Unsupervised well log reconstruction and outlier detection

Abstract

A method includes receiving well log data including a plurality of well logs; identifying one or more segments of one or more well logs in the plurality of well logs having substantially complete data; training a reconstruction neural network to reconstruct incomplete well logs based on the one or more segments of the one or more well logs having substantially complete data; and reconstructing one or more incomplete well logs in the plurality of well logs using the reconstruction neural network.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Patent Application Serial No. 62 / 897,088, filed on September 6, 2019, which is incorporated herein by reference in its entirety. Background Art

[0003] In the oil and gas industry, logging equipment is used to collect well logs during or after drilling to determine the characteristics of the formation around the well. The well logs may include gamma ray logs, resistivity logs, neutron logs, density logs, porosity logs, etc.

[0004] However, well logs may be incomplete, for example due to sensor failure and / or human error. Recently, machine learning models have been used to automatically reconstruct missing elements of well logs. Typically, the machine learning model is trained using a corpus of labeled, high-quality well logs from a specific region, which the machine learning model uses to establish a predictive link between the input, incomplete well logs and the output, or "reconstructed," well logs, in which the incomplete data is filled in.

[0005] However, there may be challenges in obtaining such labeled high-quality well log training data. For example, segments of high-quality well logs are manually determined by human operators (usually domain experts), which may be expensive and time-consuming. In addition, due to differences in subsurface geochemical properties, labeled high-quality well logs may not have universal applicability, that is, the labels of well logs from one oil field may not be useful for training a machine learning model to reconstruct well logs from another oil field. Therefore, in relatively new oil fields, there may not be any available labeled training data or there may not be a sufficient amount of available labeled training data. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate the present teachings and, together with the described embodiments, serve to explain the principles of the present teachings. In the drawings:

[0007] Figure 1A , Figure 1B , Figure 1C , Figure 1D , Figure 2 , Figure 3A and Figure 3B A simplified schematic diagram of an oil field and its operations is shown according to one embodiment.

[0008] Figure 4 A flow chart is shown of a workflow for well log reconstruction, for example using a neural network such as an autoencoder, according to one embodiment.

[0009] Figure 5 According to one embodiment, for example, a method for performing Figure 4 Schematic diagram of the workflow of a neural network.

[0010] Figure 6 A flow chart illustrating a workflow for well log reconstruction and outlier detection, for example using a neural network such as an autoencoder, according to one embodiment.

[0011] Figure 7 A reconstruction error histogram is shown according to one embodiment.

[0012] Figure 8 A schematic diagram of a computing system for performing one or more of the methods disclosed herein is shown, according to one embodiment. DETAILED DESCRIPTION

[0013] Reference will now be made in detail to the embodiments, examples of which are illustrated in the accompanying drawings. In the following detailed description, many specific details are set forth to provide a deeper understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without these specific details. In other cases, well-known methods, processes, components, circuits, and networks are not described in detail so as not to unnecessarily obscure aspects of the embodiments.

[0014] It should also be understood that although the terms first, second, etc. can be used to describe various elements in this article, these elements should not be limited to these terms. These terms are only used to distinguish one element from another element. For example, without departing from the scope of the present invention, a first object can be referred to as a second object, and similarly, a second object can be referred to as a first object. The first object and the second object are two objects respectively, but they should not be considered as the same object.

[0015] The terms used in the description of the present invention are only used for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, the singular forms "one", "an" and "the" are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any possible combination of one or more of the associated listed items. It should also be understood that the terms "include", "comprise" and / or "contain" when used in this specification provide for the presence of stated features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or their groups. In addition, as used herein, the term "if" can be interpreted as meaning "at the time of" or "after" or "in response to determining" or "in response to detecting", depending on the context.

[0016] Attention is now directed to processes, methods, techniques, and workflows according to some embodiments.Some operations in the processes, methods, techniques, and workflows disclosed herein may be combined and / or the order of some operations may be changed.

[0017] Figures 1A to 1D A simplified schematic diagram of an oil field 100 having a subsurface formation 102 containing a reservoir 104 therein is shown in accordance with implementations of various techniques and technologies described herein. Figure 1A A survey operation is shown being performed by a survey tool, such as a seismic vehicle 106.1, to measure properties of underground formations. The survey operation is a seismic survey operation for generating acoustic vibrations. Figure 1A In the embodiment of the present invention, one such acoustic vibration (e.g., acoustic vibration 112 generated by source 110) reflects off a horizon 114 in an earth formation 116. A set of acoustic vibrations is received by sensors located on the earth's surface, such as geophone-receivers 118. The received data 120 is provided as input data to a computer 122.1 of the seismic vehicle 106.1, and in response to the input data, the computer 122.1 generates a seismic data output 124. This seismic data output may be stored, transmitted, or further processed as desired, such as by data reduction.

