or three dimensions. The depth, size, and locations of interfaces between geological formations can be deduced from such a representation
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, and Dorn.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
One of ordinary skill would have motivation to combine Ram, and Dorn that provide improved resolution of the seismic trace by compressing the selected trace using system computer ((Dorn [Col 8, lines 3-49]).
In regard to claim 16. (Previously Presented)
Ram discloses:
- wherein [[the]] each grouping of subsurface image data is at three-dimensional grouping of subsurface image data.
In [0167] :
In picking, selecting can occur by positioning a cursor on a rendered seismic image at a particular location that is interpreted to be a point of interest (e.g., a seismic event, etc.). Such information can include coordinates of the point, which may be a pixel or a voxel of a seismic image dataset, which can include associated seismic information (e.g., coordinates, amplitude, time, etc.). As an example, a series of selections can be associated with a particular structural feature of a subsurface region.
In [0072]:
seismic data may be processed in a technique called “depth imaging” to form an image (e.g., a depth image) of reflection amplitudes in a depth domain for a particular target structure (e.g., a geologic subsurface region of interest).
In [0151]:
training of a neural network system (NNS) is performed on a relatively small portion of a single seismic cube (volumetric seismic data) to generate a trained NNS and where the trained NNS is utilized for prediction on one or more other portions of the single seismic cube.
In [0152]:
a method can involve a regional approach where, for example, training of a NNS on multiple similar seismic cubes (e.g. cubes from an area) is performed to generate a trained NNS and where the trained NNS is utilized for prediction on another similar cube (e.g., a newly acquired seismic cube).
In regard to claim 26. (Previously Presented)
Ram discloses:
- wherein the selected point is at the center of each training set grouping of subsurface image data, each horizon extending through the selected point in the respective training set grouping,
In [0239]:
a method can include training a convolution neural network to generate a trained convolution neural network. In such an example, training can include interpreting a portion of seismic image data to generate labeled training data and processing the labeled training data to generate the trained convolution neural network. As an example, labeling can include depth labeling, for example, to provide information for a depth channel. As an example, labeling can include interpretation labeling to provide information as to an interpretation channel. As an example, an interpretation channel can include information such as whether or not a location in a seismic image is a reflector or not a reflector. A location that is in the middle of a layer of material with relatively homogenous properties
(BRI: the context of machine learning for subsurface image analysis (such as seismic interpretation), an interpretation channel that provides a location in the middle of a relatively homogeneous layer typically represents the center or a centroidal point of a training set grouping)
- wherein identifying points includes detecting points in the respective subsurface image data, the detected points of a geological age approximately equal to the geological age of the formation at the center of the respective training grouping.
in [0162]:
As shown in FIG. 10, the seismic image 1000 can be rendered using seismic image data that can be in the form of seismic traces, illustrated approximately in a graphic that includes waveforms of amplitude with respect to depth where traces are acquired with respect to time using seismic acquisition equipment
In [0169]:
As to depth, a scale is shown in FIG. 10 ranging from z meters to z+Δz meters. In such an example, z meters can be based on a reference location, which may be, for example, the surface of the Earth. As mentioned, time may be a proxy for depth (e.g., traveltime, etc.). In the example of FIG. 10, various windows (e.g., tiles) may be depth referenced with respect to a common reference location.
PNG
media_image2.png
682
543
media_image2.png
Greyscale
(BRI: the method leverages the fact that a horizon is a, physically continuous boundary; hence, training sets often contain sections where the horizon, if it extends through a specific location(commo reference location) , must be tracked, facilitating the identification of that horizon )
In [0132]:
Stratigraphy involves the study of the history, composition, relative ages and distribution of strata, and the interpretation of strata to elucidate Earth history for one or more purposes. The comparison, or correlation, of separated strata can include study of their lithology, fossil content, and relative or absolute age
In [0045]:
Furthermore, data may include depth and thickness maps stemming from facies variations (e.g., due to seismic unconformities) assumed to following geological events (“iso” times) and data may include lateral facies variations (e.g., due to lateral variation in sedimentation characteristics).
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
- wherein the horizon for each respective training set grouping is determined by identifying points in one or more approximately contiguous lines extending from the center of the respective training grouping of subsurface image data,
In [0099] :
Referring again to FIG. 4, the wellsite system 400 can include one or more sensors 464 that are operatively coupled to the control and/or data acquisition system 462. As an example, a sensor or sensors may be at surface locations. As an example, a sensor or sensors may be at downhole locations. As an example, a sensor or sensors may be at one or more remote locations that are not within a distance of the order of about one hundred meters from the wellsite system 400. As an example, a sensor or sensor may be at an offset wellsite where the wellsite system 400 and the offset wellsite are in a common field (e.g., oil and/or gas field).
