Geosteering control framework

Abstract

A system and method can include receiving data acquired by a downhole tool of a tool string disposed at least partially within a borehole in a subsurface region. The system and method can also include inverting the data using a trained machine learning model to generate structural features of the subsurface region. The system and method can also include controlling an operation of the tool string based at least in part on the structural features of the subsurface region.

Description

[0001] Cross-references to related applications

[0002] This application claims priority to U.S. Nonprovisional Application No. 18 / 433842, filed February 6, 2024, the entire contents of which are incorporated herein by reference and shall be considered a part of this specification. Background Technology

[0003] Geological steering can utilize downhole geological logging data to provide directional control of the drill bit in the drill string, for example, to keep the directional wellbore within the producing formation. In various scenarios, geological steering can be used to keep the wellbore within a specific portion of the reservoir to minimize gas or water channeling and maximize hydrocarbon production. Summary of the Invention

[0004] A method may include: receiving data acquired by a downhole tool of a tool string at least partially located within a wellbore in a subsurface area; inverting the data using a trained machine learning model to generate structural features of the subsurface area; and controlling the operation of the tool string at least partially based on the structural features of the subsurface area. A system may include: a processor; processor-accessible memory; and processor-executable instructions stored in the memory and executable by the processor to instruct the system to perform the following actions: receiving data acquired by a downhole tool of a tool string at least partially located within a wellbore in a subsurface area; inverting the data using a trained machine learning model to generate structural features of the subsurface area; and controlling the operation of the tool string at least partially based on the structural features of the subsurface area. One or more non-transitory computer-readable storage media may include processor-executable instructions to instruct the processor to perform the following actions: receiving data acquired by a downhole tool of a tool string at least partially located within a wellbore in a subsurface area; inverting the data using a trained machine learning model to generate structural features of the subsurface area; and controlling the operation of the tool string at least partially based on the structural features of the subsurface area. Various other apparatuses, systems, methods, etc., are also disclosed herein.

[0005] The present invention is provided to introduce some concepts further described in the detailed embodiments below. The present invention is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used as an auxiliary basis for limiting the scope of the claimed subject matter. Attached Figure Description

[0006] The features and advantages of the described implementation can be more easily understood by referring to the accompanying drawings and the following description.

[0007] Figure 1 An example of equipment in a geological environment is shown;

[0008] Figure 2 Examples of the system and well types are shown;

[0009] Figure 3 An example of a geological environment with a wellbore is shown, as well as an example of a portion of a drill string that may include various components;

[0010] Figure 4 An example of a portion of a drill string that may include various components is shown;

[0011] Figure 5 An example of a well logging curve is shown;

[0012] Figure 6 An example of the workflow is shown;

[0013] Figure 7 An example of the workflow is shown;

[0014] Figure 8 An example of the workflow is shown;

[0015] Figure 9 An example of the workflow is shown;

[0016] Figure 10 An example of data augmentation is shown;

[0017] Figure 11 An example of the workflow is shown;

[0018] Figure 12 Examples of environments and frameworks are shown;

[0019] Figure 13 Examples of methods and systems are shown; and

[0020] Figure 14 Examples of computing and networking devices are shown. Detailed Implementation

[0021] The following description includes embodiments of the best mode currently contemplated for implementing the described implementations. This description should not be construed as limiting, but merely as an attempt to describe the general principles of the implementations. The scope of the described implementations should be determined with reference to the claims.

[0022] As mentioned above, geological steering can utilize downhole geological logging measurements to provide directional control of the drill bit in the drill string, for example, to keep the directional wellbore within the producing formation. In various cases, geological steering can be used to keep the wellbore within a specific portion of the reservoir to minimize gas or water channeling and maximize hydrocarbon production.

[0023] A wellbore may be referred to as a wellbore and may include an open-hole section or an uncased section and / or may include a casing section. The wellbore may be defined by a wellbore wall composed of the rock that defines the wellbore. A well or wellbore, whether used for exploration, sensing, production, injection, or other operations, can be planned. Such a process is often referred to as well planning, which involves planning a path within the geological environment. Such a path may be referred to as a trajectory, which may include coordinates in a three-dimensional coordinate system, where the measurement along the trajectory may be measured depth (MD), total vertical depth (TVD), or another type of measurement.

[0024] As an example, drilling may include the use of one or more logging tools that may perform one or more logging operations (e.g., at rest, during descent, during tripping, etc.) with the drill string during drilling or otherwise. As an example, drilling and / or one or more other operations may be performed in response to measurement results. For example, a logging while drilling operation may acquire measurement results and adjust the drilling based at least in part on such measurements. In such an example, adjustment may be made by actuating one or more geosteering actuators, which may be used to orient the drill bit of the drill string.

[0025] Figure 1 An example of geological environment 120 is shown. Figure 1 In this context, the geological environment 120 may be a sedimentary basin comprising layers (e.g., bedding), including reservoir 121, and may be traversed, for example, by faults 123 (e.g., one or more faults). As an example, the geological environment 120 may be equipped with various sensors, detectors, actuators, etc. For example, device 122 may include communication circuitry to receive and / or transmit information relating to one or more networks 125. Such information may include information associated with downhole device 124, which may be a device for information acquisition, assisting resource recovery, etc. For example, downhole device 124 may be located in a wellbore 142 formed by the wellbore walls of one or more types of rock. Other devices 126 may be located remotely from the well site and include sensing, detection, transmission, or other circuitry. Such devices may include storage and communication circuitry to store and transmit data, instructions, etc. As an example, one or more devices may be used to measure, collect, communicate, store, analyze, etc., data (e.g., for one or more produced resources). As an example, one or more satellites may be provided for communication, data acquisition, geolocation, etc. For example, Figure 1Satellite 150 communicating with network 125 is shown. The satellite can be configured for communication. It should be noted that satellite 150 may additionally or alternatively include circuitry for imaging (e.g., spatial, spectral, temporal, radiometric measurements, etc.).

[0026] Figure 1 The geological environment 120 is also shown to optionally include devices 127 and 128 associated with well 144, which includes a basic horizontal portion that may intersect with one or more fractures 129. For example, consider a well in a shale formation, which may include natural fractures, artificial fractures (e.g., hydraulic fractures), or a combination of natural and artificial fractures. As an example, wells may be drilled for reservoirs that extend laterally. In such examples, properties, stresses, etc., may vary laterally, and assessment of such variations can aid in planning, operations, etc., for reservoir development (e.g., through fracturing, injection, production, etc.). As an example, devices 127 and / or 128 may include components, systems, multiple systems, etc., for fracturing, seismic sensing, seismic data analysis, NMR logging, assessment of one or more fractures, injection, production, etc. As an example, devices 127 and / or 128 may be used to measure, collect, communicate, store, analyze, etc., data such as formation data, fluid data, production data (e.g., for one or more produced resources). As an example, one or more satellites, such as Satellite 150, can be provided for communication, data acquisition, etc.

[0027] Figure 1 Examples of devices 170 and 180 are also shown. Such devices can be systems of components and can be adapted for use in geological environment 120. Although devices 170 and 180 are shown as land-based, various components can be adapted for use in marine systems. Figure 1 As shown, device 180 can be a mobile device carried by a vehicle; it should be noted that device 170 can be assembled, disassembled, transported, and reassembled.

[0028] Equipment 170 includes a platform 171, a derrick 172, a crane 173, a wire rope 174, a traveling block 175, a winch 176, and a work platform 177 (e.g., a second-level platform). As an example, the wire rope 174 can be controlled at least partially via the winch 176, causing the traveling block 175 to move vertically relative to the platform 171. For example, by tightening the wire rope 174, the winch 176 can allow the wire rope 174 to travel through the crane 173 and lift the traveling block 175 upwards away from the platform 171; while by releasing the wire rope 174, the winch 176 can allow the wire rope 174 to travel through the crane 173 and lower the traveling block 175 toward the platform 171. When the traveling block 175 carries a string of tubing (e.g., casing), tracking the movement of the traveling block 175 can provide an indication of how much tubing has been lowered. As shown in the figure, the movement of the traveling carriage assembly 175 can move the equipment into and out of the wellbore 178 in the formation 179.

[0029] The derrick can be a structure for supporting a crane and a traveling block, the traveling block being at least partially operably connected to the crane via wire ropes. The derrick can be pyramidal in shape and provide a suitable strength-to-weight ratio. The derrick can be moved as a whole or in parts (e.g., for assembly and disassembly).

[0030] As an example, a winch may include a drum, brake, power source, and various auxiliary devices. The winch can controllably release and wind up a wire rope. The wire rope may loop around the overhead crane and be connected to the traveling block to obtain mechanical gain via a "pulley block" or "pulley". Releasing and winding up the wire rope can lower the traveling block (e.g., and any object that may be suspended below it) into or out of the wellbore. Releasing the wire rope can be gravity-driven, while winding up can be driven by a motor, engine, etc. (e.g., an electric motor, a diesel engine, etc.).

[0031] As an example, a crane may include a set of pulleys located at or near the top of the derrick or mast, with a wire rope passing through them. A traveling block may include a set of pulleys that can be moved up and down within the derrick or mast by the wire rope passing through both the traveling block's pulleys and the crane's pulleys. The crane, traveling block, and wire rope can form a pulley system for the derrick or mast that allows heavy loads (e.g., drill string, casing, casing, liner, etc.) to be lifted up or lowered into the wellbore. As an example, the wire rope may have a diameter of about one centimeter to about five centimeters, and may be, for example, a steel cable. By using a set of pulleys, such a wire rope can carry a heavier load than it could support as a single strand.

[0032] As an example, a derrickman can be a drilling rig operator working on a platform attached to a derrick or mast. The derrick may include a work platform on which the derrickman can stand. As an example, such a work platform may be located approximately 10 meters or higher above the drilling rig. In an operation known as tripping out of the well (TOH, or POOH), the derrickman may wear a safety harness that allows him to lean out from the work platform (e.g., a second platform) to reach the tubing string located at or near the center of the derrick or mast, and to wrap a rope around the tubing string and pull it back into its storage position (e.g., a finger beam), for example, until it is necessary to lower the tubing string back into the wellbore. As an example, the drilling rig may include automated tubing string handling equipment, allowing the derrickman to control the machinery rather than physically moving the tubing string.