[0018] Figure 1BThe drilling operation being performed by a drilling tool 106.2 is shown, which is suspended by a drilling rig 128 and advanced into a subterranean formation 102 to form a wellbore 136. A mud pit 130 is used to draw drilling mud into the drilling tool via a flow line 132 for circulating the drilling mud downward through the drilling tool and then upward through the wellbore 136 and back to the surface. The drilling mud is usually filtered and returned to the mud pit. A circulation system can be used to store, control or filter the flowing drilling mud. The drilling tool is advanced into the subterranean formation 102 to reach a reservoir 104. Each well can target one or more reservoirs. The drilling tool is suitable for measuring downhole properties using a logging while drilling tool. The logging while drilling tool can also be suitable for taking a core sample 133 as shown.

[0019] The computer facility may be located at various locations around the oil field 100 (e.g., surface unit 134) and / or at a remote location. The surface unit 134 may be used to communicate with the drilling tool and / or off-site operations, as well as other surface or downhole sensors. The surface unit 134 is capable of communicating with the drilling tool to send commands to the drilling tool and receive data from the drilling tool. The surface unit 134 may also collect data generated during the drilling operation and generate data output 135, which may then be stored or transmitted.

[0020] Sensors (S) (such as strain gauges) may be positioned around the oilfield 100 to collect data related to various oilfield operations as previously described. As shown, the sensors (S) are positioned in one or more locations in the drilling tool and / or at the drilling rig 128 to measure drilling parameters such as bit weight, drill bit torque, pressure, temperature, flow, composition, rotation speed, and / or other field operation parameters. The sensors (S) may also be positioned in one or more locations in the circulation system.

[0021] The drilling tool 106.2 may include a bottom hole assembly (BHA) (not shown), generally located close to the drill bit (e.g., within a number of drill collar lengths from the drill bit). The bottom hole assembly includes capabilities for measuring, processing, and storing information, as well as communicating with the surface unit 134. The bottom hole assembly also includes drill collars for performing various other measurement functions.

[0022] The bottom hole assembly may include a communication subassembly that communicates with the surface unit 134. The communication subassembly is adapted to send and receive signals to and from the surface using a communication channel such as mud pulse telemetry, electromagnetic telemetry, or wired drill pipe communication. The communication subassembly may include, for example, a transmitter that generates a signal (such as an acoustic or electromagnetic signal) representative of the measured drilling parameter. Those skilled in the art will appreciate that a variety of telemetry systems may be employed, such as wired drill pipe, electromagnetic, or other known telemetry systems.

[0023] Typically, a wellbore is drilled according to a drilling plan established prior to drilling. The drilling plan typically sets forth the equipment, pressures, trajectories, and / or other parameters of the drilling process that define the wellsite. Drilling operations may then be performed according to the drilling plan. However, as information is gathered, drilling operations may need to deviate from the drilling plan. Additionally, as drilling or other operations are performed, subsurface conditions may change. As new information is gathered, the earth model may also need to be adjusted.

[0024] The data collected by the sensor (S) may be collected by the surface unit 134 and / or other data collection sources for analysis or other processing. The data collected by the sensor (S) may be used alone or in combination with other data. The data may be collected in one or more databases and / or transmitted on-site or off-site. The data may be historical data, real-time data, or a combination thereof. Real-time data may be used in real time or may be stored for later use. The data may also be combined with historical data or other inputs for further analysis. The data may be stored in separate databases or may be combined into a single database.

[0025] The surface unit 134 may include a transceiver 137 to allow communication between the surface unit 134 and various parts or other locations of the oil field 100. The surface unit 134 may also be provided with one or more controllers (not shown) for actuating the mechanisms at the oil field 100 or functionally connected to the one or more controllers. The surface unit 134 may then send a command signal to the oil field 100 in response to the received data. The surface unit 134 may receive the command via the transceiver 137 or may execute the command to the controller itself. A processor may be provided to analyze the data (locally or remotely), make decisions and / or actuate the controller. In this way, the oil field 100 may be selectively adjusted based on the collected data. This technology can be used to optimize (or improve) various parts of the mine operation, such as controlling drilling, drilling pressure, pump rate or other parameters. These adjustments may be made automatically based on computer protocols and / or manually by an operator. In some cases, the well plan may be adjusted to select the best (or improved) operating conditions or to avoid problems.

[0026] Figure 1C Wireline operations are shown being performed by wireline tool 106.3, which is suspended from drilling rig 128 and enters the Figure 1B The wireline tool 106.3 is adapted to be deployed into the wellbore 136 for generating well logs, performing downhole tests and / or collecting samples. The wireline tool 106.3 may be used to provide another method and apparatus for performing seismic survey operations. The wireline tool 106.3 may, for example, have an explosive, radioactive, electrical or acoustic energy source 144 that sends and / or receives electrical signals to the surrounding underground formation 102 and the fluid therein.