In [0100]:
As an example, one or more of the sensors 464 can be provided for tracking pipe, , tracking movement of at least a portion of a drillstring, etc.
In [0133]:
Rocks that are placed within a major division of a system are said to constitute a series, which may be called lower, middle, upper, or which may be given a geographic name. In parts of the geologic section, nomenclature can be utilized to assign strata to still smaller divisions, and hence stages can be used as smaller and/or more local divisions within a series.
(BRI: assigning strata within a larger series is a stratigraphic interpretation. The algorithm identifies "roughly contiguous lines" (points that are close together and form a linear pattern) that represent, for example, pipes or strata interfaces in 3D)
In regard to claim 29 (Previously Presented)
Ram discloses:
- wherein the algorithmic model is trained using a convolutional neural network
In [0208]:
As mentioned, a method such as the method 1270 of FIG. 12 can implement a supervised
framework to identify and segment stratigraphic units in seismic data. As mentioned, a neural network such as a convolution neural network (CNN) may be trained by passing in seismic sections (e.g., as tiles) along with labeled stratigraphic units (e.g., training information). Once the neural network has been trained, it can be utilized for prediction on a desired seismic section and output segmented stratigraphic units from the input seismic section.
(BRI: to be consistent with the amendments to claim 15 for groupings in the training set , using a Convolutional Neural Network (CNN) to train seismic tiles with labeled, grouped subsurface image data provides a powerful, automated, and accurate, method for seismic interpretation)
In regard to claim 30 (Previously Presented)
Ram discloses:
- wherein the algorithmic model is trained using a convolutional neural network
In [0208]:
As mentioned, a method such as the method 1270 of FIG. 12 can implement a supervised
framework to identify and segment stratigraphic units in seismic data. As mentioned, a neural network such as a convolution neural network (CNN) may be trained by passing in seismic sections (e.g., as tiles) along with labeled stratigraphic units (e.g., training information). Once the neural network has been trained, it can be utilized for prediction on a desired seismic section and output segmented stratigraphic units from the input seismic section.
(BRI: to be consistent with the amendments to claim 9 for groupings in the training set , using a Convolutional Neural Network (CNN) to train seismic tiles with labeled, grouped subsurface image data provides a powerful, automated, and accurate, method for seismic interpretation)
In regard to claim 31. (New)
Ram discloses:
- wherein the horizon for each grouping of training subsurface image data is determined using a geological age at the selected point retrieved from a geological age model.
In [0045]:
Furthermore, data may include depth and thickness maps stemming from facies variations (e.g., due to seismic unconformities) assumed to following geological events (“iso” times) and data may include lateral facies variations (e.g., due to lateral variation in sedimentation characteristics).
In [0047]:
Modeling may also model geometry with respect to time, for example, to account for changes stemming from geological events (e.g., deposition of material, erosion of material, shifting of material, etc.).
(BRI: geological events are fundamental to determining geologic age, providing both relative sequences and absolute (numeric) ages. Events like sediment deposition, volcanic eruptions, and faulting are recorded in rock layers (strata), allowing scientists to use the law of superposition and radiometric dating to determine when events occurred in Earth's history)
Claims 2, 7, 21, and 24 are rejected under 35 U.S.C. 103 as being unpatentable over
Mats Ramfjord et.al (hereinafter Ram) US 2020/0301036 A1,
In view of Geoffrey Dorn (hereinafter Dorn ) US 5894417 A
in view of Matthias Imhof et.al (hereinafter Imhof) US 2010/0149917 A1.
In regard to claim 2. (Previously Presented)
Ram and Dorn do not explicitly disclose:
- wherein selecting a point within the shared geometric shape includes selecting for each grouping based on a peak, trough, or zero crossing identified in the grouping.
However, Imhof discloses:
- wherein selecting a point within the shared geometric shape includes selecting for each grouping based on a peak, trough, or zero crossing identified in the grouping.
In [0015]:
a method of horizon patch formation and merging by common membership in clusters of waveforms and patch properties. The method picks horizons by extracting, e.g., all peaks, but correlates them by clustering of waveforms. Picks belonging to the same cluster are used to define horizons patches which are merged into larger horizons by properties such as cluster indices, position, or seismic attributes. Specifically, the method defines with sub-sample precision the positions of seismic horizons through an extrema representation of a 3D seismic input volume. For each extrema, it derives coefficients that represent the shape of the seismic waveform in the vicinity of the extrema positions and sorts the extrema positions into groups that have similar waveform shapes by using unsupervised or supervised classification of these coefficients. It then extracts surface primitives as surface segments that are both spatially continuous along the extrema of the seismic volume and continuous in class index in the classification volume.