[0033] As an example, tripping in and out of the wellbore can refer to the actions of pulling equipment out of the wellbore (e.g., tripping out (POOH)) and / or placing equipment into the wellbore (e.g., running in (RIH)). As an example, equipment may include a drill string that can be pulled out of the wellbore and / or placed or repositioned into the wellbore. As an example, tripping in and out of the drill string can be performed when the drill bit has become dull or otherwise ceased to drill efficiently and a drill bit replacement is required. As an example, tripping in and out of the drill string can be performed when changing the well section diameter, for example, when changing equipment to drill a section with a smaller wellbore diameter after completing a section with a larger wellbore diameter.

[0034] Figure 2 An example of a well site system 200 is shown (e.g., located at a well site that may be onshore or offshore). As shown, the well site system 200 may include: a mud tank 201 for containing mud and other materials (e.g., where the mud may be a drilling fluid that helps transport cuttings, etc.); a suction line 203 serving as the inlet of a mud pump 204 for pumping mud from the mud tank 201 to a vibrating hose 206; a winch 207 for winding the drilling wire rope 212; a riser 208 for receiving mud from the vibrating hose 206; a square drill pipe hose 209 for receiving mud from the riser 208; one or more gooseneck pipes 210; a traveling block 211; and a crane 213 for carrying the traveling block 211 via one or more drilling wire ropes 212 (see example...). Figure 1 173 (the overhead crane); 214 (see example) Figure 1 The derrick 172); quay drill pipe 218 or top drive 240; quay drill pipe filler 219; rotary table 220; drill platform 221; bell mouth 222; one or more blowout preventers (BOP) 223; drill string 225; drill bit 226; casing head 227; and flow pipe 228 for conveying mud and other materials to, for example, mud tank 201.

[0035] exist Figure 2In the example system, a wellbore 232 is formed in the underground formation 230 by rotary drilling; it should be noted that various example embodiments may also use directional drilling or one or more other types of drilling.

[0036] like Figure 2 As shown in the example, drill string 225 is suspended within wellbore 232 and has drill string assembly 250 comprising drill bit 226 at its lower end. As an example, drill string assembly 250 may be bottom hole assembly (BHA).

[0037] The well site system 200 can be used for the operation of the drill string 225 and other operations. As shown, the well site system 200 includes a platform 215 and a derrick 214 located above the wellbore 232. As described above, the well site system 200 may include a rotary table 220 through which the drill string 225 passes.

[0038] like Figure 2 As shown in the example, the well site system 200 may include a crisscross drill pipe 218 and associated components, or a top drive 240 and associated components. For the crisscross drill pipe example, the crisscross drill pipe 218 may be a square or hexagonal metal / alloy rod with holes drilled inside to serve as a mud flow path. The crisscross drill pipe 218 can be used to transmit rotational motion from the rotary table 220 to the drill string 225 via a crisscross drill pipe insert 219, while allowing the drill string 225 to descend or rise during rotation. The crisscross drill pipe 218 may pass through the crisscross drill pipe insert 219, which may be driven by the rotary table 220. As an example, the rotary table 220 may include a main insert operatively coupled to the crisscross drill pipe insert 219, such that rotation of the rotary table 220 causes the crisscross drill pipe insert 219 to rotate, and consequently the crisscross drill pipe 218. The square drill pipe filler 219 may include an internal profile that matches the external profile (e.g., square, hexagonal, etc.) of the square drill pipe 218; however, its size is slightly larger, allowing the square drill pipe 218 to move freely up and down within the square drill pipe filler 219.

[0039] For the top drive example, top drive 240 can provide functions performed by the kelly and rotary table. Top drive 240 can rotate drill string 225. As an example, top drive 240 may include one or more motors (e.g., electric motors and / or hydraulic motors) connected via a suitable gear mechanism to a short section of pipe called a top drive quill, which can be screwed into a protective joint or the drill string 225 itself. Top drive 240 may be suspended on traveling block 211, thereby allowing the rotary mechanism to move freely up and down along derrick 214. As an example, top drive 240 may allow drilling to be performed using more standoffs than with the kelly / rotary table method.

[0040] exist Figure 2In the example, mud tank 201 can contain mud, which can be one or more types of drilling fluid. As an example, a wellbore can be drilled to produce fluid, inject fluid, or both (e.g., hydrocarbons, minerals, water, etc.).

[0041] exist Figure 2 In the example, drill string 225 (e.g., including one or more downhole tools) may consist of a series of pipes threaded together to form a long tube, with drill bit 226 located at its lower end. As drill string 225 is advanced into the wellbore for drilling, at some point before or simultaneously with drilling, mud may be pumped by pump 204 from mud tank 201 (e.g., or other source) via lines 206, 208, and 209 to a port of angular drill pipe 218, or, for example, to a port of top drive 240. Subsequently, the mud may flow through one or more channels in drill string 225 and exit from a port located on drill bit 226 (see, for example, directional arrows). As the mud exits drill string 225 through the port in drill bit 226, it may then circulate upwards through an annular region between one or more outer surfaces of drill string 225 and one or more surrounding walls (e.g., open hole, casing, etc.), as indicated by directional arrows. In this way, the mud lubricates the drill bit 226 and carries thermal energy (e.g., frictional energy or other energy) and formation cuttings to the surface, where the mud (e.g., and the cuttings) can be returned to the mud tank 201, for example, for recycling (e.g., by processing to remove cuttings, etc.).

[0042] The drilling mud pumped into the drill string 225 by pump 204 forms a mud cake lining the wellbore after leaving the drill string 225. This mud cake, among other functions, reduces friction between the drill string 225 and one or more surrounding walls (e.g., the wellbore, casing, etc.). This reduction in friction facilitates the advancement or withdrawal of the drill string 225. During drilling operations, the entire drill string 225 can be pulled out of the wellbore and optionally replaced, for example, with a new or sharpened drill bit, a smaller diameter drill string, etc. As described above, the action of pulling the drill string out of or reinserting it into the wellbore is called tripping. Depending on the direction of tripping, tripping can be referred to as upward tripping or outward tripping, or downward tripping or inward tripping.

[0043] As an example, consider a down-running drill, where, when the drill bit 226 of the drill string 225 reaches the bottom of the wellbore, mud is pumped to lubricate the drill bit 226 for use in enlarging the wellbore through drilling. As described above, the mud can be pumped by pump 204 into the channels of the drill string 225, and while filling the channels, the mud can be used as a transmission medium to transmit energy, for example, energy that can encode information, as in mud pulse telemetry.

[0044] As an example, a mud pulse telemetry device may include downhole equipment configured to change the pressure in the mud to generate one or more acoustic waves on which information can be modulated. In such an example, information from the downhole equipment (e.g., one or more components of drill string 225) can be transmitted to a surface device, which can then relay such information to other devices for processing, control, etc.

[0045] As an example, the telemetry device can operate by transmitting energy via the drill string 225 itself. For example, consider a signal generator that applies an encoded energy signal to the drill string 225 and a repeater that can receive such energy and repeat it for further transmission of the encoded energy signal (e.g., information, etc.).

[0046] As an example, drill string 225 may be equipped with telemetry device 252, which includes: a rotatable drive shaft; a turbine impeller mechanically coupled to the drive shaft such that the drilling mud can rotate the turbine impeller; a modulator rotor mechanically coupled to the drive shaft such that the rotation of the turbine impeller causes the modulator rotor to rotate; a modulator stator mounted near or adjacent to the modulator rotor such that the rotation of the modulator rotor relative to the modulator stator generates pressure pulses in the drilling mud; and a controllable brake for selectively braking the rotation of the modulator rotor to modulate the pressure pulses. In such an example, an alternator may be coupled to the aforementioned drive shaft, wherein the alternator includes at least one stator winding electrically connected to a control circuit to selectively short-circuit the at least one stator winding, thereby electromagnetically braking the alternator and thus selectively braking the rotation of the modulator rotor to modulate the pressure pulses in the drilling mud.

[0047] exist Figure 2 In one example, the wellhead control and / or data acquisition system 262 may include circuitry for sensing pressure pulses generated by the telemetry device 252, and for example, transmitting the sensed pressure pulses or information derived therefrom for processing, control, etc.

[0048] The component 250 shown in the example includes a logging-while-drilling (LWD) module 254, a measurement-while-drilling (MWD) module 256, an optional module 258, a rotary steerable system (RSS) and / or a motor 260, and a drill bit 226. Such components or modules may be referred to as tools, and the drill string may include multiple tools.

[0049] For RSS, it relates to techniques used in directional drilling. Directional drilling involves drilling into the ground to create an skewed wellbore, such that the trajectory of the wellbore is not vertical; instead, the trajectory deviates from the vertical along one or more sections of the wellbore. As an example, consider a target located at a certain lateral distance from a surface location where a drilling rig can be placed. In such an example, drilling can begin in a vertical section and then deviate from the vertical, so that the wellbore is aligned with and eventually reaches the target. Directional drilling can be implemented in situations where the target may not be accessible from a vertical location on the surface, where there are materials underground that may hinder drilling or otherwise adversely affect it (e.g., consider salt domes), where the formation extends extensively laterally (e.g., consider relatively thin but extensively lateral reservoirs), where multiple wellbores need to be drilled from a single surface wellbore, or where it is desirable to create a rescue well.

[0050] One approach to directional drilling involves mud motors; it should be noted that mud motors can present some challenges depending on factors such as the rate of penetration (ROP) and the transfer of weight to the drill bit due to friction (e.g., weight on bit (WOB)). A mud motor can be a positive displacement motor (PDM) that operates during directional drilling to drive the drill bit. The PDM operates as drilling fluid is pumped through it, converting the hydraulic power of the drilling fluid into mechanical power to rotate the drill bit. When the drill string is not rotating from the surface, the PDM can operate in a so-called slippery mode.

[0051] RSS (Remote Suspension Controller) can be used in directional drilling in the presence of continuous rotation from surface equipment, which can mitigate slippage of the steerable motor (e.g., PDM). RSS can be deployed when performing directional drilling (e.g., deviated wells, horizontal wells, or extended reach wells). RSS can be designed to minimize interaction with the wellbore wall, which can help maintain wellbore quality. RSS can be designed to apply relatively consistent lateral forces, similar to a stabilizer rotating with the drill string, or to orient the drill bit in the desired direction while rotating continuously at the same rate as the drill string.