[0027] The cable tool 106.3 may be operably connected to, for example, Figure 1A The wireline tool 106.3 may be positioned at different depths in the wellbore 136 to provide survey information or other information related to the subsurface formation 102.

[0028] Sensors (S) such as strain gauges may be positioned around the oilfield 100 to collect data related to various field operations as previously described. As shown, the sensors S are positioned in the wireline tool 106.3 to measure downhole parameters related to, for example, porosity, permeability, fluid composition, and / or other field operation parameters.

[0029] Figure 1D Production operations are shown being performed by a production tool 106.4 deployed from a production unit or Christmas tree 129 and into a completion wellbore 136 for pumping fluid from a downhole reservoir into a surface facility 142. The fluid passes from the reservoir 104 through perforations in the casing (not shown) and into the production tool 106.4 in the wellbore 136 and reaches the surface facility 142 via a gathering network 146.

[0030] Sensors (S) (such as strain gauges) may be positioned around the oilfield 100 to collect data related to various field operations as previously described. As shown, sensors (S) may be positioned in production tools 106.4 or associated equipment (such as Christmas trees 129, gathering pipe networks 146, surface facilities 142 and / or production facilities) to measure fluid parameters such as fluid composition, flow, pressure, temperature and / or other production operation parameters.

[0031] Production may also include injection wells for enhanced recovery.One or more gathering facilities may be operably connected to one or more of the well sites for selectively collecting downhole fluids from the one or more well sites.

[0032] Although Figures 1B to 1D Tools for measuring properties of an oil field are shown, but it will be appreciated that the tools may be used in conjunction with non-oil field operations such as gas fields, mines, aquifers, reservoirs, or other underground facilities. Additionally, while certain data acquisition tools are depicted, it will be appreciated that various measurement tools capable of sensing parameters of underground formations and / or geological formations thereof such as seismic two-way travel time, density, resistivity, yield, etc. may be used. Various sensors (S) may be located at various locations along the wellbore and / or monitoring tools to collect and / or monitor the desired data. Other data sources may also be provided from off-site locations.

[0033] Figures 1A to 1D The oilfield configuration of FIG. 1 is intended to provide a brief description of an example of an oilfield that can be used with the oilfield application framework. Part or all of the oilfield 100 can be located on land, on water, and / or offshore. In addition, although a single oilfield measured at a single location is depicted, the oilfield application can be utilized with any combination of one or more oilfields, one or more processing facilities, and one or more well sites.

[0034] Figure 2 A schematic diagram of a partial cross-section of an oil field 200 is shown having data acquisition tools 202.1, 202.2, 202.3, and 202.4 positioned at various locations along the oil field 200 for collecting data of a subsurface formation 204 in accordance with implementations of various techniques and technologies described herein. The data acquisition tools 202.1-202.4 may be respectively Figures 1A to 1D The data acquisition tools 106.1-106.4 or other data acquisition tools not depicted are the same. As shown, the data acquisition tools 202.1-202.4 generate data graphs or measurement values ​​208.1-208.4 respectively. These data graphs are depicted along the oil field 200 to show the data generated by various operations.

[0035] The data graphs 208.1-208.3 are examples of static data graphs that can be generated by the data acquisition tools 202.1-202.3, respectively; however, it should be understood that the data graphs 208.1-208.3 can also be data graphs that are updated in real time. These measurements can be analyzed to better define the properties of one or more formations and / or to determine the accuracy of the measurements and / or for error checking. The graphs of each of the corresponding measurements can be aligned and scaled to compare and verify the properties.

[0036] Static data graph 208.1 is a seismic bidirectional response over a period of time. Static graph 208.2 is core sample data measured from a core sample of formation 204. The core sample can be used to provide data such as a graph of density, porosity, permeability, or some other physical property of the core sample over the length of the core. Density and viscosity tests can be performed on the fluid in the core at different pressures and temperatures. Static data graph 208.3 is a well log trace that typically provides resistivity or other measurements of the formation at various depths.

[0037] The production decline curve or graph 208.4 is a dynamic data graph of fluid flow rate over time. The production decline curve generally provides production rate over time. As the fluid flows through the wellbore, fluid properties such as flow rate, pressure, composition, etc. are measured.

[0038] Other data may also be collected, such as historical data, user input, economic information, and / or other measurement data and other parameters of interest. As described below, static and dynamic measurements may be analyzed and used to generate a model of the subsurface formation to determine the characteristics of the subsurface formation. Similar measurements may also be used to measure changes in aspects of the formation over time.

[0039] The underground structure 204 has a plurality of geological formations 206.1-206.4. As shown, the structure has several formations or layers, including a shale layer 206.1, a carbonate layer 206.2, a shale layer 206.3, and a sand layer 206.4. A fault 207 extends through the shale layer 206.1 and the carbonate layer 206.2. The static data acquisition tool is suitable for taking measurements and detecting characteristics of the formations.