In [0024] :
In step (a) above, the seismic reflections may be picked by correlating reflection events between neighboring traces in the seismic data volume. The correlation may connect data peaks and troughs using cross-event semblance or correlation coefficient as a correlation measure, wherein a connection is accepted if the correlation measure is greater than a pre-selected threshold but rejected if less than the threshold. In some embodiments of the invention, only unique correlations are accepted. Alternatively, there may be identified and also accepted multiply correlated connections characterized by two or more correlations from a single peak, trough or zero crossing all exceeding the threshold
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Imhof.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Imhof teaches zer-crossing.
One of ordinary skill would have motivation to combine Ram, Dorn and Imhof to provide improved speed of execution (Imhof [0119]).
In regard to claim 7: (Previously Presented)
Ram and Dorn do not explicitly disclose:
- wherein the interpretation module processor is further configured to determine [the]]a presence of a hydrocarbon reservoir, a site for carbon storage, a site for hydrogen storage, an aquifer, or a geothermal resource using the predicted horizon variable for each grouping.
a site for carbon storage, a site for hydrogen storage, an aquifer, or a geothermal resource using horizon predicted for each grouping of input subsurface image data
However, Imhof discloses:
- determine [[the ]] a presence of a hydrocarbon reservoir,
in [0253] :
"managing hydrocarbons" includes hydrocarbon extraction, hydrocarbon production, hydrocarbon exploration, identifying potential hydrocarbon resources, identifying well locations, determining well injection and/or extraction rates, identifying reservoir connectivity, acquiring, disposing of and/or abandoning hydrocarbon resources, reviewing prior hydrocarbon management decisions, and any other hydrocarbon-related acts or activities.
- a site for carbon storage, a site for hydrogen storage, an aquifer, or a geothermal resource using the predicted horizon variable for each grouping.
in [0253]:
The term "hydrocarbon management" is also used for the injection or storage of hydrocarbons or CO.sub.2, for example the sequestration of CO.sub.2, such as reservoir evaluation, development planning, and reservoir management.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Imhof.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Imhof teaches reservoir.
One of ordinary skill would have motivation to combine Ram, Dorn and Imhof to provide improved speed of execution (Imhof [0119]).
In regard to claim 21. (Previously Presented)
Ram and Dorn do not explicitly disclose:
- determining a presence of a hydrocarbon reservoir, a site for carbon storage, a site for hydrogen storage, an aquifer, or a geothermal resource using the predicted horizon for each grouping of input subsurface image data.
However, Imhof discloses:
- determine the presence of a hydrocarbon reservoir,
in [0253] :
"managing hydrocarbons" includes hydrocarbon extraction, hydrocarbon production, hydrocarbon exploration, identifying potential hydrocarbon resources, identifying well locations, determining well injection and/or extraction rates, identifying reservoir connectivity, acquiring, disposing of and/or abandoning hydrocarbon resources, reviewing prior hydrocarbon management decisions, and any other hydrocarbon-related acts or activities.
- a site for carbon storage, a site for hydrogen storage, an aquifer, or a geothermal resource using the predicted horizon variable for each grouping.
in [0253]:
The term "hydrocarbon management" is also used for the injection or storage of hydrocarbons or CO.sub.2, for example the sequestration of CO.sub.2, such as reservoir evaluation, development planning, and reservoir management.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Imhof.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Imhof teaches reservoir.
One of ordinary skill would have motivation to combine Ram, Dorn and Imhof to provide improved speed of execution (Imhof [0119]).
In regard to claim 24. (Previously Presented)
Ram and Dorn do not explicitly disclose:
- determining a presence of a hydrocarbon reservoir, a site for carbon storage, a site for hydrogen storage, an aquifer, or a geothermal resource using the predicted horizon variable for each grouping.
However, Imhof discloses:
- determine the presence of a hydrocarbon reservoir,
in [0253] :
"managing hydrocarbons" includes hydrocarbon extraction, hydrocarbon production, hydrocarbon exploration, identifying potential hydrocarbon resources, identifying well locations, determining well injection and/or extraction rates, identifying reservoir connectivity, acquiring, disposing of and/or abandoning hydrocarbon resources, reviewing prior hydrocarbon management decisions, and any other hydrocarbon-related acts or activities.
