CLOSED LOOP FRACTURING MODELS AND OPERATIONS

Description

TECHNICAL FIELD

Some implementations relate to subsurface wells. More specifically, some implementations relate to modelling and controlling hydraulic-fracturing-related operations in the well.

BACKGROUND

Hydraulic fracturing (commonly known as “fracking”) is a well-stimulation technique used in the oil and gas industry to enhance the extraction of hydrocarbons from low-permeability rock formations (such as shale). The process may involve injecting a high-pressure mixture of water, sand, and chemicals into a wellbore to create small fractures in the rock. These fractures may significantly increase production rates by enabling oil and gas to flow more freely into the well. The sand (referred to as “proppant”) may create a continuous flow of hydrocarbons by keeping fractures open after the pressure is released.

Some fracking operations may involve multiple wells and various sensors that measure conditions in and around the wells. These measurements may be used by processes that model and control fracking operations in the wells.

BRIEF DESCRIPTION OF THE DRAWINGS

Implementations of the disclosure may be better understood by referencing the accompanying drawings.

FIG. 1 is a diagram illustrating operations for utilizing machine-learning models to achieve near well and far field uniformity for hydraulic fracturing.

FIG. 2 is a diagram that shows more operations for utilizing machine-learning models to achieve near well and far field uniformity for hydraulic fracturing.

FIG. 3 is a diagram illustrating an example artificial neural network included in some implementations.

FIG. 4 is a block diagram illustrating a computer system that may be utilized with some implementations.

FIG. 5 is an illustration depicting an example multi-well system, according to some implementations.

DESCRIPTION OF IMPLEMENTATIONS

The description that follows may include example systems, methods, techniques, and program flows that embody implementations of the disclosure. However, this disclosure may be practiced without these specific details. For clarity, some well-known instruction instances, protocols, structures, and techniques may not be shown in detail.

Overview

The technological development of both surface and downhole installed sensor-based systems has provided a significant uplift in how oil and gas operators monitor hydraulic fracture stimulations. These sensor systems may include, but are not limited to, pressure and temperature gauges, accelerometers, geophones, hydrophones, and distributed acoustic sensing acquisitions.

With the increase in sensor data recorded and stored for hydraulic fracture stimulations across North American unconventional basins, there may be a desire to utilize the derived answer products to optimize the completions in both a real-time and pseudo real-time manner. Some implementations develop machine-learning model workflows, in a regression and classification sense, and generate recommender systems that propose changes based on ML responses to well stimulation parameters at the surface. The learnings derived from the machine-learning models may enable focused extraction of physics-based fracture modeling insights to provide the recommender system with hybrid optimized stimulation changes to ensure the most consistent frac on a stage-by-stage basis.

Some implementations may relate to the far-field and near-field stimulation effectiveness in how consistently perforations are eroding as the hydraulic fracture stimulation progresses in time and/or how the hydraulic fracture propagation responds correspondingly. Some implementations facilitate consistency, where they seek to stimulate and erode every perforation equally over the course of the stimulation and to achieve a uniformly propagating fracture corridor from the fracturing stage. Some implementations aim to utilize answer products as targets to generate models which provide understanding of key features in the data which lead to inconsistencies in the fractures. Inconsistencies may be defined as answer product responses which do not indicate the hydraulic fracturing completion was uniformly distributed to the subsurface formation. Supervised learning models may extract inconsistencies related to temporal patterns of one or more surface pumping parameters. After inconsistencies are processes, stages with similar inconsistencies can be clustered and forward models of fracture propagation can be built which will provide the physical relationship between any changes which will mitigate the inconsistencies provided by the machine-learning model. The forward models can be validated by adjacent consistently completed stages with uplift in answer product response, which will allow for a hybrid data-physics recommender system to be generated. The recommender system will be a database (also referred to herein as a “digital twin”) for possible remediation strategies based on the input feature space.