[0040] Although a particular underground formation with a particular geological structure is depicted, it will be appreciated that oilfield 200 may contain a variety of geological structures and / or formations, sometimes with great complexity. At some locations, typically below the waterline, fluids may occupy pore spaces in the formation. Each of the measurement devices may be used to measure the properties of the formation and / or its geological characteristics. Although each acquisition tool is shown in oilfield 200 as being at a particular location, it will be appreciated that one or more types of measurements may be made at one or more locations across one or more fields or other locations for comparison and / or analysis.

[0041] Then, data from, for example, Figure 2 The data collected by various sources of the data acquisition tools of the present invention are collected by the geophysicists. Typically, the seismic data displayed in the static data graph 208.1 from the data acquisition tool 202.1 is used by the geophysicists to determine the characteristics and features of the underground formations. Geologists typically use the core data shown in the static graph 208.2 and / or the logging data from the logging 208.3 to determine various characteristics of the underground formations. Reservoir engineers typically use the production data from the curve graph 208.4 to determine the fluid flow reservoir characteristics. Modeling techniques can be used to analyze the data analyzed by the geologists, geophysicists and reservoir engineers.

[0042] Figure 3A An oil field 300 is shown for performing production operations according to implementations of various techniques and technologies described herein. As shown, the oil field has a plurality of well sites 302 operatively connected to a central processing facility 354. Figure 3A The oilfield configuration is not intended to limit the scope of the oilfield application system. Part or all of the oilfield may be located on land and / or offshore. In addition, although a single oilfield with a single processing facility and multiple well pads is depicted, any combination of one or more oilfields, one or more processing facilities, and one or more well pads may exist.

[0043] Each well site 302 has equipment for forming a wellbore 336 in the earth. The wellbore extends through underground formations 306 including reservoirs 304. These reservoirs 304 contain fluids, such as hydrocarbons. The well site extracts fluids from the reservoirs and delivers them to processing facilities via a surface network 344. The surface network 344 has pipes and control mechanisms for controlling the flow of fluids from the well site to the processing facilities 354.

[0044] Now turn your attention to Figure 3B , which shows a side view of a marine-based survey 360 of a subsurface subsurface 362 according to one or more implementations of various techniques described herein. The subsurface 362 includes a shallow seafloor 364. A seismic source 366 may include a marine source, such as a vibrator or an air gun, which may propagate seismic waves 368 (e.g., energy signals) into the earth over an extended period of time or at nearly instantaneous energy provided by a pulsed source. The marine source may propagate the seismic waves as a frequency sweep signal. For example, a vibrator-type marine source may initially emit seismic waves at a low frequency (e.g., 5 Hz) and increase the seismic waves to a high frequency (e.g., 80 Hz-90 Hz) over time.

[0045] One or more components of the seismic waves 368 may be reflected and converted by the shallow seafloor 364 (i.e., reflectors), and the seismic wave reflections 370 may be received by a plurality of seismic receivers 372. The seismic receivers 372 may be disposed on a plurality of streamers (i.e., streamer array 374). The seismic receivers 372 may generate electrical signals representative of the received seismic wave reflections 370. The electrical signals may be embedded with information about the subsurface 362 and captured as a record of seismic data.

[0046] In one implementation, each streamer may include a streamer steering device not shown in this application, such as a bird, a deflector, a tail buoy, etc. The streamer steering device may be used to control the position of the streamers according to the techniques described herein.

[0047] In one implementation, seismic wave reflections 370 may travel upward and reach the water / air interface at the water surface 376, and a portion of the reflections 370 may then be reflected downward again (i.e., surface ghost waves 378) and received by a plurality of seismic receivers 372. Surface ghost waves 378 may be referred to as surface multiples. The point on the water surface 376 at which the wave is reflected downward is generally referred to as a downward reflection point.

[0048] The electrical signal may be transmitted to the ship 380 via a transmission cable, wireless communication, etc. The ship 380 may then transmit the electrical signal to a data processing center. Alternatively, the ship 380 may include an onboard computer capable of processing the electrical signal (i.e., seismic data). Those skilled in the art who have the benefit of this disclosure will appreciate that this illustration is highly idealized. For example, the survey may be for a formation deep below the surface. The formation may typically include a plurality of reflectors, some of which may include a drop event, and multiple reflections (including wave conversions) may be generated for reception by the seismic receiver 372. In one implementation, the seismic data may be processed to generate a seismic image of the subsurface 362.

[0049] The marine seismic acquisition system tows each streamer in the streamer array 374 at the same depth (e.g., 5-10 m). However, the ocean-based survey 360 may tow each streamer in the streamer array 374 at a different depth so that seismic data can be acquired and processed in a manner that avoids destructive interference effects due to sea surface ghost waves. For example, Figure 3B The ocean-based survey 360 shows eight streamers towed at eight different depths by a vessel 380. The depth of each streamer may be controlled and maintained using birds disposed on each streamer.