- a site for carbon storage, a site for hydrogen storage, an aquifer, or a geothermal resource using the predicted horizon variable for each grouping.
in [0253]:
The term "hydrocarbon management" is also used for the injection or storage of hydrocarbons or CO.sub.2, for example the sequestration of CO.sub.2, such as reservoir evaluation, development planning, and reservoir management.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Imhof.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Imhof teaches reservoir.
One of ordinary skill would have motivation to combine Ram, Dorn and Imhof to provide improved speed of execution (Imhof [0119]).
Claims 6, and 14 are rejected under 35 U.S.C. 103 as being unpatentable over
Mats Ramfjord et.al (hereinafter Ram) US 2020/0301036 A1,
In view of Geoffrey Dorn (hereinafter Dorn ) US 5894417 A,
further in view of Valentin Tschannen et.al (hereinafter Tschan) Extracting horizon surfaces from 3D seismic data using deep learning, GEOPHYSICS, VOL. 85, NO. 3 (MAY-JUNE 2020); Pages . N17–N26.
further in view of Matthias Imhof et.al(hereinafter Imhof) US 2010/0149917 A1.
in regard to claim 6. ( (Previously Presented)
Ram and Dorn do not expliclity disclose:
- wherein the selected point is at the center of each training set grouping of subsurface image data, each horizon extending through the selected point in the respective training set grouping,
However, Tschan discloses:
- wherein the selected point is at the center of each training set grouping of subsurface image data, each horizon extending through the selected point in the respective training set grouping,
In [ABSTRACT, Page N17]:
We also determine how our algorithm can be used to extend horizons to the prestack domain by following reflections across offsets planes, even in the presence of residual moveout.
In [Extension to prestack seismic data, Page N22]:
Extending horizons to the prestack domain is useful to perform an improved amplitude versus angle analysis because the gradient is strongly sensitive to moveout effects
In [INTRODUCTION, Page N17]:
The tracker uses those hints to extract a 2D surface from the 3D data by finding related waveforms between traces using similarity measures.
In [DISCUSSION, Page N22]:
classification networks associate one label to the center of a fixed-size input
In [DISCUSSION, Page N23]:
patch, and obtaining a prediction over the entire volume requires us to evaluate the network on a few overlapping patches equal to the number of voxels in the data.
In [DISCUSSION, Page N23]:
The multidimensional and multiscale nature of CNNs also make them robust classifiers.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Tschan.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Tschan teaches center of training set grouping.
One of ordinary skill would have motivation to combine Ram, Dorn and Tschan that can improve the results of picking fine training the network in complex areas (Tschan [Abstract, Page N17]).
Ram, Dorn and Tschan does not explicitly disclose:
- wherein the horizon for each respective training set grouping is determined by identifying points in one or more approximately contiguous lines extending from the center of the respective training grouping of subsurface image data,
However, Imhof discloses:
- wherein the horizon for each respective training set grouping is determined by identifying points in one or more approximately contiguous lines extending from the center of the respective training grouping of subsurface image data,
In [0031]:
In another aspect, a method of transforming geologic data relating to a subsurface region between a geophysical depth domain and a geologic age domain is disclosed. Seismic data is obtained. A set of topologically consistent surfaces is obtained that correspond to the seismic data. The surfaces are enumerated in the depth domain. An age is assigned to each of the surfaces in the depth domain. The age corresponds to an estimated time of deposition of the respective surface. An age mapping volume is generated. An extent of the age domain is chosen
In [0233]:
In depth mapping volumes, however, there is no discontinuity or gap when crossing a listric fault, an unconformity, or a hiatus. Instead, the vertical derivative simply vanishes on depth mapping volumes because greatly different ages are located at essentially the same depth. A small or vanishing vertical derivative indicates a high probability for an age gap or rapid age change corresponding, for example, to a listric fault, an unconformity, (differential) compaction or a hiatus in sedimentation. These features cause tremendous stretching on seismic data transformed to the age domain, i.e., flattened data, because just a few samples that are contiguous in the depth domain are required to cover a long age span in the age domain.
- wherein identifying points includes detecting points in the respective subsurface image data,
In [0014]:
stratigraphic interpretation of a seismic image for determination of the sedimentary history of the subsurface. The method involves automatically tracking events creating at least one horizon, selecting horizons with similar seismic attributes extracted from a window at or near the horizons, and flattening the seismic volume along the selected horizons.
In [0234]:
Another application of aspects discussed herein is computing thicknesses or isopachs, which is defined as a contour which denotes points of equal thickness of a rock type, formation, groups of formations, or the like. Given two ages, two points in the subsurface, or one of each, the age mapping volume or the depth mapping volume can be used to compute a thickness map for this interval
- the detected points of a geological age approximately equal to the geological age of the formation at the center of the respective training grouping.