Some implementations address the need for a closed-loop fracturing system that utilizes sensor data recorded in real-time/pseudo-real-time to optimize hydraulic fracturing operations. In some implementations, a closed-loop fracturing system may analyze surface and downhole sensor data to verify uniformity of ongoing completions in both the near and far field and recommend changes to the treating parameters based on the temporary variant sensor responses. Some implementations generate both an exploratory machine learning model and a hybrid physics and data-driven model to identify key features within the hydraulic fracturing process that exhibit nonuniformity. The plurality of models may be used to implement remediation strategies in real-time. The exploratory machine learning models may identify dominant fracturing feature data that are temporally related to the hydraulic fracture stimulation and that are directly related to either/both near and far field fracturing uniformity or nonuniformity. The quantification of temporal occurrences of nonuniformity may enable some implementations to generate and deploy real-time mitigation strategies via a digital twin. These mitigation strategies may themselves be developed by hybrid physics and machine learning modeling.

Some implementations provide technical improvements over traditional methods by reducing large-scale physics models which may be complex and may not fully represent the observed physical processes demonstrated by surface and downhole sensor data. By leveraging large data collected from a variety of sensors in both near and far field, some implementations generate machine learning models to infer key information which drives variability in the sensor data around fracture growth uniformity and how changes of the well completion might be implemented such that the uniformity is optimized. These sets of machine learning models may be included in a feedback loop that updates fracture importance and remediation strategies as more data is collected.

Example Implementations

Some implementations automate operations for optimizing hydraulic fracturing processes which lead to near wellbore and far field uniformity by employing a cascading workflow of trained machine learning models. With the extraction of defining temporal features, some implementations isolate inefficiencies and explore mitigation strategies via fracture modelling processes. After the mitigation strategies are defined, a separate classification machine learning model may be created to probabilistically identify the inefficiencies in real-time and call a digital twin model of fracture modelling results to enable operations that adjust well treatments. This workflow may be refined and updated as more data becomes available and as more field trials are performed using the mitigation techniques defined by the feature modelling to further optimize the process.

Some implementations determine useful features. The extraction of features which are critical to inefficiencies in the hydraulic fracturing process may be determined via a supervised learning process. The supervised learning process may contain several time variant and global features related to surface pumping parameters, which may include one or more of spread pressure, rate, proppant concentration, elapsed stage timing, perforation diameter, and total number of perforations. In addition to the perforation and pumping time series data, a variety of engineered features may be constructed from the pumping data to aid in model convergence and highlight statistical features of interest in time. The engineered features may include one or more of rolling metrics of mean/variance, autocorrelation, skewness, derivatives, seasonal decomposition, lag features, Fourier/wavelet information, approximate entropy, and sliding window amplitude ratios.

Some implementations determine one or more regression targets. The regression targets to be used relate to one or more metrics derived from an answer product (such a stage efficiency index derived from a pressure pulse related answer product the uniformity index derived from a digital acoustic sensing tool acquisition (such as StimWatch DAS acquisition), fracture propagation velocities from offset well low frequency DAS, and inverted fracture corridor width uniformity. Multiple exploratory models may require generation for each answer product, as both global and localized effects must be understood, thus regression targets may be treated in both an absolute sense and relative sense. The near-field and far-field answer products may operate on varying time scales (such as time durations, frequency content, and others). Thus, the engineered features may adequately capture this variance.

Some implementations create one or more supervised machine-learning models. By implementing one or more supervised machine-learning models which cater to time series data (such as a transformer or convolutional neural network), the models may emphasize features which best explain the target during a cost minimization.

After the machine-learning models have converged to an acceptable residual between target and prediction, a variety of techniques can be used in conjunction with one another to pull the feature class and the temporal relation that best determines the target value. Techniques such as a Shapley analysis and the gradient class maps from convolutional neural networks may isolate features and times of interest across each machine-learning model. The features and times of interest may then be weighted and grouped together via clustering with binned feature groups. These grouped features may be utilized with fracture modelling to explain the inefficiencies in the fracture stimulation and the potential remediation efforts.

After clustering stages which display similar far field and near field answer products, where each individual stage does not need to have acquired both near and far field metrics, some implementations may optionally perform forward fracture modelling to match the expected answer product responses. In a hybrid data-physics approach, some implementations utilize the clusters of stages which show an incremental uplift in the answer product response to perturb the fracture model and observe the improvement in the dominant temporal pumping features. As a result, some implementations generate a database of inverted remediation strategies which may be applied in real-time, and this database can be transitioned to a digital twin for optimized utilization in a field setting. Some of the remediation strategies may include configuring equipment, moving equipment at the surface or in the subsurface, operations involving physically altering an aspect of the well or well environment, and/or other operations.