[0052] LWD module 254 can be housed in a suitable type of drill collar and may include one or more selected types of logging tools (e.g., one or more NMR units, etc.). It should also be understood that one or more LWD and / or MWD modules may be employed at one or more locations. The LWD module may include capabilities for measuring, processing, and storing information, as well as for communicating with surface equipment. In the illustrated example, LWD module 254 may include seismic measuring devices, NMR measuring devices, etc.

[0053] MWD module 256 can be housed in a suitable type of drill collar and may include one or more devices for measuring the characteristics of drill string 225 and drill bit 226. As an example, MWD module 256 may include devices for generating electricity, for example, to power various components of drill string 225. As an example, MWD module 256 may include telemetry device 252, for example, where a turbine impeller can generate electricity through the flow of mud; it should be understood that other power sources and / or battery systems may be used to power various components. As an example, MWD module 256 may include one or more measuring devices of the following types: drill pressure measuring device, torque measuring device, vibration measuring device, impact measuring device, stick-slip measuring device, direction measuring device, and well inclination measuring device.

[0054] As an example, one or more NMR measurement devices (e.g., NMR units, etc.) may be included in a drill string (e.g., BHA, etc.), wherein, for example, the measurement results may support one or more of the following: geological guidance, geological shutdown, trajectory optimization, etc. As an example, motion characterization data may be used to control NMR measurements (e.g., acquisition, processing, quality assessment, etc.).

[0055] Figure 2 Examples of drillable wellbore types are also shown. For example, consider 272, 274, 276, and 278.

[0056] As an example, drilling operations may include directional drilling, where, for example, at least a portion of the well includes a curved axis. For example, consider a radius defining the curvature, where the well inclination angle relative to the vertical direction can vary up to an angle between approximately 30 degrees and approximately 60 degrees, or, for example, an angle of approximately 90 degrees, or possibly greater than approximately 90 degrees. As an example, the trajectory and / or drilling may be characterized in part by dogleg degree (DLS), which can be a two-dimensional parameter expressed in degrees per 30 meters (e.g., or degrees per 100 feet).

[0057] As an example, directional wells can include various shapes, each designed to meet specific operational requirements. As an example, drilling processes can be performed based on information relayed to drilling engineers. As an example, well inclination and / or direction can be modified based on information received during drilling.

[0058] As an example, wellbore deflection can be achieved in part by using downhole motors and / or turbines. For motors, consider drill strings that may include positive displacement motors (PDMs).

[0059] As an example, the system may be a steerable system and include equipment for performing methods such as geosteering. As described above, a steerable system may be or include an RSS. As an example, a steerable system may include a PDM and / or turbine in the lower part of the drill string and may have a bend mounted directly above the drill bit. The geosteering equipment of the drill string may include one or more geosteering actuators that can be used to orient the drill bit of the drill string. For example, the actuator may include a piston that moves a pad to provide a force that can be applied to the wellbore wall, thereby steering the bottomscrew assembly (e.g., orienting the drill bit of the bottomscrew assembly). As an example, the actuator may be a bend downhole motor that can be actuated via one or more processes. As an example, a bend drilling motor may have a fixed bend angle that cannot be changed during normal operation, or a variable bend angle that can be changed, for example, based on a geosteering command. As an example, for a variable bend angle drilling motor, one or more actuators may be included that can be configured to generate or change the bend angle, thereby affecting the steering behavior of the steerable system. As an example, the actuator may be a downhole actuator that can adjust the orientation downhole, and / or the actuator may be a surface actuator that performs actions above ground (e.g., on the surface) to adjust the orientation downhole.

[0060] As an example, above the PDM, an MWD device and / or LWD device can be installed to provide real-time or near-real-time data of interest (e.g., well inclination, direction, pressure, temperature, actual weight on the drill bit, torque stress, etc.). For the latter, the LWD device enables the transmission of various types of data of interest to the surface, including, for example, geological data (e.g., gamma ray logging, resistivity logging, density logging, and sonic logging, etc.).

[0061] Combining sensors that provide real-time or near-real-time information about well trajectory with one or more logging curves, for example, that geologically characterize the formation, allows for the implementation of geosteering methods. Such methods can include navigation in subsurface environments, for example, to reach a desired target or multiple targets along a desired route.

[0062] As an example, a drill string may include one or more of the following: an azimuth density neutron (ADN) tool for measuring density and porosity; a MWD tool for measuring well inclination, azimuth, and impact; a compensated dual resistivity (CDR) tool for measuring resistivity and gamma-ray related phenomena; a combined magnetic resonance (CMR) tool for measuring properties (e.g., relaxation properties, etc.); one or more variable diameter stabilizers; one or more bends; and a geological steering tool that may include a motor and optionally include devices for measuring and / or responding to one or more of well inclination, resistivity, and gamma-ray related phenomena.

[0063] As an example, tools such as ECOSCOPE (SLB, Houston, Texas) can be used to acquire measurement results. Such tools may include one or more pulsed neutron generators (PNGs) and associated detectors. These tools may include features for one or more of the following measurements: resistivity, neutron porosity, azimuth gamma rays, density, elemental trapping spectrum, and sigma. For example, consider features for one or more of the following: 2 MHz and 400 kHz propagating resistivity, elemental trapping spectrum, neutron-gamma density, trapping cross section (sigma), azimuth bulk density, azimuth photoelectric factor, azimuth natural gamma rays, density caliper, ultrasonic caliper, downhole annular pressure and temperature, triaxial shock and vibration, and near-bit borehole inclination. Such tools can be operatively coupled to one or more telemetry systems that can be used for real-time acquisition and, for example, real-time decision-making, graphical rendering, etc. As an example, such tools can be operatively coupled to one or more types of circuits that can, for example, perform calculations downhole using measurements acquired in the well.

[0064] As an example, tools such as PERISCOPE (SLB, Houston, Texas) can be used to acquire measurement results. For instance, consider measurements such as resistivity, which can be acquired using one or more types of receivers. As an example, the receiver can be or include an antenna. For example, PERISCOPE tools can include tilted antennas, axial antennas, and lateral antennas. As an example, data acquired from such tools can be used to identify layers, the number of layers, and the location of one or more layers at a distance of 1 meter or greater (e.g., reaching or exceeding 8 meters).

[0065] For Sigma measurements (e.g., Sigma data), Sigma is the macroscopic cross-section, or capture cross-section, of thermal neutron absorption by a volume of matter, measured in capture units (cu). Sigma logging profiles are the primary output of pulsed neutron capture logging and can be used for one or more purposes.

[0066] As an example, one or more types of nuclear measurements can be acquired by one or more tools, wherein such nuclear measurements may include electron density (ρ). e One or more of the following: hydrogen index (HI) and thermal neutron capture cross section (sigma or Σ).

[0067] As an example, geological steering may include intentional directional control of the wellbore based on downhole geological logging measurements to keep the directional wellbore within a desired area, zone (e.g., producing layer), etc. As another example, geological steering may include guiding the wellbore to remain within a specific portion of the reservoir, for example, to minimize gas and / or water channeling, and for example, to maximize the economic production of the well including the wellbore.

[0068] Refer again Figure 2 The well site system 200 may include one or more sensors 264 operatively coupled to a control and / or data acquisition system 262. As an example, the one or more sensors may be located at a surface location. As an example, the one or more sensors may be located at a downhole location. As an example, the one or more sensors may be located at one or more remote locations not within approximately one hundred meters of the well site system 200. As an example, the sensors may be located at an off-site well site, wherein the well site system 200 and the off-site well site are located in the same oil field (e.g., an oil field and / or a gas field).

[0069] As an example, one or more of the sensors 264 may be provided for tracking the movement of the tubing, tracking the movement of at least a portion of the drill string, etc.

[0070] As an example, system 200 may include one or more sensors 266 that can sense signals and / or transmit signals to fluid conduits such as drilling fluid conduits (e.g., drilling mud conduits). For example, in system 200, one or more sensors 266 may be operatively coupled to portions of riser 208 through which mud flows. As an example, downhole tools may generate pulses that can propagate through mud and be sensed by one or more of the one or more sensors 266 (e.g., consider mud pulse telemetry). In such an example, the downhole tool may include associated circuitry, such as encoding circuitry, that can encode the signal, for example, to reduce transmission requirements. As an example, surface circuitry may include decoding circuitry to decode encoded information transmitted at least partially via mud pulse telemetry. As an example, surface circuitry may include encoder and / or decoder circuitry, and downhole circuitry may include encoder and / or decoder circuitry. As an example, system 200 may include a transmitter that can generate signals that can be transmitted downhole via mud (e.g., drilling fluid) as a transmission medium.

[0071] Analysis of formation information acquired by one or more tools can reveal characteristics such as dissolution pores, dissolution surfaces (e.g., dissolution along bedding planes), stress-related features, dip events, etc. As an example, the tools can acquire information that helps characterize the reservoir, optionally a fractured reservoir, where fractures can be natural and / or artificial (e.g., hydraulic fractures). The reservoir can be a porous formation where fluids can reside within the various pores of the porous formation and are readily mobile (e.g., to produce fluids from the reservoir). As an example, information acquired by one or more tools can be analyzed using a framework such as the TECHLOG framework (SLB, Houston, Texas). As an example, the TECHLOG framework can interoperate with one or more other frameworks, such as the PETREL framework (SLB, Houston, Texas). As an example, computational environments such as the DELFI environment (SLB, Houston, Texas) can be utilized, which can be used to leverage the PETREL framework and other frameworks, optionally in an interdependent manner.

[0072] Figure 3 An example of a drilling assembly 300 in a geological setting 301 is shown, the geological setting 301 including a wellbore 303, wherein the drilling assembly 300 (e.g., a drill string) includes a drill bit 304 and a motor section 310, wherein the motor section 310 can drive the drill bit 304 (e.g., to rotate the drill bit 304 and deepen the wellbore 303).