[0050] Embodiments of the present disclosure may provide systems and methods for automating the well log quality control analysis and well log reconstruction process, thereby facilitating the ability and efficiency of managing high-quality oilfield data. In some embodiments, the systems and methods may employ a workflow that utilizes the well logs being correlated and utilizes this correlation via one or more neural networks (e.g., autoencoders). Autoencoders are particularly useful in data denoising and reconstruction.

[0051] Figure 4 The illustrated workflow 400 may begin by acquiring well log data, which may include a plurality of well logs, as at 402. The well logs may be a priori unlabeled, or may be partially labeled, or may be fully labeled. In this context, "labeled" means that a human user has reviewed the well logs and indicated incomplete or erroneous segments of the data, which may provide a basis for supervised learning, as a machine learning model may make predictions based on the conclusions of the human user. However, because the workflow 400 may not rely on supervised learning, the workflow 400 may operate regardless of whether the well logs are labeled. This may avoid time-consuming and potentially non-repeatable (e.g., subjective, error-prone and mislabeled) human user interactions with raw well log data. The workflow 400 may proceed in two general stages: an outlier detection stage 403A and a well log reconstruction stage 403B.

[0052] In the outlier detection stage 403A, as at 403, segments of the well log for which the well log data is substantially complete are identified. By way of example, measurements may be taken between depths of 1000 feet and 1500 feet. Between depths of 1100 feet and 1105 feet, the density well log is filled with -999, otherwise indicating missing values. These may be incomplete segments; in contrast, those segments that are not filled with -999 may be identified as complete. A variety of other ways to identify missing well logs or segments thereof may also be used.

[0053] Thus, "complete" segments may be those with zero missing well logs. "Substantially complete" segments are those with fewer missing segments, such that the data is sufficient to employ these segments in further analysis. In some embodiments, the workflow 400 may identify those segments that are completely complete (no missing well logs), or those segments that have, for example, less than a threshold number of missing entries. For example, the well log data may include a combination of well logs (e.g., two or more types of well logs selected from the group of density, neutron, gamma, resistivity, and acoustic Δt (sound waves)). Although some types of well logs are considered more descriptive than other types of well logs (e.g., density well logs), for the same depth interval, a greater number of different types of well logs generally corresponds to a higher accuracy. Thus, if one or more of two types of well logs are missing for a certain depth interval, but a density well log exists, the segment may still be considered substantially complete because the input may still be used to achieve the result. However, at a certain number (e.g., three) of missing well logs, the missing well logs may be considered incomplete even if a density well log exists.

[0054] As at 404, these identified well log segments with substantially complete (or completely complete) data can be extracted, and as at 406, these identified well log segments are used to train an outlier detection neural network, such as a first or "outlier detection" autoencoder. Once trained, the outlier detection autoencoder can be configured to detect outliers in the remaining (substantially) complete well log data. Typically, an outlier is defined as a segment in a well log that deviates from adjacent segments to a certain degree, which will generally indicate a false or erroneous signal or data point. As will be described below, this determination can be made statistically. Outliers are removed from the data that can be used to train the second autoencoder (described below). Removing outliers can be used to increase the accuracy of the second autoencoder by avoiding training with abnormal or other non-representative training data. Therefore, the result of the outlier detection stage is an "estimated" well log that does not include incomplete well log segments and outlier well log segments.

[0055] Figure 5An example of a neural network / autoencoder is shown, and reference to it may help to more fully understand the outlier detection phase 403A and the well log reconstruction phase 403B. The neural network may have an input layer that may receive well logs (e.g., gamma rays, neutrons, resistivity, etc.). The trained neural network may then apply one or more series of layers in which the weights in the connections have been updated by back-propagation to minimize the prediction error. The result may be an output layer that may provide an estimate of the well log data.

[0056] Thus, the autoencoder can be trained to learn the internal correlations between the input well logs. Thus, a cost function for the autoencoder can be set that measures the difference between the estimated well logs and the original well logs. A backpropagation process is then used to minimize the cost function during the training process. Backpropagation may include calculating the gradient of the cost function and updating the weights of the autoencoder in a direction that reduces the cost function. After hundreds of generations of training, the cost function may reach a stable state and may not be reduced further.

[0057] In some embodiments, there is a risk that the autoencoder simply maps the input to the output layer without learning the internal associations. To avoid this, the bottleneck layer can be narrower than the input layer. By squeezing the bottleneck, the autoencoder breaks the same mapping relationship and is forced to learn internal associations. Using backpropagation and a narrower bottleneck layer, the autoencoder is trained to learn the associations between the well logs.