FIG. 1 is a diagram illustrating operations for utilizing machine-learning models to achieve near well and far field uniformity for hydraulic fracturing. At block 102, some implementations may generate temporally variant (and/or non-temporally variant) engineered features from any suitable data source such as surface pumping spread data for each stage. The surface pumping spread data may include one or more of pressure, slurry rate, proppant concentration of various diameters, and others. Temporally variant engineered features may include one or more of the following with respect to any of the engineered features described herein: rolling statistical measures, derivatives, cumulative summations, lag features, Fourier information, wavelet information, entropy, and amplitude ratios. Some implementations may generate data from which the engineered features are generated and/or by which any one or more of the models described herein are trained. In some implementations, such data may be generated by only physics-based models, only by AI models, and/or by hybrid AI and physics-based models. In some implementations, such data is only field-acquired data such as surface pumping spread data.

At block 104, some implementations may define machine learning model architectures which track and account for temporal changes in model parameter updates (transformers, LSTM, and others).

At block 106, some implementations may perform regression analysis with engineered features as model inputs and answer products as targets. Some implementations may generate a variety of models for near field and far field measurements in both an absolute and relative sense. The one or models may be segregated by geologic formation. Some implementations may compute both epistemic and aleatoric uncertainty with each model (probabilistic).

At block 106, some implementations may extract feature importance metrics for each trained model with respect to both feature and time (attention weights, Shapley, gradient class activation maps, and others). Some implementations may extract answer products related to in situ and surface sensor measurements to be utilized as targets. These answer products may include global and/or time variant metrics (Pressure/temperature gauge, fiber optic, proxy sensor measurement) (see block 110).

At block 108, some implementations may extract feature importance metrics for each trained model with respect to both feature and time (attention weights, Shapley, gradient class activation maps, and others). The flow 100 continues in FIG. 2 at block 112.

FIG. 2 is a diagram that shows more operations for utilizing machine-learning models to achieve near well and far field uniformity for hydraulic fracturing. At block 112, some implementations may cluster stages that exhibit similar far field and near field answer product responses (individual states need not have acquired both near and far field metrics). Additionally, some implementations may define one or more cost functions to generate metrics of weighted feature importances (computed probabilities and temporally extracted feature importance) to be utilized in forward modeling.

At block 114, some implementations may perform fracture modelling on stages (to a varying degree of physics) from each cluster to match the expected answer product responses with associated uncertainty.

At block 116, some implementations may utilize the clustered stages which display percentage uplift in answer product response (spread data measurements, intervention from operator, and more) to perturb fracture model to observe improvement in dominant temporal pumping features.

At block 118, some implementations may generate a database of clustered features and inverted remediation strategies and develop a digital twin for optimized utilization of field settings.

Example Computer Systems

Some implementations include machine-learning models. Machine-learning modelling may include machine-implemented algorithms that learn from and make predictions (or decisions) based on input data. The input data may be collected from a variety of sources. The input data may include time series, image, text, and more. The input data may be cleaned via a suite of signal processing routines before engineered features are extracted and tailored to the problem at hand.

In some implementations, the machine-learning models may be trained with input data that spans the entire solution space, thereby creating a generalized model that produces accurate results based on input data that was not used in training. Input data used for training machine-learning models may include field data, synthetic data, artificial intelligence (AI) generated field data, and more. Field Data may include data that was recorded by sensors of any suitable type, whether they be installed at the surface or subsurface. The field data may be labeled in terms of the current time state (stage number, well name, etc.) or may be unlabeled depending on whether supervised or unsupervised learning is applied, respectively. The field data may either be raw sensor data or have a variety of signal processing applied prior to any feature engineering processes or machine learning model training.

Synthetic Data may be generated by physics-based models that produce pseudo sensor data to be used in machine-learning model training. This data may be utilized in generating a suite of physical scenarios, where the machine-learning modelling may be targeted at speeding up the inversion process (digital twin development). AI-generated field data may have been developed using means such as generative AI and other similar techniques. Recorded sensor field data may be input into one or more generative AI models whose output is synthetically derived field data (also referred to as AI-generated field data). The synthetically derived field data may be built in massive quantities and used to train subsequent machine learning models.