[0073] As shown in the figure, the motor section 310 may include a vent valve 312, a power section 314, a ground-adjustable housing 316, a transmission assembly 318, a bearing section 320, and a drive shaft 322, which can be operatively connected to a drill bit such as a drill bit 304. Figure 3 The motor segment 310 can be a POWERPAK series motor segment (SLB, Houston, Texas) or another type of motor segment.

[0074] The power section converts the hydraulic energy from the drilling fluid into mechanical power to rotate the drill bit. Consider, for example, a reverse application of the Moino pump principle. During operation, drilling fluid can be pumped at pressure to rotate the rotor within the stator into the power section, where the rotational force is transmitted to the drill bit via a drive shaft and a transmission shaft.

[0075] Figure 3Examples of components 340 are also shown, such as sensor 350, circuitry 360, and geoguided actuator 370. As shown, sensor 350 may include conductivity and dielectric sensors 352, gamma sensors 354, and one or more other sensors 356. As shown, circuitry 360 may include processor 362, memory 364, and one or more other types of circuitry 366. As shown, geoguided actuator 370 may be operatively coupled to circuitry 360 and sensor 350. For example, circuitry 360 may process signals from sensor 350 (e.g., measurement results or sensor data) to generate one or more commands for actuating geoguided actuator 370. Figure 3 In the example, the geological steering actuator 370 can be used for one or more of PDM actuation and elbow actuation, for example, to orient the drill bit 304.

[0076] Figure 4 An example of a drilling assembly 400 (e.g., a portion of a drill string) is shown, which includes a drill bit 404 and a rotary steerable system (RSS) 410. As described above, the RSS can be used for directional drilling, including geological steering. As an example, the RSS 410 may include one or more features of a POWERDRIVE ARCHER RSS (SLB, Houston, Texas).

[0077] Figure 4 Examples of components 440 are also shown, such as sensor 450, circuitry 460, and geoguided actuator 470. As shown, sensor 450 may include conductivity and dielectric sensors 452, gamma sensors 454, and one or more other sensors 456. As shown, circuitry 460 may include processor 462, memory 464, and one or more other types of circuitry 466. As shown, geoguided actuator 470 may be operatively coupled to circuitry 460 and sensor 450. For example, circuitry 460 may process signals from sensor 450 (e.g., measurement results or sensor data) to generate one or more commands for actuating geoguided actuator 470. Figure 4 In the example, the geological guidance actuator 470 can be used for RSS actuation, for example, to orient the drill bit 404.

[0078] As an example, drilling assembly 400 may include one or more of the following: near-bit continuous deviation and azimuth measurement unit or sub, near-bit azimuth gamma ray measurement unit or sub, and one or more other types of measurement unit or sub.

[0079] As an example, a drilling assembly may include one or more types of circuitry. For instance, consider a processing unit with a processor and associated memory, where one or more sensors can generate signals that can be received by the processing unit. In such an example, the processing unit can perform calculations using information from the signals (e.g., measurement results, etc.) to generate commands for geosteering. In such an example, the drilling assembly may be able to perform downhole geosteering at least partially based on geosteering commands generated downhole, without transmitting information uphole to a controller and subsequently downhole to the geosteering equipment. In such an example, at least some types of geosteering processes can be executed more quickly in response to sensor signals. For example, consider sensor signals indicating one or more of the presence, amount, type of clay, and boundaries as interfaces between layers, where the downhole geosteering equipment can steer the drill bit based on one or more of such sensor signals.

[0080] As an example, an electromagnetic conductivity measurement tool (ECM tool) can be implemented as a cable tool and / or as an LWD tool to generate dielectric constant and conductivity measurements at each of one or more frequencies, which can be interpreted using a rock physics model. In such an example, the model's output parameters may include aquifer porosity (and therefore, water saturation if total porosity is known) and water salinity. As an example, parameters output from measurements taken using an ECM tool (e.g., induction, propagation, etc.) may include one or more of the following: overall formation cation exchange capacity (CEC), water saturation (Sw), native water salinity, Archie cementation index, and Archie saturation index.

[0081] Figure 5 An example of logging curve 500 from a dielectric scanner tool is shown. Various types of tools can be used to acquire logging curves; for example, consider tools used for the following data: lateral logging operation data at 35 Hz and 280 Hz, drill bit resistivity (RAB) operation data at 1.5 kHz, array induction tool (AIT) operation data at 26 kHz and 52 kHz, compensated array resistivity (ARC) operation data at 0.4 MHz and 2 MHz, dielectric propagation tool (DPT) operation data at 25 MHz, array dielectric tool (ADT) operation data at 0.9 GHz, and electromagnetic propagation tool (EPT) operation data at 1.1 GHz.

[0082] exist Figure 5In the example, in logging curve 500, a depth-related salinity logging curve is shown together with a depth-related resistivity logging curve and a porosity logging curve, wherein the resistivity logging curve shows array inductive resistivity (2 ft, A90), invasion zone resistivity and dielectric scanner invasion zone resistivity, and wherein the porosity logging curve includes total porosity and dielectric scanner water-bearing porosity, and optionally includes hydrocarbons (e.g., at least in part based on water-bearing porosity).

[0083] As an example, the framework may include components for implementing a deep learning method for two-dimensional inversion of electromagnetic data. In such an example, the framework could be used to generate output for one or more geologically guided operations.

[0084] As an example, a toolset can include various tools that can acquire downhole measurements within a wellbore in a region for calculating a resistivity profile of that region. As another example, the framework can generate one or more resistivity profiles in a region with acceptable accuracy within seconds. Such a method can be referred to as a real-time approach, or, for example, a near-real-time approach (e.g., generating output in less than 60 seconds).

[0085] As an example, the framework can leverage a trained deep learning model (e.g., a deep learning-type machine learning model) that can accelerate the generation of outputs such as one or more resistivity profiles. For input, the framework can receive resistivity measurements from one or more tools (e.g., consider the ECOSCOPE tool, the PERISCOPE tool, etc.).

[0086] As an example, a workflow could include generating a trained deep learning model. For instance, such a workflow could include generating input and output data for training the deep learning model, where the input data can come from one or more types of tools, and the output data is a corresponding resistivity profile in a region including the wellbore. As an example, the input and output data could be organized into one or more matrices.

[0087] As mentioned above, deep learning models can be utilized. As an example, the Transformer can be used, where the Transformer can be a deep learning architecture based on a multi-head attention mechanism. As an example, the Transformer can be an architecture that does not include one or more recurrent units. In such an example, the training time can be less than that of a recurrent neural architecture (e.g., Long Short-Term Memory (LSTM)). For the multi-head approach, the input can be split into encoded tokens, where each token can be converted into a vector. As an example, at each layer, each token can establish contextual relationships with other tokens via a parallel multi-head attention mechanism, allowing one or more signals from one or more tokens to be amplified, while one or more signals from one or more other tokens to be attenuated. As an example, the Transformer can be viewed as the backbone of a neural network architecture that can be used to train deep learning models.

[0088] As an example, the framework can be used to implement a trained deep learning model to generate predictions. For instance, consider a framework that can receive real-time downhole measurements and make predictions based on those measurements using a trained deep learning model, where such predictions can be generated in real-time or near real-time (e.g., in less than 60 seconds).

[0089] As an example, the framework can be adapted to be implemented using surface equipment and / or downhole equipment. As an example, a trained deep learning model can be adapted to be embedded in surface equipment and / or downhole equipment, where downhole measurement results (e.g., downhole data) can be generated and received to produce output.

[0090] As an example, a trained deep learning model can be used to perform multidimensional inversion of electromagnetic data. As an example, a trained deep learning model can be used to receive one or more other types of data. As an example, multidimensional inversion of electromagnetic data can be used to perform two-dimensional inversion of electromagnetic data to provide characteristics of a region adjacent to downhole tools located within a wellbore in said region. In such an example, the region may be or includes a reservoir. In such an example, geosteering can be implemented by controlling the orientation of the drill bit on the drill string, at least in part based on the output of the trained deep learning model.

[0091] As described above, reflection seismology can be utilized in regional seismic surveys to generate seismic data, which can, for example, be inverted to generate an acoustic model of the region. In such an example, the region may include one or more reservoirs. As an example, at least seismic data can be used to plan a wellbore trajectory, where the wellbore trajectory connects a surface location (e.g., land surface, seabed surface, etc.) to a reservoir. As an example, such a planned wellbore trajectory can serve as the basis for drilling, where the drilling can be further guided via geosteering. As described above, a wellbore can be drilled into the reservoir, where the drilling may include geosteering such that the wellbore reaches the reservoir and / or one or more portions of the wellbore contact the reservoir, for example, to maximize the contact between the wellbore and the reservoir for the purpose of producing and / or injecting fluids.

[0092] The use of geosteering can make wellbore drilling relative to the reservoir more accurate. For example, seismic data may have low resolution, where various subsurface structures may only be known within a range on the order of meters, such as ±10 to 20 meters. For various types of reservoirs, uncertainties on the order of about 10 meters in reservoir location can lead to missed reservoirs, which in turn can reduce the ability to produce fluids from and / or inject fluids into the reservoir (e.g., considering water injection, chemical injection, carbon sequestration, etc.). As mentioned above, geosteering can improve drilling accuracy by reducing the uncertainties caused by the resolution of seismic data, which in turn can improve wellbore construction and field operations (e.g., production, etc.).

[0093] As an example, inversion results generated by a trained deep learning model can be used to adjust the wellbore trajectory during drilling. As mentioned above, this approach can be used to adjust the drilling in real time during drilling, where, for example, the drill bit can be guided by one or more mechanisms (e.g., a geological steering mechanism, etc.).

[0094] As an example, a workflow could include preparing data for training with input and output data already collected. In such an example, the input and output data can be used to train a deep learning model; it should be noted that data can be reserved for testing, hyperparameter tuning, and one or more other purposes. In such an example, once the trained deep learning model is generated, it can be deployed for field operations. For example, consider local deployment as a model embedded in surface and / or downhole equipment, and / or remote hosting on devices such as dedicated servers, cloud platforms, etc. Once deployed, the trained deep learning model can be used to receive input data collected by one or more downhole tools on a toolchain at least partially located within a wellbore in an area, and to predict characteristics of that area, where, for example, the drill bit of the toolchain can be controlled in terms of orientation and / or one or more other parameters to improve one or more of wellbore path, bit wear, mechanical rate of penetration, depth of cut, wellbore quality, etc.