[0058] Return to Figure 4 , the workflow 400 may proceed to a well log reconstruction phase 403B. This phase 403B may be performed by a second autoencoder (also referred to herein as a "well log reconstruction" or "reconstruction" autoencoder). As at 408, the results of the outlier detection phase 403A, i.e., estimated well logs, such as input well logs from which incomplete segments and outlier segments have been extracted, may be used to train the second autoencoder. Next, the incomplete well logs received at 402 (and extracted from consideration during the initial outlier detection phase 403A) may be introduced into the trained second autoencoder and reconstructed. For example, individual well logs may be considered to include missing data and may be reconstructed by the second autoencoder.

[0059] Therefore, the autoencoder-based well log reconstruction workflow is an unsupervised modeling workflow that avoids the shortcomings inherent in supervised modeling discussed above. That is, for example, when reconstructing well logs, the workflow may not use labeled data. This workflow can be applied to a wide range of well log data sets and improve the efficiency of well log quality control analysis and well log reconstruction.

[0060] Figure 6 A flow chart of another workflow 600 for well log reconstruction according to one embodiment is shown. The workflow 600 may be a more detailed version of an embodiment of the workflow 400, and thus the two workflows 400, 600 should not be considered mutually exclusive.

[0061] Workflow 600 may include working steps 402, 403, 404 and 406, such as Figure 4 As shown, for example, as part of the outlier detection stage 403A. In addition, as at 602, the outlier stage 403A can include generating an estimated well log using the trained outlier detection autoencoder and the well log data. The estimated well log can be generated based on the original well log data (e.g., by the outlier detection autoencoder attempting to complete or replace the extracted incomplete segment).

[0062] Workflow 600 may then include determining an estimated error between the original well log and the estimated well log, such as by comparing the estimated data to the original data, as at 604. The estimated error may be stored as a histogram, such as Figure 7 As shown. Based on the estimated error, an outlier error threshold can be determined. In some embodiments, the outlier error threshold can be determined as the maximum value between a hard cutoff (e.g., static and / or predetermined) and a soft (e.g., dynamic) percentile value. In some embodiments, the hard cutoff value is an heuristic value from experience. The soft percentile value is determined by the error distribution in the well logging training data. For example, the soft percentile value can be a value corresponding to the 90th, 95th, or 98th percentile of the estimated error distribution. As at 606, the workflow 600 can then use the outlier error threshold to identify outlier segments, which are those segments that differ from the original segment by more than a threshold amount.

[0063] In some embodiments, an uncertainty analysis can be performed on the results of the outlier detection phase 403A, such as well logs that do not include incomplete segments and identified outliers. For example, a Monte Carlo dropout method can be implemented for uncertainty analysis. During Monte Carlo dropout, a random number of neural nodes can be deactivated, and then a model output sample can be generated. After multiple repetitions of Monte Carlo dropout, a series of output samples are generated. The confidence interval of the result can be calculated as [μ-2*σ, μ+2*σ], where μ is the mean and σ is the standard deviation.

[0064] Workflow 600 may then proceed to implementation / reconstruction phase 403B. Figure 4 Similar to stage 403B, as at 408, Figure 6This stage 403B of the process may include training a second reconstruction autoencoder to reconstruct the well log based on the results of the outlier detection stage 403A (e.g., the well log with the incomplete and outlier segments removed). Thereafter, as at 409, the second autoencoder may reconstruct the well log. In at least some embodiments, "reconstructing" the well log means predicting the information that would have been included in the incomplete segments of the well log if the well log were not incomplete, and then adding the predicted portions to the well log. In some embodiments, the results of the reconstruction stage 403B (i.e., the reconstructed well log) may be subjected to uncertainty analysis in either of the workflows 400 and / or 600. As discussed above, such uncertainty analysis may be performed using a Monte Carlo dropout technique.

[0065] Thus, embodiments of the present disclosure may provide an integrated, autoencoder-based well log data outlier detection workflow that may facilitate unsupervised training and implementation of a reconstruction autoencoder (or any other type of neural network). Thus, the present embodiment may, for example, provide data outlier detection and well log reconstruction simultaneously. The workflow may be applied to a wide range of well log data sets, and may improve the efficiency of well log quality control and well log reconstruction. As a result, the reconstructed well log may be displayed or otherwise transmitted (visualized) to a user. Well logs have many practical applications in many different petroleum technology workflows, such as determining location, technology, and otherwise informing hydrocarbon extraction processes, such as exploration, drilling, processing, intervention, production, etc. Thus, the present disclosure provides a more accurate reconstruction of well logs that may improve the effects of gaps in well log data. Furthermore, even at the same processing power, this improved workflow may be several orders of magnitude faster than previous attempts, thereby reducing processing time from days to minutes.