In [0031]:
In another aspect, a method of transforming geologic data relating to a subsurface region between a geophysical depth domain and a geologic age domain is disclosed. Seismic data is obtained. A set of topologically consistent surfaces is obtained that correspond to the seismic data. The surfaces are enumerated in the depth domain. An age is assigned to each of the surfaces in the depth domain. The age corresponds to an estimated time of deposition of the respective surface. An age mapping volume is generated. An extent of the age domain is chosen
In [0233]:
A small or vanishing vertical derivative indicates a high probability for an age gap or rapid age change corresponding, for example, to a listric fault, an unconformity, (differential) compaction or a hiatus in sedimentation. These features cause tremendous stretching on seismic data transformed to the age domain, i.e., flattened data, because just a few samples that are contiguous in the depth domain are required to cover a long age span in the age domain.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn, Tschan and Imhof.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Tschan teaches center point of the training set grouping.
Imhof teaches contiguous lines.
One of ordinary skill would have motivation to combine Ram, Dorn and Imhof to provide improved speed of execution (Imhof [0119]).
In regard to claim 14. (Currently Amended)
Ram and Dorn do not expliclity disclose:
- wherein the selected point is at the center of each training set grouping of subsurface image data, each horizon extending through the selected point in the respective training set grouping,
However, Tschan discloses:
- wherein the selected point is at the center of each training set grouping of subsurface image data, each horizon extending through the selected point in the respective training set grouping,
In [ABSTRACT, Page N17]:
We also determine how our algorithm can be used to extend horizons to the prestack domain by following reflections across offsets planes, even in the presence of residual moveout.
In [Extension to prestack seismic data, Page N22]:
Extending horizons to the prestack domain is useful to perform an improved amplitude versus angle analysis because the gradient is strongly sensitive to moveout effects
In [INTRODUCTION, Page N17]:
The tracker uses those hints to extract a 2D surface from the 3D data by finding related waveforms between traces using similarity measures.
In [DISCUSSION, Page N22]:
classification networks associate one label to the center of a fixed-size input
In [DISCUSSION, Page N23]:
patch, and obtaining a prediction over the entire volume requires us to evaluate the network on a few overlapping patches equal to the number of voxels in the data.
In [DISCUSSION, Page N23]:
The multidimensional and multiscale nature of CNNs also make them robust classifiers.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Tschan.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Tschan teaches center of training set grouping.
One of ordinary skill would have motivation to combine Ram, Dorn and Tschan that can improve the results of picking fine training the network in complex areas (Tschan [Abstract, Page N17]).
Ram, Dorn and Tschan does not explicitly disclose:
- wherein the horizon for each respective training set grouping is determined by identifying points in one or more approximately contiguous lines extending from the center of the respective training grouping of subsurface image data,
However, Imhof discloses:
- wherein the horizon for each respective training set grouping is determined by identifying points in one or more approximately contiguous lines extending from the center of the respective training grouping of subsurface image data,
In [0031]:
In another aspect, a method of transforming geologic data relating to a subsurface region between a geophysical depth domain and a geologic age domain is disclosed. Seismic data is obtained. A set of topologically consistent surfaces is obtained that correspond to the seismic data. The surfaces are enumerated in the depth domain. An age is assigned to each of the surfaces in the depth domain. The age corresponds to an estimated time of deposition of the respective surface. An age mapping volume is generated. An extent of the age domain is chosen
In [0233]:
In depth mapping volumes, however, there is no discontinuity or gap when crossing a listric fault, an unconformity, or a hiatus. Instead, the vertical derivative simply vanishes on depth mapping volumes because greatly different ages are located at essentially the same depth. A small or vanishing vertical derivative indicates a high probability for an age gap or rapid age change corresponding, for example, to a listric fault, an unconformity, (differential) compaction or a hiatus in sedimentation. These features cause tremendous stretching on seismic data transformed to the age domain, i.e., flattened data, because just a few samples that are contiguous in the depth domain are required to cover a long age span in the age domain.
- wherein identifying points includes detecting points in the respective subsurface image data,
In [0014]:
stratigraphic interpretation of a seismic image for determination of the sedimentary history of the subsurface. The method involves automatically tracking events creating at least one horizon, selecting horizons with similar seismic attributes extracted from a window at or near the horizons, and flattening the seismic volume along the selected horizons.