AI generated hybrid field-synthetic data may be developed using means such as generative AI and other similar techniques. Recorded sensor field data may include inputs into one or more generative AI models whose output is hybrid field-synthetic data. An underlying physics model (included in the generative AI model) may restrict the space into which the generative AI model can generate synthetic field data. By restricting the space, the generative AI model may produce more realistic and constrained data (hybrid field-synthetic data). The hybrid field-synthetic data may be built in massive quantities and used to train subsequent machine learning models.

One or more of models described herein may include an artificial neural network. FIG. 3 is a diagram illustrating an example artificial neural network included in some implementations. In FIG. 3, a fracking module 300 includes the artificial neural network (ANN) 302. The ANN 302 may include a plurality of neurons 304. The ANN 302 also may include an input layer having any suitable number of neurons 304 (supporting any suitable number of features). The input layer may intake features indicating aspects related to hydraulic fracturing (as described herein). The ANN 302 also may include an output layer that may predict a target (as described herein) and perform any of the operations described herein. The output layer may include any suitable number of neurons.

In some implementations, the plug movement detector may be integrated into a computer system. FIG. 4 is a block diagram illustrating a computer system that may be utilized with some implementations. In FIG. 4, the computer system 400 may include one or more processors 402 connected to a system bus 404. The system bus 404 may be connected to memory 408 and a network interface 405. The memory 408 may include any suitable memory random access memory (RAM), non-volatile memory (e.g., magnetic memory device), and/or any device for storing information and instructions executable by the processor(s) 402. The network interface 405 may provide connectivity to any suitable network, such as a wired network, wireless network, satellite network, etc.

The computer system 400 may include additional peripheral devices. For example, the computer system 400 may include multiple external multiple processors. In some implementations, any of the components may be integrated or subdivided.

The computer system 400 also may include the fracking module 800. The fracking module 800 may implement any one or more of the methods and operations described herein. The fracking module 800 may include an ANN 802 or other logic for performing machine-learning and classification operations described herein. In some implementations, the computer system 400 may be included in a well system (such as the well system described with reference to FIG. 10) and may cooperate with other components and/or systems to perform the functionality described herein.

The computer system 400 also may include a sensor controller 412 configured to perform operations for capturing sensor data and processing the sensor data (as described herein). The sensor controller 412 may transmit sensor data to the fracking module 800 or any other component in or external to the computer system 400.

The computer system 400 also may include a fracking controller 410 configured to perform operations for controlling hydraulic fracturing in a well. The fracking controller 410 may respond to output (such as predictions, classifications, or other outputs) from the fracking module 800. For example, the fracking controller 410 may alter at least one physical aspect of a fracking stage in response to a prediction (or classification) generated by the plug movement detector 118 (such as by causing a subsurface operation in a well).

Although the components are shown separately, any of the components of the computer system 400 may be further combined or subdivided. For example, the fracking module 800 and fracking controller 410 may be combined into a single component or subdivided into three or more components. Any component of the computer system 400 may be implemented as hardware, firmware, and/or machine-readable media including computer-executable instructions for performing the operations described herein. For example, some implementations include one or more non-transitory machine-readable media including computer-executable instructions including program code configured to perform functionality described herein. Machine-readable media includes any mechanism that provides (e.g., stores and/or transmits) information in a form readable by a machine (e.g., a computer system). For example, tangible machine-readable media includes read only memory (ROM), random access memory (RAM), magnetic disk storage media, optical storage media, flash memory machines, etc. Machine-readable media also includes any media suitable for transmitting software over a network.