[0095] As an example, a trained deep learning model can undergo further training, testing, tuning, etc. For instance, as additional data becomes available, at least a portion of the additional data can be used to improve the trained deep learning model. As an example, additional data can be acquired during drilling guided by the output of the trained deep learning model, where background processes can be executed to revise the trained deep learning model and replace the existing trained deep learning model when the revised trained deep learning model is indicated to provide better output relative to the existing trained deep learning model.

[0096] As an example, the framework can operate for various scenarios, such as applying a trained deep learning model to perform inversion; it should be noted that, for example, in situations involving input and output data with a fixed format, one or more other tasks can be appropriately performed. As an example, as more and more data is collected and received, the framework can provide higher-quality job deliverables, allowing one or more models to be retrained, further tuned, etc., to achieve improvements beyond what the current model could possibly offer. As an example, the framework can be used to implement a continuous improvement loop, where data can be received for improvement, and such data can be collected for current jobs, parallel jobs, historical jobs where data becomes available, synthetic jobs (e.g., simulated jobs), augmentation jobs (e.g., via data augmentation, etc.).

[0097] As an example, the framework can provide time-saving explanations, knowledge accumulation, and more consistent product quality, where, for example, the wellbore can be viewed as a manufactured product. As mentioned above, the operation can involve geological steering, where, for example, a trained deep learning model can be used to automate one or more aspects of geological steering. For example, consider a control loop that generates downhole data and uses said downhole data to predict one or more geological steering parameters, such that geological steering can be at least partially automated in the control loop (e.g., consider a closed loop or a loop with personnel who can review and validate control commands, etc.).

[0098] As an example, a framework can reduce the time required for one or more types of geosteering decisions. Consider, for instance, a manual approach where a driller visually reviews a graph presented on a monitor, such as a 2D graph generated by an iterative inversion process, where the 2D graph corresponds to a slice across a region including the wellbore, and within the wellbore is a toolchain configured with geosteering capabilities. In such an approach, the driller may first have to wait for the iterative inversion process to complete before visually assessing the boundary between the reservoir and another layer. Once the driller has made such a visual assessment, they can then enter one or more geosteering commands to drill into the reservoir, which may include drilling in a direction away from another layer. In contrast, a framework that can implement a trained deep learning model can rapidly generate 2D graphs and / or one or more geosteering commands by referencing a multidimensional space. Such a framework can significantly save time and cost, especially because it can operate using a trained deep learning model instead of an iterative inversion process. As described above, given that trained deep learning models can generate outputs rapidly, such as in real-time or near real-time (e.g., less than 60 seconds), latency can be relatively insignificant and at a level sufficient for real-time control (e.g., considering the rate of drilling and the time required to effectively implement geosteering commands). As also described above, tools or toolchains can include embedded trained deep learning models, where latency in data acquisition and model-based prediction and / or geosteering command generation can be considerably low. In this manner, one or more aspects of geosteering can be automated downhole at a speed that can be faster than telemetry (e.g., mud pulse telemetry) used for uphole and / or downhole transmission.

[0099] Figure 6Examples of workflows 610 and 620 are shown. As illustrated, workflow 610 involves receiving data from a data framework (e.g., INTERACT framework, SLB, Houston, Texas, etc.) and performing an iterative inversion based on an inversion computation block 621 using high-performance computing (HPC) and / or a cloud platform engine ecosystem (EESy), where the iterative inversion may take minutes to hours to generate results. As illustrated, workflow 620 includes: an inversion computation block 621; a data block 622 for a local data pool; a training block 623 for training a deep learning model; a deployment block 624 for deploying the trained deep learning model; a prediction block 625 for generating predictions of the inversion results, which can be executed in seconds (e.g., less than 60 seconds); and a visualization and / or control block 626 for visualizing and / or controlling the predictions.

[0100] Figure 7 An example of a workflow 700 that can be implemented before and / or during field operations involving drilling in a region is shown. For example, consider a workflow that may include pre-operation planning, which can be used to generate an appropriately trained machine learning model, such as a deep learning model.

[0101] like Figure 7 As shown, workflow 700 may include utilizing a computing engine 710, which can receive data and generate points that can be received by optimization service 720. As shown, optimization service 720 can generate outputs that can be combined with corresponding data to generate multiple model variants, such as a model optimized using a first point, a model optimized using the first and second points, a model optimized using the first, second, and third points, and so on, up to multiple points represented by n. As shown, each model can be used to generate corresponding points derived from the model, these points in... Figure 7 The numbers in the middle represent artificial intelligence (AI) points 1 to n.

[0102] As an example, during a real-time job, multiple inverters can be utilized, which, for example, may include one or more machine learning model-based inverters and one or more iterative inverters, wherein, for example, the machine learning model-based inverter generates inversion results faster than the iterative inverter. In such an example, a background process may utilize one or more iterative inverters, while a foreground process may utilize one or more machine learning model-based inverters. In this manner, the output of one or more iterative inverters can be used to additionally train a version of the currently implemented machine learning model-based inverter, wherein tests can be performed to evaluate the performance of the additionally trained machine learning model-based inverter relative to the currently implemented version. In such an example, once performance is improved (e.g., according to one or more evaluation criteria, which may include factors such as memory utilization, processor requirements, etc.), the currently implemented version can be replaced with the improved version.

[0103] As an example, the input and output data of the newly computed inversion result can be used to retrain a machine learning model to generate an optimized machine learning model (e.g., for a specific point in time). As described above, the optimized machine learning model can then be used to generate predictions (e.g., the inversion result) in response to received real-time data. In such an example, the predictions (e.g., the inversion result) of the optimized machine learning model can be more accurate than those of previous versions (e.g., the default machine learning model, a previously optimized version, etc.).

[0104] Figure 8An example of a workflow 800 is shown, which includes receiving data from a receiving area, where a machine learning model can generate inversion results. As shown, data can be acquired using one or more tools positioned within a wellbore in a geological environment 810, where an AI inversion block 820 can perform an inversion at least partially based on the acquired data to generate inversion results. As shown, the inversion results can be received by an automatic boundary picking component 830, which generates a set of points having points corresponding to various boundaries, which may be layer boundaries. As shown, the point set can be utilized by a 3D model generator 840 to generate a surface, which can be utilized by a predictive geological steering component 850, where updated surfaces can be generated for a drilling target update component 860 to receive, thereby updating the drilling trajectory target, for example, based on a more accurate assessment of one or more structures within the reservoir and / or region. As shown in the figure, given an update target, the real-time trajectory replanning component 870 can be used to generate one or more updated drilling commands, for example, for an updated trajectory. One or more of these updated drilling commands can be utilized by the adjustment component 880 to adjust the drilling direction in real time (e.g., real-time control of drilling). As described above, the wellbore trajectory can be updated and provided as input to an automatic geological steering system, which can output one or more drilling commands to adjust the drill bit direction in real time or near real time.

[0105] Figure 9 An example of workflow 900 is shown, which includes a pre-job portion 910 and a real-time portion 950. In this manner, the pre-job portion 910 can be executed before the machine learning model is implemented in the real-time portion; it should be noted that one or more aspects of the pre-job portion can be executed as background processes during the execution of the real-time portion.

[0106] For the pre-work portion 910 of workflow 900, it may include providing data, such as tool specifications, trajectory specifications, and stratigraphic characteristics in the area. As shown, the pre-work portion 910 may include generating potential geological structures 912 based at least in part on the data. In such an example, the pre-work portion 910 may be used to generate training data. For example, consider using one or more geological structures to perform forward modeling 914 to generate simulated measurement data. In such an example, a dataset including data about the geological structures and corresponding simulated measurement data may be generated. As shown, the dataset may be split into a validation set (e.g., a test set) and a training set, thereby allowing the machine learning model to be trained, retrained, and / or validated. As shown, training 916 may be used to generate a pre-trained AI model (e.g., a trained ML model). As shown, the pre-work portion 910 may be used to generate a trained machine learning model suitable for implementation 952 by the real-time portion 950 of workflow 900.

[0107] As shown in example workflow 900, the real-time portion 950 may include acquiring real-time measurements using one or more downhole tools on a toolchain configured for geological guidance, wherein a trained machine learning model may be used to generate formation predictions, which may be generated in less than 60 seconds, and for example less than 10 seconds, or for example less than 1 second, in response to receiving real-time measurements. As an example, the real-time measurements may be windowed relative to time and / or distance. For example, Figure 9 Various data channels are shown regarding the phase angle in degrees relative to the measured depth (MD) in meters. Figure 9 In the examples, the window can be for a specific distance that can be measured from the current location to a previous location, for example, considering the current location at the current time and the previous location at a previous time. As an example, the data can be averaged on the window and / or subjected to one or more other statistical techniques. As an example, the trained machine learning model can receive data over a period of time and / or a distance, for example, as sequence data and / or as a pattern. As an example, the trained machine learning model can receive data at the current depth of measurement (MD) and generate results based on said data without requiring previous data from previous depths of measurement (MD).

[0108] As described above, the input to a trained machine learning model can be an array, which can be a vector as a one-dimensional array or a matrix as a multi-dimensional array. As described above, the output of a trained machine learning model can be an array, for example, consider a two-dimensional array. As an example, the data can include measurements corresponding to positions before and / or after the current measurement depth. As described above, various types of tools can be used to acquire multi-dimensional data, which can, for example, be within a field such as an electromagnetic field. As described above, the tool can include one or more transmitters that can emit energy such as electromagnetic energy within a space in the region, where the measurements can correspond to the space as a whole, or the measurements can correspond to increments of the space (e.g., consider techniques that can provide spatial encoding via gradients, etc.).

[0109] As an example, the input to a trained machine learning model can be in the form of a two-dimensional array, which can be mapped to another two-dimensional array as the output. In such an example, the input can correspond to measurement results and spatial information, and the output can correspond to formation properties and spatial information. As mentioned above, the region can include various boundaries that define multiple layers, where each layer can be a different type of formation (e.g., rocks with different properties). During geological steering, accurately understanding the layer boundaries can help drill a well along the desired path to reach the target.