In [0234]:
Another application of aspects discussed herein is computing thicknesses or isopachs, which is defined as a contour which denotes points of equal thickness of a rock type, formation, groups of formations, or the like. Given two ages, two points in the subsurface, or one of each, the age mapping volume or the depth mapping volume can be used to compute a thickness map for this interval
- the detected points of a geological age approximately equal to the geological age of the formation at the center of the respective training grouping.
In [0031]:
In another aspect, a method of transforming geologic data relating to a subsurface region between a geophysical depth domain and a geologic age domain is disclosed. Seismic data is obtained. A set of topologically consistent surfaces is obtained that correspond to the seismic data. The surfaces are enumerated in the depth domain. An age is assigned to each of the surfaces in the depth domain. The age corresponds to an estimated time of deposition of the respective surface. An age mapping volume is generated. An extent of the age domain is chosen
In [0233]:
A small or vanishing vertical derivative indicates a high probability for an age gap or rapid age change corresponding, for example, to a listric fault, an unconformity, (differential) compaction or a hiatus in sedimentation. These features cause tremendous stretching on seismic data transformed to the age domain, i.e., flattened data, because just a few samples that are contiguous in the depth domain are required to cover a long age span in the age domain.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn, Tschan and Imhof.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Tschan teaches center point of the training set grouping.
Imhof teaches contiguous lines.
One of ordinary skill would have motivation to combine Ram, Dorn and Imhof to provide improved speed of execution (Imhof [0119]).
Claims 5, and 13 are rejected under 35 U.S.C. 103 as being unpatentable over
Mats Ramfjord et.al (hereinafter Ram) US 2020/0301036 A1,
In view of Geoffrey Dorn (hereinafter Dorn ) US 5894417 A
Further in view of Sergey Skripkin et.al (hereinafter Skripkin) US 2022/0114302 A1.
In regard to claim 5. (Previously Presented)
Ram and Dorn do not explicitly disclose:
- wherein fitting a training set to an algorithmic model includes applying classification and linear regression analysis to the training set
However, Skripkin discloses:
- wherein fitting a training set to an algorithmic model includes applying classification and linear regression analysis to the training set
In [0202]:
As an example, in a method such as the method 1500, types of machine learning models may be the same from iteration to iteration as batches are analyzed or a dynamic approach may adjust types of machine learning models utilized. For example, where training and/or testing determine that one or more types of machine learning models are unlikely to provide acceptable prediction accuracy, such types may be dropped for a subsequent iteration. As an example, where particular complexity is observed in data, a type of machine learning model may be adjusted as to its complexity,
In [0235]:
The plot 1630 shows a one-dimensional example where the actual underlying system is in actuality more complex. As such, the five samples shown in the plot 1630 do not provide for a reasonable fit (e.g., training) of a predictive model. In such an example, additional samples may assist with fitting such that a fit predictive model can adequately represent the underlying system.
In [0147]:
As to types of machine learning models, consider one or more of a support vector machine (SVM) model, a k-nearest neighbors (KNN) model, an ensemble classifier model, a neural network (NN) model, etc. As an example, a machine learning model can be a deep learning model (e.g., deep Boltzmann machine, deep belief network, convolutional neural network, stacked auto-encoder, etc.), an ensemble model (e.g., random forest, gradient boosting machine, bootstrapped aggregation, AdaBoost, stacked generalization, gradient boosted regression tree, etc.), a neural network model (e.g., radial basis function network, perceptron, back-propagation, Hopfield network, etc.), a regularization model (e.g., ridge regression, least absolute shrinkage and selection operator, elastic net, least angle regression), a rule system model (e.g., cubist, one rule, zero rule, repeated incremental pruning to produce error reduction), a regression model (e.g., linear regression, ordinary least squares regression)
In [0148]:
the DLT provides convolutional neural networks (ConvNets, CNNs) and long short-term memory (LSTM) networks to perform classification and regression on image
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn, and Skripkin.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Skripkin teaches linear regression.
One of ordinary skill would have motivation to combine Ram, Tschan , and Skripkin improve model generation (Skripkin [0127]).
In regard to claim 13. (Previously Presented)
Ram and Tschan do not explicitly disclose:
- wherein fitting a training set to an algorithmic model includes applying classification and linear regression analysis to the training set
However, Skripkin discloses:
- wherein fitting a training set to an algorithmic model includes applying classification and linear regression analysis to the training set
In [0202]:
As an example, in a method such as the method 1500, types of machine learning models may be the same from iteration to iteration as batches are analyzed or a dynamic approach may adjust types of machine learning models utilized. For example, where training and/or testing determine that one or more types of machine learning models are unlikely to provide acceptable prediction accuracy, such types may be dropped for a subsequent iteration. As an example, where particular complexity is observed in data, a type of machine learning model may be adjusted as to its complexity,
In [0235]:
The plot 1630 shows a one-dimensional example where the actual underlying system is in actuality more complex. As such, the five samples shown in the plot 1630 do not provide for a reasonable fit (e.g., training) of a predictive model. In such an example, additional samples may assist with fitting such that a fit predictive model can adequately represent the underlying system.