Example Environment

The computer system 400 may be part of a larger system for drilling and fracturing well. FIG. 5 is an illustration depicting an example multi-well system, according to some implementations. In particular, FIG. 5 is a schematic of a multi-well system 500 that includes a wellbore 502 and a wellbore 508 in a subsurface formation 501. The wellbore 502 includes casing 506 and a number of perforations 590A-590H being made in the casing 506 at different depths to allow reservoir fluids (i.e., oil, water, and gas) from the subsurface formation 501 to flow into the wellbore 502. Similarly, the wellbore 508 includes casing 510 and a number of perforations 580A-580H being made in the casing 510 to allow reservoir fluids (i.e., oil, water, and gas) from the subsurface formation 501 to flow into the wellbore 508. During hydraulic fracturing operations of the wellbores 502 508, fracturing fluid, with or without sand, may be pumped into the subsurface formation 501, via the perforations 590A-590H and perforations 580A-580H, to hydraulically fracture the rock such that reservoir fluid may flow into the wellbore 502, 508, respectfully.

In some implementations, one or more sensors may be positioned in a wellbore to obtain measurements while an offset well is being hydraulically fractured. For example, the wellbore 502 may include a fiber optic cable 520 to obtain strain measurements, temperature measurements, derived pressure measurements (from strain measurements), etc. of the subsurface formation 501 while the wellbore 508 is being hydraulically fractured. The fiber optic cable 520 may extend from the wellhead 514 on the surface 511 to the subsurface along the wellbore 502. The fiber optic cable 520 may be cemented in place in the annular space between the casing 506 of the wellbores 502 and the subsurface formation 501. The fiber optic cable 520 may be clamped to the outside of the casing 506 during deployment and protected by centralizers and cross coupling clamps. The fiber optic cables 520 may be included with coiled tubing, wireline, loose fiber using coiled tubing, or gravity deployed fiber coils that unwind the fiber as the coils are moved in the wellbore 502. The fiber optic cable 520 also may be deployed with pumped down coils and/or self-propelled containers. Additional deployment options for the fiber optic cable 520 may include coil tubing and wireline deployed coils where the fiber optic cables 520 are anchored at the toe of the wellbore. In such implementations the fiber optic cable 520 may be deployed when the wireline or coiled tubing is removed from the well. The fiber optic cable 520 may house one or more optical fibers, and the optical fibers may be single mode fibers, multi-mode fibers, or a combination of single mode and multi-mode optical fibers. The distribution of sensors shown in FIG. 5 is for example purposes only. Any suitable sensor deployment may be used.

The fiber optic cable 520 may be used for distributed sensing where acoustic, vibration, strain, and temperature measurements may be collected downhole in the wellbores 502. The measurements may be collected at various positions distributed along the fiber optic cable 520. For example, data may be collected every 1-3 ft along the full length of the fiber optic cable 520 downhole along the horizontal section of the wellbore. Fiber optic interrogation unit 522 of the wellbore 502 may be located on the surface 511 of the multi-well system 500. The fiber optic interrogation units 522 may be directly coupled to the fiber optic cables 520. Alternatively, the fiber optic interrogation units 522 may be coupled to a fiber stretcher module, wherein the fiber stretcher module is coupled to the fiber optic cable 520. The fiber optic interrogation unit 522 may receive measurement values taken and/or transmitted along the length of the fiber optic cable 520 such as acoustic, temperature, strain, etc. The fiber optic interrogation unit 522 may be electrically connected to a digitizer to convert optically transmitted measurements into digitized measurements.

The fiber optic interrogation unit 522 may operate using various sensing principles including but not limited to amplitude-based sensing systems like DTS, DAS, Low Frequency Distributed Acoustic Sensing (LFDAS), Distributed Vibration Sensing (DVS), and Distributed Strain Sensing (DSS). For example, the DTS system may be based on Raman and/or Brillouin scattering. A DAS system may be a phase sensing-based system based on interferometric sensing using homodyne or heterodyne techniques where the system may sense phase or intensity changes due to constructive or destructive interference. The DAS system may also be based on Rayleigh scattering and in particular coherent Rayleigh scattering. A DSS system may be a strain sensing system using dynamic strain measurements based on interferometric sensors or static strain sensing measurements using Brillouin scattering. DAS systems based on Rayleigh scattering may also be used to detect dynamic strain events. Temperature effects may in some cases be subtracted from both static and/or dynamic strain events, and temperature profiles may be measured using Raman based systems and/or Brillouin based systems capable of differentiating between strain and temperature, and/or any other optical and/or electronic temperature sensors, and/or any other optical and/or electronic temperature sensors, and/or estimated thermal events.

