Methods for Managing Pump-lifted Wells in Hydrocarbon Fields

Abstract

A methodology for integrated pump and well data for managing pump-lifted wells in hydrocarbon fields is provided. Rod pumps are widely applied to various types of wells to provide uplift and may play a dominant role in the flow of hydrocarbon from reservoir to surface. Assess the state of each well and its components (including the associated rod pump) is important for optimal well and reservoir management for fields with artificially lifted wells. As such, a methodology is disclosed that performs machine learning in order to generate a machine-learned model using pump card data and at least one of well data, reservoir data, flowline data, user input data, or data generated from analysis of one or more of the well data, the reservoir data, the flowline data, or the user input data.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This application claims the benefit of U.S. Provisional Application Ser. No. 63/267,113, entitled “Methods for Managing Pump-lifted Wells in Hydrocarbon Fields,” filed Jan. 25, 2022, the disclosure of which is hereby incorporated by reference in its entirety.

FIELD OF THE INVENTION

The present application relates generally to the field of hydrocarbon production. Specifically, the disclosure relates to a methodology for managing pump-lifted wells in hydrocarbon fields.

BACKGROUND OF THE INVENTION

This section is intended to introduce various aspects of the art, which may be associated with exemplary embodiments of the present disclosure. This discussion is believed to assist in providing a framework to facilitate a better understanding of particular aspects of the present disclosure. Accordingly, it should be understood that this section should be read in this light, and not necessarily as admissions of prior art.

A hydrocarbon well may be utilized to produce hydrocarbons from a subterranean formation. Often, a wellbore liquid may build up within one or more portions of the hydrocarbon well. This wellbore liquid, which may include water, condensate, and/or liquid hydrocarbons, may impede flow of the gaseous hydrocarbons from the subterranean formation to a surface region via the hydrocarbon well, thereby reducing and/or completely blocking gaseous hydrocarbon production from the hydrocarbon well.

Sucker rod pumping (also known as rod pumping) is the most widely artificial lift method to provide uplift and oftentimes plays a dominant role to flow hydrocarbon from a reservoir to the surface. In particular, a reciprocating system, which may include a sucker rod pump for pumping liquids from a wellbore, is an example system to provide the artificial lift. A sucker rod pump typically includes a rocking beam with one end coupled to a pump motor by a crank assembly. The crank assembly has a counterweight intended to balance the loading of the motor by offsetting at least part of the weight of the pump connecting rods, which are cantilevered on the opposite end of the rocking beam. Nevertheless, as the rods to the downhole pump are raised and lowered, the loading of the motor passes through a cycle during which potential energy is stored as the pump rods are lifted, and released as the pump rods are lowered. An example of a rod pump is disclosed in US Patent Application Publication No. 2020/0190965 A1, incorporated by reference herein in its entirety.

The wells may be part of a well pad, which may represent an area where the ground surface has been prepared for drilling and completion operations, and may comprise a collection of wells, pipeline(s), and one or more rod pumps. Wells are frequently instrumented for purposes of assessing operational parameters. The fluid flow rate produced by the well is an advantageous parameter to measure, and can be measured using flow rate sensors at any point along the conduits through which the fluid is pumped. The fluid pressures produced in the well by the pump can also be monitored, and used to develop additional information, such as the rate at which the geological formation is refilling the pump, and other aspects of well performance. One way to sense well fluid pressure indirectly is to sense tension and compression of the moving pump structures, for example using strain gauges mounted on such structures or load cells coupled between them.

There are a number of aspects of one or more of the well, the well pad, or the pump performance that may be pertinent to issues of efficiency, maintenance, capacity, switching between operational modes and the like. A monitoring system and controller may sense conditions and adjust operational parameters such as the frequency of cyclic operation, the manner in which power is coupled to the motor windings and so forth.

Thus, the monitoring system and controller, configured to perform optimal well and reservoir management for fields with artificially lifted wells using rod pumps, may assess the state of each well and its components (including the rod pump) using the available data. Various data is available including pressure and temperature information in the wellbore, casing, or flowline, reservoir, or pump. Such data includes, but limited to, tubing pressure and temperature, casing pressure and temperature, flowline pressure and temperature, vibration, downhole pressure and temperature, liquid levels, and reservoir pressure. The available data may also include rod pump data such as fillage, pump speed, motor power, and dynamometer card.

In particular, the dynamometer card represents the relationship between load and distance travelled by the rod (e.g., a measurement of rod loading vs. position). In each stroke, based on the relationship between the load applied to rod string and the position of rod, the “pump card” may be extracted. In practice, measurements are gathered at the surface and typically transformed to downhole dynagraph cards by using variations of the wave equation, which may be a second-order linear partial differential equation for the description of waves as they occur in classical physics, such as those that occur in fluid dynamics. The downhole cards may then be evaluated to determine whether the pumping system is operating properly, and if not, how it can be improved.

SUMMARY OF THE INVENTION

In one or some embodiments, a computer-implemented method for managing one or more of a well, a rod pump, or a well pad for hydrocarbon extraction is disclosed. The method includes: performing machine learning in order to generate a machine-learned model using pump card data and at least one of well data, reservoir data, flowline data, user input data, or data generated from analysis of one or more of the well data, the pump card data, the reservoir data, the flowline data, or the user input data; predicting, by the machine-learned model, at least one aspect of the well, the rod pump, or the well pad; and using the predicted at least one aspect for hydrocarbon management.