[0066] In some embodiments, any of the methods of the present disclosure may be performed by a computing system. Figure 8An example of such a computing system 800 is shown according to some embodiments. The computing system 800 may include a computer or computer system 801A, which may be a single computer system 801A or an arrangement of distributed computer systems. The computer system 801A includes one or more analysis modules 802, which are configured to perform various tasks according to some embodiments, such as one or more methods disclosed herein. To perform these various tasks, the analysis module 802 executes independently or in conjunction with one or more processors 804 connected to one or more storage media 806. Processor 804 is also connected to a network interface 809 to allow computer system 801A to communicate with one or more additional computer systems and / or computing systems such as 801B, 801C and / or 801D via data network 809 (it should be noted that computer systems 801B, 801C and / or 801D may or may not share the same architecture as computer system 801A and may be located in different physical locations, for example, computer systems 801A and 801B may be located in a processing facility while communicating with one or more computer systems such as 801C and / or 801D located in one or more data centers and / or in different countries on different continents).

[0067] A processor may include a microprocessor, a microcontroller, a processor module or subsystem, a programmable integrated circuit, a programmable gate array, or another control or computing device.

[0068] Storage medium 806 may be implemented as one or more computer-readable or machine-readable storage media. Figure 8 In the exemplary embodiment of the present invention, the storage medium 806 is depicted as being within the computer system 801A, but in some embodiments, the storage medium 806 may be distributed within and / or across multiple internal and / or external enclosures of the computing system 801A and / or another computing system. The storage medium 806 may include one or more different forms of memory, including semiconductor memory devices, such as dynamic or static random access memory (DRAM or SRAM), erasable 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 compact disks (CDs) or digital video disks (DVDs), A disk or other type of optical storage device, or other type of storage device. It should be noted that the instructions discussed above may be provided on one computer-readable or machine-readable storage medium, or alternatively, may be provided on multiple computer-readable or machine-readable storage media distributed in a larger system that may have multiple nodes. Such one or more computer-readable or machine-readable storage media are considered to be part of an article (or product). An article or product may refer to any manufactured single component or multiple components. One or more storage media may be located in a machine running machine-readable instructions or at a remote location where machine-readable instructions can be downloaded over a network for execution.

[0069] In some embodiments, computing system 800 includes one or more neural network modules 808. In the example of computing system 800, computer system 801A includes neural network module 808. In some embodiments, a neural network module can be used to perform some or all aspects of one or more embodiments of the method. In alternative embodiments, multiple neural network modules can be used to perform some or all aspects of the method.

[0070] It should be understood that computing system 800 is only one example of a computing system, and computing system 800 may have more or fewer components than shown, and may be combined. Figure 8 Additional components not depicted in the exemplary embodiment of the present invention and / or the computing system 800 may have Figure 8 Different configurations or arrangements of the components depicted in . Figure 8 The various components shown may be implemented in hardware, software, or a combination of hardware and software, including one or more signal processing and / or application specific integrated circuits.

[0071] In addition, the steps in the processing method described herein can be implemented by running one or more functional modules in an information processing device (such as a general-purpose processor or a dedicated chip (such as an ASIC, FPGA, PLD or other suitable device)). These modules, the combination of these modules and / or their combination with general-purpose hardware are all included in the protection scope of the present invention.

[0072] The geological interpretation, model and / or other interpretation aids may be refined in an iterative manner; this concept may be applicable to embodiments of the present method discussed herein. This may include the use of a feedback loop executed on an algorithmic basis, such as on a computing device (e.g., computing system 800, Figure 8 ) at the location, and / or through manual control of a user, the user can make a determination as to whether a given step, action, template, model, or set of curves has become sufficiently accurate for use in evaluating the three-dimensional subsurface geological formation under consideration.

[0073] For the purpose of explanation, the foregoing description has been described with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the invention to the precise form disclosed. In view of the above teachings, many modifications and variations are possible. In addition, the order of the elements of the method shown and described can be rearranged, and / or two or more elements can occur simultaneously. These embodiments are selected and described in order to best explain the principles of the invention and its practical application, so that other technical personnel in the art can best utilize the present invention and various embodiments with various modifications suitable for the specific purpose expected.

Claims

1. A method comprising: receiving well log data including a plurality of well logs; identifying a plurality of segments of one or more well logs in the plurality of well logs having substantially complete data; Using a first segment of the plurality of segments, training an outlier detection neural network to identify outliers in a remaining portion of the plurality of segments; applying the outlier detection neural network to the remaining portion of the plurality of segments to identify outlier segments; removing the outlier segment from the remaining portion of the plurality of segments to obtain a training portion of the plurality of segments; using the training portion, training a reconstruction neural network to reconstruct incomplete well logs based on the plurality of segments of the one or more well logs having substantially complete data; as well as One or more incomplete well logs in the plurality of well logs are reconstructed using the reconstruction neural network.

2. The method of claim 1 further comprises extracting one or more incomplete segments of the one or more incomplete well logging records in the plurality of well logging records before training the reconstruction neural network, so that the one or more incomplete segments are not used to train the reconstruction neural network.