[0110] Figure 10 An example of data 1000 is shown, which includes augmented data. For example, consider data comprising real data that can represent true values, where additional data can be generated by implementing one or more augmentation techniques. For example, consider one or more geometric techniques, such as horizontal flipping, downward translation, upward translation, inserting faults with discontinuities, etc. In various cases, the amount of data used for training, testing, tuning, etc., of one or more machine learning models can be increased by applying one or more augmentation techniques that can be based on real-value data. (See also: Regarding...) Figure 9 As described in the pre-job section of workflow 900, one or more geological structures can be used to generate synthetic data. As an example, a geological structure can be subjected to one or more techniques to provide an enhanced geological structure, wherein the enhanced geological structure can be used to generate synthetic data, and wherein the enhanced geological structure and its corresponding synthetic data can be used to train, test, regulate, etc., one or more machine learning models.

[0111] As an example, data augmentation can be used to attempt to improve one or more data uncertainties in measurement data and / or structural models. For instance, some measurement data may apply to multiple different structural models; such uncertainty can be reduced by providing augmented data and the associated structural models. For example, if more types of structures and their corresponding measurement data (e.g., synthetic or other data) are considered, a trained machine learning model can map the measurement data to the actual structures more accurately.

[0112] As an example, a workflow may include training more than one machine learning model. In such an approach, the machine learning models may have different architectures and / or hyperparameter values. As an example, a common loss function such as mean squared error (MSE) may be used to compare multiple trained machine learning models. As an example, the best-performing trained machine learning model may be selected. For example, given a set of inputs, the outputs of multiple machine learning models may be compared, for instance, if the true values ​​of the given set of inputs are known. As an example, more than one trained machine learning model may be implemented as part of a real-time framework for geological guidance.

[0113] For geological structures, in geosteering, directional decisions can depend on the location of resistivity boundaries within the geological structure, the resistivity value of the layer in which the well (e.g., a horizontal well) is located, and the resistivity of one or more adjacent layers. During geosteering, clean water-bearing formations and shale can exist within the area. Generally, water-bearing formations should be avoided, and the drilling direction should be directed away from them to reduce the risk of water intrusion or water entering the well. In contrast, directional drilling can be performed close to or near shale formations without increasing the risk of consequences from water-bearing formations. As an example, the method can use measurement results to perform inversion to improve geosteering, which can be used to distinguish between two layers, resulting in a geosteering wellbore with a reduced risk of water channeling. While shale is mentioned, as an example, the area can include sandstone, which can be a formation other than shale. For example, consider a shale and sandstone region.

[0114] As an example, a machine learning model can receive multiple data input channels. For instance, consider a machine learning model that receives measurements at three frequencies for each receiver, along with various measurement types, which may include one or more of symmetry, antisymmetry, harmonic resistivity, and harmonic anisotropy. Each of these measurement types may have attenuation and phase shift components, where the phase shift component can be represented in degrees. Therefore, as an example, the machine learning model could receive approximately 18 channels for each receiver. As an example, the machine learning model can also utilize measurements such as apparent resistivity from one or more instruments. As an example, for a two-receiver operation, the machine learning model could include approximately 40 input data channels, while a three-receiver operation could include approximately 60 input data channels.

[0115] As mentioned above, a specific job can have a specific number of input channels. As an example, different jobs can use different tools (e.g., consider 1, 2, or 3 receivers). Furthermore, regarding the choice of frequency, each receiver can utilize one or more frequencies that are different from one or more frequencies of one or more other receivers.

[0116] As an example, a machine learning model can receive measurement results and generate outputs providing contrast between the reservoir and one or more adjacent layers (e.g., considering one or more reservoir boundaries). As an example, when navigating a toolchain within a wellbore in an area including a reservoir, the frame can be used to map one or more conductive layers above or below, or above and below; it should be noted that measurements and calculations can also map high-resistivity layers below or above. For example, during wellbore landing, when drilling into a shale layer above a high-resistivity reservoir, the high-resistivity reservoir below can be mapped. In such examples, higher contrast corresponds to a larger detection range.

[0117] As an example, the task may involve prioritizing the transmission of specific data. For instance, various types of telemetry may have associated latency and / or bandwidth characteristics. As an example, the task may include prioritizing real-time frame data to ensure that the frame receives measurements that provide acceptable resolution and latency. As an example, the frame may provide one or more parameter values ​​for telemetry such that relevant data points (e.g., measurement results) are included in the frame at a specific frequency (e.g., how often the frame is transmitted). Where the task may include equipment suitable for implementing wired telemetry (e.g., wired drill pipe (WDP)), latency may not be as critical a problem compared to telemetry relying on mud pulse techniques (e.g., encoded acoustic pulses transmitted via mud in the wellbore). As mentioned above, in the case of utilizing mud pulse telemetry, the measurement resolution can be considered to optimize the design frame, which may be associated with a trained machine learning model that can be used to invert measurement results (e.g., predictions of geological structures).

[0118] As an example, channels can be constructed based on EM measurements acquired by one or more tools and / or via one or more modeling techniques (e.g., synthetic data). As an example, a job can utilize one or more tools, which may have one or more different measurements defined for one or more purposes (e.g., different, the same, etc.). As an example, some measurements may be sensitive to distance to the boundary and contrast, others may be sensitive to tilt angle, still others may be sensitive to anisotropy, and yet others may be sensitive to resistivity. As an example, measurements may include shallow resistivity measurements, but their contribution may be less than that of one or more other types of data for various jobs. As an example, a framework can include different measurement types and different frequencies to plot layers (e.g., thickness and resistivity) of a region, and, for example, plot the tilt angle and anisotropy of one or more layers. As an example, for multiple channels from different measurement types, the framework can specify spacing and frequency in the mapping, where, for example, weights may be different.

[0119] As an example, machine learning models can include one or more types of components, architectures, etc. For instance, consider the ResNet model, where various models can be generated by changing the number of layers and the shape of the convolutional kernels, which can be determined through parameter fine-tuning. The ResNet model can be a deep learning model where weight layers refer to layer inputs to learn a residual function.

[0120] As an example, a machine learning model can include one or more components of the TransformerNet architecture. For instance, consider an architecture that includes multiple layers of CNNs added before features are fed into the Transformer structure to amplify those features. As mentioned above, the Transformer architecture can include multi-head mechanisms; it should be noted that the Transformer architecture can exist in a variety of natural language processing applications (e.g., consider GPT, etc.). As an example, a machine learning model can be constructed to receive input in image-like forms (e.g., pixel images), where, for example, to improve performance, the architecture can include additional layers to extract 2D information before implementing the Transformer. As an example, a machine learning model can be defined by the number of layers, the number of attention heads, and the model size, which can also be included in parameter fine-tuning.

[0121] As an example, a machine learning model may include one or more components of a Visual Transformer (ViT) architecture. As another example, a ViT can be tuned by the number of layers and the number of attention heads, for example, where one or more of these parameters can be fine-tuned.

[0122] As an example, the ViT architecture could be a Transformer architecture designed for computer vision or otherwise receiving input that can be represented in pixel form (e.g., two-dimensional images). As another example, the ViT architecture could provide components for decomposing an input image (e.g., a multidimensional array) into a series of image patches, which could then be tokenized before being applied to the Transformer architecture. In the ViT architecture, the attention mechanism could repeatedly transform the representation vectors of the image patches, thereby incorporating an increasing number of relationships between the image patches in the input array. This can be viewed in some way as analogous to the situation in natural language processing where, as representation vectors flow through a Transformer, they incorporate an increasing number of semantic relationships between words, moving from syntax to semantics.

[0123] As an example, a machine learning model can be a classifier. For instance, consider a ViT that can convert images (e.g., arrays) into a series of vector representations, where one or more network modules can be added to use these vector representations for downstream applications. For example, to use it for classification, a shallow MLP can be added on top of it, which outputs a probability distribution for each class.

[0124] As an example, a model can be defined using Y to represent the input and X to represent the output, where Y equals H(X), and the model can be H... -1The proxy model. For example, Y may include a real-time channel (e.g., a well logging curve), which, as mentioned above, may or may not be windowed. X, for example, may be a two-dimensional data structure such as a square of fixed size, representing a resistivity profile of a region in space. Given the equation Y = H(X), the iterative inversion technique aims to solve for X from Y. As mentioned above, in the case of utilizing a machine learning model, this machine learning model may be H... -1 The proxy model makes X=H -1 (Y). Therefore, a machine learning model can receive input Y and provide X as output. In this way, X can be considered as a prediction made by the machine learning model for a given input Y.

[0125] Figure 11 An example of workflow 1100 is shown, which includes: a selection block 1114 for selecting an oilfield; an acquisition block 1118 for acquiring data; and a generation block 1122 for generating a structure, wherein the data and the structure can be used to create a dataset block 1126 (which includes the dataset). As described above, this approach allows the generation of the structure from the acquired data using iterative inversion. It should be noted that one or more techniques, such as data augmentation, can be employed. As an example, for a given structure, forward modeling techniques can be used to generate synthetic measurement data. As an example, one or more aspects of the operation can be included in the workflow for generating one or more datasets. For example, the type of tool, the type of formation, etc., can be considered.

[0126] like Figure 11 As shown, it may include a provision block 1130 for providing one or more machine learning models, a training block 1134 for training the one or more machine learning models, and an output block 1138 for outputting the one or more machine learning models.

[0127] like Figure 11 As shown, it may include: a data acquisition block 1142 for acquiring data using one or more downhole tools on a toolchain, the toolchain being at least partially located within a wellbore in the region; a generation block 1146 for generating outputs using one or more of the one or more machine learning models; and a control block 1150 for controlling drilling using the toolchain, wherein the toolchain includes a drill bit that can be oriented by one or more geosteering techniques.

[0128] like Figure 11As shown, workflow 1100 can be used for continuous learning, retraining, etc. For example, the data collected from acquisition block 1142 and the output generated by generation block 1146 can be used to form one or more datasets, which can be provided to dataset block 1126 for continuous learning, retraining, etc. on one or more machine learning models; it should be noted that one or more of blocks 1114, 1118 and 1122 can provide additional data (e.g., for a selected oil field, another oil field, etc.).