In some implementations, the fiber optic interrogation unit 522 may measure changes in optical fiber properties between two points in an optical fiber at any given point, and these two measurement points move along the optical sensing fiber as light travels along the optical fiber. Changes in optical properties may be induced by strain, vibration, acoustic signals and/or temperature as a result of the fluid flow. Phase and intensity based interferometric sensing systems are sensitive to temperature and mechanical, as well as acoustically induced, vibrations. DAS data may be converted from time series data to frequency domain data using Fast Fourier Transforms (FFT) and other transforms, like wavelet transforms, also may be used to generate different representations of the data. Various frequency ranges may be used for different purposes and low frequency signal changes may be attributed to formation strain changes or fluid movement and other frequency ranges may be indicative of fluid movement. Various techniques may be applied to generate indicators of events related to the generation and/or expansion of shear induced fracture fields during hydraulic fracturing operations. Although FIG. 5 depicts the fiber optic cable 520 in the wellbore 502, a fiber optic cable 520 may also be positioned in the wellbore 508 to obtain measurements when the wellbore 502 is hydraulically fractured.

The wellbore 502 may also include pressure sensors, such as externally ported pressure sensors 530, 532, to measure the formation pressure while the offset wellbore 508 is hydraulically fractured. Although FIG. 5 depicts the externally ported pressure sensors 530, 532 at the heel and toe of the wellbore 502, respectively, the externally ported pressure sensors 530, 532 may be positioned at any suitable location in the wellbore 502. Although FIG. 5 depicts the externally ported pressure sensors 530, 532 external to the casing 506 of the wellbore 502, externally ported pressure sensors 530, 532 may also be positioned in the wellbore 508 to obtain measurements when the wellbore 502 is hydraulically fractured.

During the hydraulic fracturing operations of wellbore 502 and/or wellbore 508, shear induced fracturing fields may be generated and/or dilated. For example, the shear induced fracturing fields comprising Mode 2 and/or Mode 3 failures may form between clusters of a stage, between stages of a wellbore, between clusters and/or stages of offset wellbores, etc. In some implementations, the fiber optic cable 520 and/or the externally ported pressure sensors 530, 532 may obtain measurements of the subsurface formation 501 to detect and/or monitor the subsurface formation 501 and the shear induced fracture fields.

A computer 570 may be communicatively coupled to the fiber optic interrogation units 522, externally ported pressure sensors 530, 532, and other sensors in the multi-well system 500. The computer 580 may include a signal processor to perform various signal processing operations on signals captured by the fiber optic interrogation units 522, externally ported pressure sensors 530, 532, and/or other components of the multi-well system 500. The computer 570 may have one or more processors and a memory device to analyze the measurements and graphically represent analysis results on a display device. The computer 570 may include one or more of the components described with reference to FIG. 9. The computer 570 may be configured to identify plug movement during stages of hydraulic fracturing (as described herein). Although FIG. 5 depicts a system with multiple wellbores, embodiments described herein may also be applicable to other systems such as a single well system, multiple pads, etc.

FIGS. 1-5 and the operations described herein are examples meant to aid in understanding example implementations and should not be used to limit the potential implementations or limit the scope of the claims. None of the implementations described herein may be performed exclusively in the human mind nor exclusively using pencil and paper. None of the implementations described herein may be performed without computer components such as those described herein. Some implementations may perform additional operations, fewer operations, operations in parallel or in a different order, and some operations differently. Some implementations may perform the operations with different components.

As used herein, a phrase referring to “at least one of” a list of items refers to any combination of those items, including single members. As an example, “at least one of: a, b, or c” is intended to cover: a, b, c, a-b, a-c, b-c, and a-b-c.

The various illustrative logics, logical blocks, modules, circuits, and algorithm processes described in connection with the implementations disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. The interchangeability of hardware and software has been described in terms of functionality, and illustrated in the various illustrative components, blocks, modules, circuits and processes described throughout. Whether such functionality is implemented in hardware or software depends upon the particular application and design constraints imposed on the overall system.

The hardware and data processing apparatus used to implement the various illustrative logics, logical blocks, modules and circuits described in connection with the implementations disclosed herein may be implemented or performed with a general purpose single- or multi-chip processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor or any conventional processor, controller, microcontroller, or state machine. A processor also may be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. In some implementations, particular processes and methods may be performed by circuitry that is specific to a given function.