In [0147]:
As to types of machine learning models, consider one or more of a support vector machine (SVM) model, a k-nearest neighbors (KNN) model, an ensemble classifier model, a neural network (NN) model, etc. As an example, a machine learning model can be a deep learning model (e.g., deep Boltzmann machine, deep belief network, convolutional neural network, stacked auto-encoder, etc.), an ensemble model (e.g., random forest, gradient boosting machine, bootstrapped aggregation, AdaBoost, stacked generalization, gradient boosted regression tree, etc.), a neural network model (e.g., radial basis function network, perceptron, back-propagation, Hopfield network, etc.), a regularization model (e.g., ridge regression, least absolute shrinkage and selection operator, elastic net, least angle regression), a rule system model (e.g., cubist, one rule, zero rule, repeated incremental pruning to produce error reduction), a regression model (e.g., linear regression, ordinary least squares regression)
In [0148]:
the DLT provides convolutional neural networks (ConvNets, CNNs) and long short-term memory (LSTM) networks to perform classification and regression on image
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn, and Skripkin.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Skripkin teaches linear regression.
One of ordinary skill would have motivation to combine Ram, Tschan , and Skripkin improve model generation (Skripkin [0127]).
Claims 10 and 28 are rejected under 35 U.S.C. 103 as being unpatentable over
Mats Ramfjord et.al (hereinafter Ram) US 2020/0301036 A1,
In view of Geoffrey Dorn (hereinafter Dorn ) US 5894417 A
further in view of Mustafa Al-Ali (hereinafter Ali) US 2020/00271809 A1.
20110002194
In regard to claim 10. (Previously Presented)
Ram and Dorn do not explicitly disclose:
- wherein the selected point is at the center of the geometric shape
However, Ali discloses:
In [0089]:
In the execution process, and in preparation for seismic data acquisition or collection, the illustrated equipment of FIG. 20 are taken to the field or the drilling rig site to layout the designed acquisition geometries or patterns on the surface around the drilling rig 802. These sensors (for example, geophones 818) record seismic signals transmitted and reflected from the drilling bit 816 and drilling string 814 travelling through subsurface formations.
In [0105]:
some aspects, the same velocity model as is used in depth for the drill bit imaging, and the target point is assumed as the projection of the well head to the target horizon or target depth.
In [0086]:
FIG. 17 illustrates a symmetric acquisition geometry 1700 centered on the wellhead and targeting deeper depths. The geometry 1700 targets deeper depths of over 500 m and up to 3000 m.
In [0080]:
The main parameters used to form an acquisition geometry are: the special sampling, which is the spacing between the centers of sensor arrays; the maximum areal coverage, which is the size of the geometry pattern; and the target depth.
(BRI: Wellhead placement is determined by seismic data, which provides subsurface imaging to identify geological features and ideal drilling locations).
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn, and Ali.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Skripkin teaches linear regression.
Ali teaches image tiles.
One of ordinary skill would have motivation to combine Ram, Dorn and Ali that can provide required accuracy and spatial resolution (Ali [0105]).
In regard to claim 28.(Previously Presented)
Ram and Dorn do not explicitly disclose:
- wherein selecting groupings of input subsurface image data from the seismic data includes selecting groupings that form image tiles, with each image tile having a predetermined size and a center, wherein the selected point is at the center of the image tile.
However, Ali discloses:
- wherein selecting groupings of input subsurface image data from the seismic data includes selecting groupings that form image tiles,
In [0089]:
In the execution process, and in preparation for seismic data acquisition or collection, the illustrated equipment of FIG. 20 are taken to the field or the drilling rig site to layout the designed acquisition geometries or patterns on the surface around the drilling rig 802. These sensors (for example, geophones 818) record seismic signals transmitted and reflected from the drilling bit 816 and drilling string 814 travelling through subsurface formations.
In [0105]:
some aspects, the same velocity model as is used in depth for the drill bit imaging, and the target point is assumed as the projection of the well head to the target horizon or target depth.
- with each image tile having a predetermined size and a center, wherein the selected point is at the center of image tile.