[0129] Figure 12 Various examples of environments 1201, 1203, and 1205 that can host frameworks 1210, 1220, and 1230 are shown. For example, environment 1201 may be a cloud platform environment that can host framework 1210, which can be used to generate and / or implement one or more machine learning models; environment 1203 may be a drilling rig site environment that can host framework 1220, which can be used to generate and / or implement one or more machine learning models; and environment 1205 may be a toolchain environment that can host framework 1230, which can be used to generate and / or implement one or more machine learning models. As an example, devices at drilling rig site environment 1203 can be operatively coupled to devices in cloud platform environment 1201 and / or toolchain environment 1205.

[0130] As an example, a toolchain may include an embedded framework for downhole automation control of one or more operations, such as geological guidance. As an example, a drilling rig control system (RCS) may include an embedded framework for controlling one or more operations, such as geological guidance. In such examples, one or more levels of automation may be implemented such that the framework forms part of a control loop, which may be a closed-loop control loop and / or a human-in-the-loop (HITL) type control loop. As an example, a cloud platform may be used for one or more purposes, such as model building, model updating, data access, synthetic data generation, etc. As an example, when updating a model, the updated model may be provided via one or more environments for field implementation, such as in the drilling rig field environment and / or in the toolchain environment.

[0131] Regarding the types of machine learning models, consider one or more of the following: Support Vector Machine (SVM) models, k-Nearest Neighbors (KNN) models, ensemble classifier models, neural network (NN) models, etc. As examples, machine learning models can be deep learning models (e.g., deep Boltzmann machines, deep belief networks, convolutional neural networks, stacked autoencoders, etc.), ensemble models (e.g., random forests, gradient boosting machines, bootstrap aggregation, AdaBoost, stacked generalization, gradient boosting regression trees, etc.), neural network models (e.g., radial basis function networks, perceptrons, backpropagation, Hopfield networks, etc.), regularization models (e.g., ridge regression, minimum absolute shrinkage and selection operators, elastic networks, minimum angle regression), rule system models (e.g., Cubist, single-rule, zero-rule, repeated incremental pruning for error reduction), and regression models (e.g., linear regression, ordinary least squares regression, stepwise regression, multivariate adaptive regression splines, local estimation scatter smoothing, logistic regression, etc.). Bayesian models (e.g., Naive Bayes, Average Independent Dependency Estimator, Bayesian Belief Network, Gaussian Naive Bayes, Multinomial Naive Bayes, Bayesian Networks), decision tree models (e.g., classification and regression trees, Iterative Dichotomy 3, C4.5, C5.0, Chi-square Automatic Interaction Detection, Decision Stumps, Conditional Decision Trees, M5), dimensionality reduction models (e.g., Principal Component Analysis, Partial Least Squares Regression, Sammon Mapping, Multidimensional Scaling Analysis, Projective Pursuit, Principal Component Regression, Partial Least Squares Discriminant Analysis, Mixed Discriminant Analysis, Quadratic Discriminant Analysis, Regularized Discriminant Analysis, Flexible Discriminant Analysis, Linear Discriminant Analysis, etc.), instance models (e.g., k-Nearest Neighbors, Learned Vector Quantization, Self-Organizing Map, Locally Weighted Learning, etc.), clustering models (e.g., k-means, k-median, Expectation Maximization, Hierarchical Clustering, etc.), etc.

[0132] As an example, a machine learning model can be a machine learning model (ML model) and can be built using computational frameworks with libraries, toolboxes, etc., such as the libraries and toolboxes of the MATLAB framework (MathWorks, Inc., Natick, Massachusetts). The MATLAB framework includes toolboxes that provide supervised and unsupervised machine learning algorithms, including Support Vector Machines (SVM), boosting decision trees and bagged decision trees, k-nearest neighbors (KNN), k-means, k-centroids, hierarchical clustering, Gaussian mixture models, and hidden Markov models. Another MATLAB framework toolbox is the Deep Learning Toolbox (DLT), which provides a framework for designing and implementing deep neural networks with algorithms, pre-trained models, and applications. DLT provides convolutional neural networks (ConvNet, CNN) and long short-term memory (LSTM) networks for performing classification and regression on image, time series, and text data. DLT includes features for building network architectures such as generative adversarial networks (GANs) and Siamese networks using custom training loops, shared weights, and automatic differentiation. DLT provides model exchange with various other frameworks.

[0133] As an example, the TENSORFLOW framework (Google LLC, Mountain View, CA) can be implemented. This framework is an open-source software library for dataflow programming that includes a symbolic mathematics library that can be implemented for machine learning applications that may include neural networks. As an example, the CAFFE framework can be implemented. This framework is a DL framework developed by Berkeley AI Research (BAIR) (University of California, Berkeley, California). As another example, consider utilizing the SCIKIT platform (e.g., scikit-learn) of the Python programming language. As an example, frameworks such as the APOLLO AI framework (APOLLO.AI GmbH, Germany) can be utilized. As an example, frameworks such as the PYTORCH framework (Facebook AI Research Lab (FAIR), Facebook, Inc., MenloPark, California) can be utilized.

[0134] As an example, training methods may include various actions that can manipulate the dataset to train an ML model. As another example, the dataset may be split into training and testing data, where the testing data can be used for evaluation. Methods may include cross-validation of parameters and optimal parameters, which can be provided for model training.

[0135] The TENSORFLOW framework can run on multiple CPUs and GPUs (with optional CUDA (NVIDIA Corp., Santa Clara, California) and SYCL (The Khronos Group Inc., Beaverton, Oregon) extensions for general-purpose computing on graphics processing units (GPUs). TENSORFLOW is available for 64-bit Linux, macOS (Apple Inc., Cupertino, California), Windows (Microsoft Corp., Redmond, Washington), and mobile computing platforms including those based on Android (Google LLC, Mountain View, California) and iOS (Apple Inc.) operating systems.

[0136] TENSORFLOW computation can be represented as a stateful data flow graph; it should be noted that the name TENSORFLOW originates from the operations performed by such neural networks on multidimensional data arrays. Such arrays can be called "tensors".

[0137] As an example, a device can utilize TENSORFLOW LITE (TFL) or another type of lightweight framework. For instance, consider a gateway that can be located in the field (e.g., a work site) and can utilize TFL and / or one or more other types of lightweight frameworks. The TFL framework is a set of tools that enables on-device machine learning, where models can run on mobile devices, embedded devices, and IoT devices. The TFL framework is optimized for on-device machine learning, addressing issues of latency (no round trips to the server), privacy (personal data does not leave the device), connectivity (internet connection required), size (reducing model and binary file size), and power consumption (e.g., efficient inference and lack of network connectivity). The TFL framework offers multi-platform support, covering Android and iOS devices, embedded Linux, and microcontrollers. The TFL framework offers multi-language support, including Java, SWIFT, Objective-C, C++, and Python. The TFL framework can deliver high performance through hardware acceleration and model optimization.

[0138] Figure 13An example of method 1300 is shown, which includes: a receiving block 1310 for receiving data acquired by downhole tools of a tool string at least partially located within a wellbore in a subsurface region; an inversion block 1320 for inverting the data using a trained machine learning model to generate structural features of the subsurface region; and a control block 1330 for controlling the operation of the tool string at least partially based on the structural features of the subsurface region. As shown, method 1300 may optionally include an update block 1340 for updating the trained machine learning model.

[0139] Figure 13 Method 1300 is shown to include various computer-readable storage medium (CRM) blocks 1311, 1321, 1331 and 1341, which may include processor-executable instructions that may instruct a computing system for controlling the system to perform one or more actions described with respect to method 1300.

[0140] like Figure 13 As illustrated in the example, system 1390 may include one or more computers 1392, each computer 1392 including one or more processors 1393, a memory 1394 operatively coupled to at least one of the processors 1393, instructions 1396 that may be stored in the memory 1394, and one or more interfaces 1395 (e.g., one or more network interfaces and / or other interfaces). As an example, system 1390 may include one or more processor-readable media including processor-executable instructions executable by at least one of the processors 1393 to cause system 1390 to perform actions such as one or more actions of method 1300. As an example, instructions 1396 may include instructions of one or more CRM blocks 1311, 1321, 1331, and 1341. Memory 1394 may be or include the one or more processor-readable media, wherein the processor-executable instructions may be or include instructions. As an example, the processor-readable media may be a non-transitory computer-readable storage medium that is not a signal and is not a carrier wave.

[0141] As an example, system 1390 may include subsystems. For instance, system 1390 may include multiple subsystems that can operate using distributed devices, where each subsystem may be referred to as a system. Consider, for example, a downhole tool system and a surface system. As an example, a downhole tool system may be used to perform the operations of blocks 1310, 1320, 1330, and 1340 of method 1300. Method 1300 may be implemented using, for example, a downhole system and / or a surface system, which may be a cloud-based system or a cloud-connected system.

[0142] As an example, the method may include: receiving data acquired by a downhole tool from a tool string at least partially located within a wellbore in a subsurface region; inverting the data using a trained machine learning model to generate structural features of the subsurface region; and controlling the operation of the tool string at least in part based on the structural features of the subsurface region. In such an example, the data may include electromagnetic data (e.g., attenuation, phase, etc.), wherein the structural features may be structural features in a resistivity profile of at least a portion of the subsurface region. In such an example, the electromagnetic data may include electromagnetic data channels for multiple receivers of the downhole tool and / or may include electromagnetic data channels for multiple frequencies.

[0143] As an example, a structural feature may be or include a boundary between a reservoir and another layer in a subsurface region. In such an example, the method may include controlling the operation of the tool string, for example, by geologically guiding the drill bit of the tool string to further drill into the subsurface region. In such an example, geological guidance may turn the drill bit away from the boundary and toward the reservoir.

[0144] As an example, a trained machine learning model can be or includes a deep learning model.

[0145] As an example, a method may include training a machine learning model to generate a trained machine learning model. In such an example, training may include: accessing well logging data from a database; iteratively inverting the well logging data to generate one or more structural models; and associating the well logging data and the one or more structural models as a training set for training the machine learning model.

[0146] As an example, the method may include updating a trained machine learning model with a revised trained machine learning model. In such an example, the update may include: iteratively inverting data to generate one or more structured models, wherein the data and the one or more structured models form a training set; and further training the trained machine learning model using the training set to generate the revised trained machine learning model. In such an example, the method may include performing the iterative inversion and the further training as background processes. As an example, the method may include performing at least a portion of such background processes using a device on a cloud platform.