In one or more implementations, the functions described may be implemented in hardware, digital electronic circuitry, computer software, firmware, including the structures disclosed in this specification and their structural equivalents thereof, or in any combination thereof. Implementations of the subject matter described in this specification also may be implemented as one or more computer programs, such as one or more modules of computer program instructions stored on a computer storage media for execution by, or to control the operation of, a computing device.

If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. The processes of a method or algorithm disclosed herein may be implemented in a processor-executable instructions which may reside on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that may be enabled to transfer a computer program from one place to another. Storage media may be any available media that may be accessed by a computer. By way of example, and not limitation, such computer-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that may be used to store desired program code in the form of instructions or data structures and that may be accessed by a computer. Also, any connection may be properly termed a computer-readable medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-Ray™ disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations also may be included within the scope of computer-readable media. Additionally, the operations of a method or algorithm may reside as one or any combination or set of codes and instructions on a machine readable medium and computer-readable medium, which may be incorporated into a computer program product.

Various modifications to the implementations described in this disclosure may be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other implementations without departing from the spirit or scope of this disclosure. Thus, the claims are not intended to be limited to the implementations shown herein but are to be accorded the widest scope consistent with this disclosure, the principles and the novel features disclosed herein.

Certain features that are described in this specification in the context of separate implementations also may be implemented in combination in a single implementation. Conversely, various features that are described in the context of a single implementation also may be implemented in multiple implementations separately or in any suitable subcombination. Moreover, although features may be described as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination may in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.

Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. Further, the drawings may schematically depict one more example process in the form of a flow diagram. However, some operations may be omitted and/or other operations that are not depicted may be incorporated in the example processes that are schematically illustrated. For example, one or more additional operations may be performed before, after, simultaneously, or between any of the illustrated operations. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of various system components in the implementations described should not be understood as requiring such separation in all implementations, and the described program components and systems may be integrated together in a single software product or packaged into multiple software products. Additionally, other implementations are within the scope of the following claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve desirable results.

Some parts of this disclosure refer to the “trained plug movement detector” to clarify that the plug movement detector has completed training with the training data set. However, references to the “plug movement identifier” that do not include “trained” may nevertheless be referring to instances for which training is complete.

Example Clauses

Some implementations may include the following clauses.