3. The method of claim 1, wherein: The outlier segments are not used to train the reconstruction neural network.

4. The method of claim 2, wherein: Training an outlier detection neural network involves: generating one or more estimated well logs using the outlier detection neural network by estimating the one or more incomplete segments using the outlier detection neural network; and The outlier segments are identified at least in part by comparing the one or more estimated well logs to corresponding well logs in the plurality of well logs.

5. The method of claim 4, wherein identifying the outlier segment comprises determining an error threshold based on a comparison of the one or more estimated well logs with the corresponding well logs, wherein the error threshold comprises a hard cutoff, a soft cutoff, or a combination thereof, and wherein the outlier segment is identified based on a deviation from the corresponding well log exceeding the error threshold.

6. The method of claim 4, further comprising performing a Monte Carlo uncertainty analysis on the outlier detection neural network, the reconstruction neural network, or both.

7. The method of claim 1, wherein the training of the reconstruction neural network is unsupervised.

8. The method of claim 1, wherein the reconstructed neural network comprises an autoencoder.

9. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor of a computing system, cause the computing system to perform operations comprising: receiving well log data including a plurality of well logs; identifying a plurality of segments of one or more well logs in the plurality of well logs having substantially complete data; Using a first segment of the plurality of segments, training an outlier detection neural network to identify outliers in a remaining portion of the plurality of segments; applying the outlier detection neural network to the remaining portion of the plurality of segments to identify outlier segments; removing the outlier segment from the remaining portion of the plurality of segments to obtain a training portion of the plurality of segments; using the training portion, training a reconstruction neural network to reconstruct incomplete well logs based on the plurality of segments of the one or more well logs having substantially complete data; as well as One or more incomplete well logs in the plurality of well logs are reconstructed using the reconstruction neural network.

10. The medium of claim 9, wherein: The operations also include extracting one or more incomplete segments of the one or more incomplete well logs in the plurality of well logs before training the reconstruction neural network, so that the one or more incomplete segments are not used to train the reconstruction neural network.

11. The medium of claim 9, wherein: The outlier segments are not used to train the reconstruction neural network.

12. The medium of claim 10, wherein: Training an outlier detection neural network involves: generating one or more estimated well logs using the outlier detection neural network by estimating the one or more incomplete segments using the outlier detection neural network; and The outlier segments are identified at least in part by comparing the one or more estimated well logs to corresponding well logs in the plurality of well logs.

13. A medium as described in claim 12, wherein identifying the outlier segment includes determining an error threshold based on a comparison of the one or more estimated well log records with the corresponding well log records, wherein the error threshold includes a hard cutoff, a soft cutoff, or a combination thereof, and wherein the outlier segment is identified based on a deviation from the corresponding well log record exceeding the error threshold.

14. The medium of claim 12, wherein: The operations also include performing a Monte Carlo uncertainty analysis on the outlier detection neural network, the reconstruction neural network, or both.

15. The medium of claim 9, wherein the training of the reconstruction neural network is unsupervised, and wherein the reconstruction neural network comprises an autoencoder.

16. A computing system comprising: one or more processors; as well as a memory system comprising one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations comprising: receiving well log data including a plurality of well logs; identifying a plurality of segments of one or more well logs in the plurality of well logs having substantially complete data; Using a first segment of the plurality of segments, training an outlier detection neural network to identify outliers in a remaining portion of the plurality of segments; applying the outlier detection neural network to the remaining portion of the plurality of segments to identify outlier segments; removing the outlier segment from the remaining portion of the plurality of segments to obtain a training portion of the plurality of segments; Using the training portion, training a reconstruction neural network to reconstruct incomplete well logs based on the plurality of segments of the one or more well logs having substantially complete data; and One or more incomplete well logs in the plurality of well logs are reconstructed using the reconstruction neural network.

17. The computing system of claim 16, wherein the operation further comprises extracting one or more incomplete segments of the one or more incomplete well logs in the plurality of well logs before training the reconstruction neural network, so that the one or more incomplete segments are not used to train the reconstruction neural network.

18. The computing system of claim 16, wherein the outlier segments are not used to train the reconstruction neural network.

19. The computing system of claim 17, wherein: Training an outlier detection neural network involves: generating one or more estimated well logs using the outlier detection neural network by estimating the one or more incomplete segments using the outlier detection neural network; and The outlier segments are identified at least in part by comparing the one or more estimated well logs to corresponding well logs in the plurality of well logs.

20. The computing system of claim 19, wherein identifying the outlier segment comprises determining an error threshold based on a comparison of the one or more estimated well logs with the corresponding well logs, wherein the error threshold comprises a hard cutoff, a soft cutoff, or a combination thereof, and wherein the outlier segment is identified based on a deviation from the corresponding well log exceeding the error threshold.

Images