[0147] As an example, inversion techniques based on machine learning models can occur in less than 10 seconds. In such an example, the method may include control occurring within 60 seconds of receiving data. For example, consider a control loop that receives data, performs an inversion using at least a portion of the data and a trained machine learning model, and controls drilling at least partially based on the inversion for a period of several minutes or less (e.g., less than one minute). In such an example, the control loop can be considered a real-time control loop that can be used for real-time control (e.g., real-time geological steering control).

[0148] As an example, a toolchain may include circuitry in which a trained machine learning model may be embedded for performing inversions within the toolchain. For instance, consider a toolchain capable of performing inversions and controlling one or more components of the toolchain, at least in part, based on the inversions. As an example, a toolchain may include components adapted to implement control loops that can be real-time control loops.

[0149] As an example, the method may include receiving data via mud pulse telemetry by the drilling rig control system. For instance, consider a downhole tool that can generate data using one or more sensors, where the downhole telemetry device can transmit the data (e.g., raw, compressed, encoded, etc.) to a surface telemetry receiver using mud pulse telemetry. In such an example, mud pulse telemetry may introduce a certain amount of time delay, which may depend, for example, on the depth of the BHA in the wellbore.

[0150] As an example, the system may include a processor; processor-accessible memory; and processor-executable instructions stored in the memory and executable by the processor to instruct the system to perform the following operations: receive data acquired by downhole tools of a tool string at least partially located within a wellbore in a subsurface area; invert the data using a trained machine learning model to generate structural features of the subsurface area; and control the operation of the tool string at least in part based on the structural features of the subsurface area.

[0151] As an example, one or more non-transitory computer-readable storage media may include processor-executable instructions that can be executed to instruct a processor to perform the following actions: receive data acquired by downhole tools of a tool string located at least partially within a wellbore in a subsurface area; invert the data using a trained machine learning model to generate structural features of the subsurface area; and control the operation of the tool string at least in part based on the structural features of the subsurface area.

[0152] As an example, one or more computer-readable storage media may include processor-executable instructions for instructing a computing system to perform one or more methods. In such an example, the one or more computer-readable storage media may be a program product (e.g., a computer program product, a computer system program product, etc.).

[0153] In some embodiments, one or more methods may be executed by a computing system. Figure 14 An example of system 1400 is shown, which may include one or more computing systems 1401-1, 1401-2, 1401-3 and 1401-4, which may be operatively connected via one or more networks 1409, which may include wired networks and / or wireless networks.

[0154] As an example, the system may include a standalone computer system or an arrangement of distributed computer systems. Figure 14 In the example, computer system 1401-1 may include one or more instruction sets 1402, which may be or include processor-executable instructions, for example, executable to perform various tasks (e.g., receiving information, requesting information, processing information, simulating, outputting information, etc.).

[0155] As an example, the instruction set can be executed independently or in conjunction with one or more processors 1404 operatively connected to one or more storage media 1406 (e.g., via wired, wireless, etc.). As an example, one or more of the processors 1404 can be operatively connected to at least one of one or more network interfaces 1407. In such an example, the computer system 1401-1 can transmit and / or receive information, for example, via one or more networks 1409 (e.g., one or more of the Internet, private networks, cellular networks, satellite networks, etc.). As shown, one or more other components 1408 may be included.

[0156] As an example, computer system 1401-1 may receive information from and / or transmit information to one or more other devices, such as computer system 1401-2, etc. The devices may be located in a different physical location than the physical location of computer system 1401-1. For example, the location may be a processing facility location, a data center location (e.g., a server cluster), a drilling rig location, a well site location, a downhole location, etc.

[0157] As an example, a processor may be or include a microprocessor, a microcontroller, a processor component or subsystem, a programmable integrated circuit, a programmable gate array, or another control device or computing device.

[0158] As an example, storage medium 1406 may be implemented as one or more computer-readable or machine-readable storage media. As an example, storage devices may be distributed within and / or across multiple internal and / or external housings of a computing system and / or an auxiliary computing system.

[0159] As an example, one or more storage media may include one or more different forms of memory, including semiconductor memory devices such as dynamic random access memory or static random access memory (DRAM or SRAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), and flash memory; magnetic disks such as fixed disks, floppy disks, and removable disks; other magnetic media including magnetic tape; optical media such as optical discs (CDs) or digital video discs (DVDs), Blu-ray discs, or other types of optical storage devices; or other types of storage devices.

[0160] As an example, one or more storage media may be located in a machine that runs machine-readable instructions, or at a remote site from which machine-readable instructions can be downloaded via a network for execution.

[0161] As an example, various components of a system (such as, for example, a computer system) may be implemented in hardware, software, or a combination of both (e.g., including firmware), including one or more signal processing circuits and / or application-specific integrated circuits.

[0162] As an example, the system may include a processing device, which may be or include a general-purpose processor or a special-purpose chip (e.g., a chipset), such as an ASIC, FPGA, PLD, or other suitable device.

[0163] As an example, a device can be a mobile device that includes one or more network interfaces for information communication. For example, a mobile device may include a wireless network interface (e.g., capable of operating via IEEE 802.11, ETSI GSM, BLUETOOTH, satellite, etc.). As an example, a mobile device may include components such as a main processor, memory, display, display graphics circuitry (e.g., optionally including touch and gesture circuitry), SIM card slot, audio / video circuitry, motion processing circuitry (e.g., accelerometer, gyroscope), wireless LAN circuitry, smart card circuitry, transmitter circuitry, GPS circuitry, and a battery. As an example, a mobile device can be configured as a mobile phone, tablet computer, etc. As an example, a method can be implemented using a mobile device (e.g., wholly or partially). As an example, a system may include one or more mobile devices.

[0164] As an example, the system can be a distributed environment, such as a so-called "cloud" environment, in which various devices, components, etc., interact for purposes such as data storage, communication, and computing. As an example, a device or system may include one or more components for communicating information via one or more of the Internet (e.g., where communication occurs via one or more Internet protocols), cellular networks, satellite networks, etc. As an example, methods (e.g., wholly or partially as cloud-based services) can be implemented in a distributed environment.

[0165] As an example, information can be input from a display (e.g., consider a touchscreen), output to a display, or both. As an example, information can be output to a projector, laser device, printer, etc., making the information viewable. As an example, information can be output in a stereoscopic or holographic manner. For the printer, consider a 2D printer or a 3D printer. As an example, a 3D printer can include one or more materials that can be output to construct 3D objects. For example, data can be provided to a 3D printer to construct a 3D representation of underground strata. As an example, layers (e.g., stratigraphic layers), geological bodies, etc., can be constructed in 3D. As an example, holes, cracks, etc., can be constructed in 3D (e.g., as positive structures, as negative structures, etc.).

[0166] Although only a few examples have been described in detail above, those skilled in the art will readily understand that many modifications can be made to these examples. Therefore, all such modifications are intended to be included within the scope of this disclosure as defined by the appended claims. In the claims, the device plus function clause is intended to cover structures described herein as performing said functions, and not only structural equivalents but also equivalent structures. Thus, although nails and screws may not be structural equivalents because nails use a cylindrical surface to hold wooden parts together while screws use a helical surface, in the context of fastening wooden parts, nails and screws can be equivalent structures.

Claims

1. A method (1300) comprising: Receive data acquired by downhole tools from a tool string located at least partially within a wellbore in the subsurface area (1310). The data is inverted using a trained machine learning model to generate structural features of the underground area (1330); and The operation of the toolchain is controlled at least in part based on the structural features of the underground region (1340).

2. The method according to claim 1, wherein, The data includes electromagnetic data, and wherein the structural features include structural features in the resistivity profile of at least a portion of the subsurface region, optionally wherein the electromagnetic data includes electromagnetic data channels for multiple receivers of the downhole tool and / or electromagnetic data channels for multiple frequencies.

3. The method according to claim 1 or 2, wherein, The structural features include a boundary between a reservoir and another layer in the subsurface region. Optionally, controlling the operation of the tool string includes geologically guiding the drill bit of the tool string to drill further into the subsurface region. Optionally, the geological guidance causes the drill bit to turn away from the boundary and toward the reservoir.

4. The method according to any one of the preceding claims, wherein, The trained machine learning model includes a deep learning model.

5. The method according to any one of the preceding claims, comprising training a machine learning model to generate the trained machine learning model.

6. The method according to any one of the preceding claims, wherein, The training includes: accessing well logging data from a database; iteratively inverting the well logging data to generate one or more structural models; and associating the well logging data and the one or more structural models as a training set for training the machine learning model.

7. The method according to any one of the preceding claims, comprising updating the trained machine learning model with a revised trained machine learning model, optionally wherein, The update includes: iteratively inverting the data to generate one or more structural models, wherein the data and the one or more structural models form a training set; and further training the trained machine learning model using the training set to generate the revised trained machine learning model.

8. The method according to any one of the preceding claims, comprising performing the iterative inversion and the further training as background processes.

9. The method according to any one of the preceding claims, comprising using a device of a cloud platform to perform at least a portion of the background process.

10. The method according to any one of the preceding claims, wherein, The inversion occurs in less than 10 seconds.

11. The method according to any one of the preceding claims, wherein, The control occurs within 60 seconds after the reception.

12. The method according to any one of the preceding claims, wherein, The toolchain includes circuitry, wherein the trained machine learning model is embedded in the circuitry for performing the inversion within the toolchain.

13. The method according to any one of the preceding claims, wherein, The receiving includes receiving the data via mud pulse telemetry by the drilling rig control system.

14. A system (1390) comprising: Processor (1393); The processor-accessible memory (1394). as well as Processor-executable instructions (1396), which are stored in the memory and can be executed by the processor, to instruct the system to perform the following operations: Receive data acquired by downhole tools from a tool string located at least partially within a wellbore in the subsurface area (1311). The data is inverted using a trained machine learning model to generate structural features of the underground area (1321); and The operation of the toolchain is controlled at least in part based on the structural features of the underground region (1331).

15. A computer program product comprising computer-executable instructions for instructing a computing system to perform the method according to any one of claims 1 to 13.

Images