In [0086]:
FIG. 17 illustrates a symmetric acquisition geometry 1700 centered on the wellhead and targeting deeper depths. The geometry 1700 targets deeper depths of over 500 m and up to 3000 m.
In [0080]:
The main parameters used to form an acquisition geometry are: the special sampling, which is the spacing between the centers of sensor arrays; the maximum areal coverage, which is the size of the geometry pattern; and the target depth.
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn, and Ali.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Skripkin teaches linear regression.
Ali teaches image tiles.
One of ordinary skill would have motivation to combine Ram, Dorn and Ali that can provide required accuracy and spatial resolution (Ali [0105]).
Claims 20 and 23 are rejected under 35 U.S.C. 103 as being unpatentable over
Mats Ramfjord et.al (hereinafter Ram) US 2020/0301036 A1,
In view of Geoffrey Dorn (hereinafter Dorn ) US 5894417 A
further in view of Valentin Tschannen et.al (hereinafter Tschan) Extracting horizon surfaces from 3D seismic data using deep learning, GEOPHYSICS, VOL. 85, NO. 3 (MAY-JUNE 2020); Pages . N17–N26.
In regard to claim 20. (Previously Presented)
Ram and Dorn do not explicitly disclose:
- predicting the horizon variable using the isochron variable for each grouping and the trained algorithmic model, wherein the trained algorithmic model generates a range of probability values.
However, Tschan discloses:
In [METHODS, Page N19]:
In this work, we aim to segment a set of K horizons in a 3D seismic stack. Let h (il; xl, t) be a seismic sample expressed in the 3D coordinate system (il; xl, t), where il and xl are the spatial coordinates in the inline and crossline directions and t is the vertical temporal coordinate. We express the labels as a 4D hypercube p(il; xl, t, K+1) representing the probable locations of the K target reflectors. For every voxel, p holds the discrete probability density function
(
p
k
≥
0
)
k
=
1
,
…
K
=
1
such that
∑
k
=
1
K
+
1
p
k
=1. The terms
p
1
to
p
k
are the probabilities of each horizon , whereas
p
K
+
1
is the probability of not being any of the desired reflectors.
(BRI: the probabilities
p
1
to
p
k
are the range of probabilities)
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Tschan.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Tschan teaches center of training set grouping.
One of ordinary skill would have motivation to combine Ram, Dorn and Tschan that can improve the results of picking fine training the network in complex areas (Tschan [Abstract, Page N17]).
In regard to claim 23. (Previously Presented)
Ram and Dorn do not explicitly disclose:
predicting the horizon variable using the isochron variable for each grouping and the trained algorithmic model, wherein the trained algorithmic model generates a range of probability values.
However, Tschan discloses:
- predicting the horizon variable using the isochron variable for each grouping and the trained algorithmic model, wherein the trained algorithmic model generates a range of probability values.
In [METHODS, Page N19]:
In this work, we aim to segment a set of K horizons in a 3D seismic stack. Let h (il; xl, t) be a seismic sample expressed in the 3D coordinate system (il; xl, t), where il and xl are the spatial coordinates in the inline and crossline directions and t is the vertical temporal coordinate. We express the labels as a 4D hypercube p(il; xl, t, K+1) representing the probable locations of the K target reflectors. For every voxel, p holds the discrete probability density function
(
p
k
≥
0
)
k
=
1
,
…
K
=
1
such that
∑
k
=
1
K
+
1
p
k
=1. The terms
p
1
to
p
k
are the probabilities of each horizon , whereas
p
K
+
1
is the probability of not being any of the desired reflectors.
(BRI: the probabilities
p
1
to
p
k
are the range of probabilities)
The examiner interprets the theme of the invention is to predict seismic horizon using neural network training with determination of horizon common point on the training data set.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ram, Dorn and Tschan.
Ram teaches grouping of subsurface, prediction of horizon and horizon points and its references to geological age and label
Dorn teaches apparatus and modules to perform the methods for embodiments of the core invention and to receive the seismic data.
Tschan teaches center of training set grouping.
One of ordinary skill would have motivation to combine Ram, Dorn and Tschan that can improve the results of picking fine training the network in complex areas (Tschan [Abstract, Page N17]).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TIRUMALE KRISHNASWAMY RAMESH whose telephone number is (571)272-4605. The examiner can normally be reached by phone.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at
http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Li B Zhen can be reached on phone (571-272-3768). The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users.
To file and manage patent submissions in Patent Center, visit:
https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format.
For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/TIRUMALE K RAMESH/Examiner, Art Unit 2121
/Li B. Zhen/
Supervisory Patent Examiner, Art Unit 2121