• • Clause 1: A method comprising: generating temporally variant engineered features from data for fracking stages; determining a plurality of machine-learning models based on the engineered features; training the machine-learning models via regression analysis with the engineered features as inputs and answer product responses as targets of the machine-learning models; clustering certain of the fracking stages into clusters having similar answer product responses for both far field and near field; performing, via a physics-based fracture model, fracture modeling on each cluster to determine uncertainty for the answer product responses; determining improvements to the physics-based fracture model based on the clusters of fracking stages; generating a database of clustered temporal features and remediation strategies; and generating, based on the database, a digital twin to control fracking operations in a well.
  • Clause 2: The method of clause 1 further comprising: performing one or more of the fracking operations in the well; after performing the fracking operations, retraining the machine-learning model via regression analysis based, at least in part, on additional answer product responses and sensor measurements arising from the hydraulic fracturing process.
  • Clause 3: The method of any one or more of clauses 1-2, wherein the data for fracking stages includes one or more of pressure, slurry rate, and proppant concentration and wherein the temporal variance includes one or more of rolling statistical measures, derivatives, cumulative summations, lag features, Fourier information, wavelet information, entropy, and amplitude ratios.
  • Clause 4: The method of any one or more of clauses 1-3, wherein the answer product responses are based, at least in part, on surface sensor measurements and in situ sensor measurements of the hydraulic fracturing processes.
  • Clause 5: The method of any one or more of clauses 1-4, wherein at least one of the clusters do not include a near field measurement or does not include a far field measurement.
  • Clause 6: The method of any one or more of clauses 1-5 further comprising: generating data with which the temporally variant engineered features are generated; generating the answer product responses; wherein the data and the answer product responses are generated via one or more of a machine-learning model and/or a physics-based model
  • Clause 7: The method of any one or more of clauses 1-6, wherein the clustering certain of the fracking stages is based on Shapely analysis and gradient class maps.
  • Clause 8: The method of any one or more of clauses 1-7, further comprising: defining a cost function to generate a metric of weighted importances of the stages; and applying the cost function to the stages.
  • Clause 9: One or more machine-readable mediums including computer-executable instructions that cause one or more processors to perform operations for hydraulic fracturing in a well, the instructions comprising: instructions to generate temporally variant engineered features from data for fracking stages; instructions to determine a plurality of machine-learning models based on the engineered features; instructions to train the machine-learning models via regression analysis with the engineered features as inputs and answer product responses as targets of the machine-learning models; instructions to cluster certain of the fracking stages into clusters having similar answer product responses for both far field and near field; instructions to perform, via a physics-based fracture model, fracture modeling on each cluster to determine uncertainty for the answer product responses; instructions to determine improvements to the physics-based fracture model based on the clusters of fracking stages; instructions to generate a database of clustered temporal features and remediation strategies; and instructions to generate, based on the database, a digital twin to control fracking operations in a well.
  • Clause 10: The one or more machine-readable mediums of clause 9, the instructions further comprising: instructions to perform one or more of the fracking operations in the well; instructions to, after performing the fracking operations, retrain the machine-learning model via regression analysis based, at least in part, on additional answer product responses and sensor measurements arising from the hydraulic fracturing process.
  • Clause 11: The one or more machine-readable mediums of any one or more of clauses 9-10, wherein the data for fracking stages includes one or more of pressure, slurry rate, and proppant concentration, rolling statistical measures, derivatives, cumulative summations, lag features, Fourier information, wavelet information, entropy, and amplitude ratios.
  • Clause 12: The one or more machine-readable mediums of any one or more of clauses 9-11, wherein the answer product responses are based, at least in part, on surface sensor measurements and in situ sensor measurements of the hydraulic fracturing processes.
  • Clause 13: The one or more machine-readable mediums of any one or more of clauses 9-12, wherein at least one of the clusters do not include a near field measurement or does not include a far field measurement.
  • Clause 14: The one or more machine-readable mediums of any one or more of clauses 9-13, the instructions further comprising: modifying a physical aspect of the well based on one of the remediation strategies provided by the digital twin.
  • Clause 15: The one or more machine-readable mediums of any one or more of clauses 9-14, wherein the clustering certain of the fracking stages is based on Shapely analysis and gradient class maps.
  • Clause 16: The one or more machine-readable mediums of any one or more of clauses 9-15, the instructions further comprising: defining a cost function to generate a metric of weighted importances of the stages; and applying the cost function to the stages.
  • Clause 17: A system comprising: one or more processors; and one or more machine-readable mediums including computer-executable instructions that cause the one or more processors to perform operations for hydraulic fracturing in a well, the instructions including instructions to generate temporally variant engineered features from data for fracking stages; instructions to determine a plurality of machine-learning models based on the engineered features; instructions to train the machine-learning models via regression analysis with the engineered features as inputs and answer product responses as targets of the machine-learning models; instructions to cluster certain of the fracking stages into clusters having similar answer product responses for both far field and near field; instructions to perform, via a physics-based fracture model, fracture modeling on each cluster to determine uncertainty for the answer product responses; instructions to determine improvements to the physics-based fracture model based on the clusters of fracking stages; instructions to generate a database of clustered temporal features and remediation strategies; and instructions to generate, based on the database, a digital twin to control fracking operations in a well.
  • Clause 18: The system of clause 17, the instructions further including: instructions to perform one or more of the fracking operations in the well; instructions to, after performing the fracking operations, retrain the machine-learning model via regression analysis based, at least in part, on additional answer product responses and sensor measurements arising from the hydraulic fracturing process.
  • Clause 19: The system of any one or more of clauses 17-18, wherein the data for fracking stages includes one or more of pressure, slurry rate, and proppant concentration, rolling statistical measures, derivatives, cumulative summations, lag features, Fourier information, wavelet information, entropy, and amplitude ratios.
  • Clause 20: The system of any one or more of clauses 17-19, wherein the answer product responses are based, at least in part, on surface sensor measurements and in situ sensor measurements of the hydraulic fracturing processes.