BRIEF DESCRIPTION OF THE DRAWINGS

The present application is further described in the detailed description which follows, in reference to the noted plurality of drawings by way of non-limiting examples of exemplary implementations, in which like reference numerals represent similar parts throughout the several views of the drawings. In this regard, the appended drawings illustrate only exemplary implementations and are therefore not to be considered limiting of scope, for the disclosure may admit to other equally effective embodiments and applications.

FIG. 1 is a block diagram of the different types of data as input to the pump status diagnostics.

FIG. 2 is a flowchart for training and retraining.

FIG. 3 is a table of example pump card labels by category.

FIG. 4 is a table of example features in a model.

FIGS. 5A-D illustrate an example of various aspects of a rod pump diagnostics results with FIG. 5A illustrating pump stroke inferred rate over time with darker dots indicative of the rate associated with a bad pump card and light dots indicative of the rate associated with a good pump card, and with FIGS. 5B-D illustrates surface (darker) and downhole (lighter) cards at different points.

FIG. 6A is an example model prediction in the gas interference label category.

FIG. 6B is an example model prediction in the beam transducer (BT) wire label category.

FIG. 6C is an example model prediction in the traveling valve or plunger leak label category.

FIG. 6D is an example model prediction in the hang up label category.

FIG. 6E is an example model prediction in the SAM load calibration label category.

FIG. 6F is an example model prediction in the reference error label category.

FIGS. 7A-F illustrate another example various aspects of a rod pump diagnostics results with FIG. 7A illustrating pump stroke inferred rate over time with darker dots indicative of the rate associated with a bad pump card and light dots indicative of the rate associated with a good pump card, and with FIGS. 7B-D illustrates surface (darker) and downhole (lighter) cards at different points including FIG. 7B illustrating a bad pump card indicative of hang up, FIG. 7C illustrating a good pump card, and FIG. 7B illustrating a bad pump card indicative of flumping.

FIG. 7E illustrates a bad pump card indicative of gas interference and FIG. 7F illustrates a bad pump card indicative of plunger leak.

FIG. 8 is an example user interface illustrating various data.

FIG. 9 are examples of data being displayed in time-series.

FIG. 10 is a diagram of an exemplary computer system that may be utilized to implement the methods described herein.

DETAILED DESCRIPTION OF THE INVENTION

The methods, devices, systems, and other features discussed below may be embodied in a number of different forms. Not all of the depicted components may be required, however, and some implementations may include additional, different, or fewer components from those expressly described in this disclosure. Variations in the arrangement and type of the components may be made without departing from the spirit or scope of the claims as set forth herein. Further, variations in the processes described, including the addition, deletion, or rearranging and order of logical operations, may be made without departing from the spirit or scope of the claims as set forth herein.

It is to be understood that the present disclosure is not limited to particular devices or methods, which may, of course, vary. It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only, and is not intended to be limiting. As used herein, the singular forms “a,” “an,” and “the” include singular and plural referents unless the content clearly dictates otherwise. Furthermore, the words “can” and “may” are used throughout this application in a permissive sense (i.e., having the potential to, being able to), not in a mandatory sense (i.e., must). The term “include,” and derivations thereof, mean “including, but not limited to.” The term “coupled” means directly or indirectly connected. The word “exemplary” is used herein to mean “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects. The term “uniform” means substantially equal for each sub-element, within about ±10% variation.

The term “seismic data” as used herein broadly means any data received and/or recorded as part of the seismic surveying and interpretation process, including displacement, velocity and/or acceleration, pressure and/or rotation, wave reflection, and/or refraction data. “Seismic data” is also intended to include any data (e.g., seismic image, migration image, reverse-time migration image, pre-stack image, partially-stack image, full-stack image, post-stack image or seismic attribute image) or interpretation quantities, including geophysical properties such as one or more of: elastic properties (e.g., P and/or S wave velocity, P-Impedance, S-Impedance, density, attenuation, anisotropy and the like); and porosity, permeability or the like, that the ordinarily skilled artisan at the time of this disclosure will recognize may be inferred or otherwise derived from such data received and/or recorded as part of the seismic surveying and interpretation process. Thus, this disclosure may at times refer to “seismic data and/or data derived therefrom,” or equivalently simply to “seismic data.” Both terms are intended to include both measured/recorded seismic data and such derived data, unless the context clearly indicates that only one or the other is intended. “Seismic data” may also include data derived from traditional seismic (e.g., acoustic) data sets in conjunction with other geophysical data, including, for example, gravity plus seismic; gravity plus electromagnetic plus seismic data, etc. For example, joint-inversion utilizes multiple geophysical data types.

The term “geophysical data” as used herein broadly includes seismic data, as well as other data obtained from non-seismic geophysical methods such as electrical resistivity. In this regard, examples of geophysical data include, but are not limited to, seismic data, gravity surveys, magnetic data, electromagnetic data, well logs, image logs, radar data, or temperature data.

The term “formation” refers to any definable subsurface region. The formation may contain one or more hydrocarbon-containing layers, one or more non-hydrocarbon containing layers, an overburden, and/or an underburden of any geologic formation.

The term “geological features” (interchangeably termed geo-features) as used herein broadly includes attributes associated with a subsurface, such as any one, any combination, or all of: subsurface geological structures (e.g., channels, volcanos, salt bodies, geological bodies, geological layers, etc.); boundaries between subsurface geological structures (e.g., a boundary between geological layers or formations, etc.); or structure details about a subsurface formation (e.g., subsurface horizons, subsurface faults, mineral deposits, bright spots, salt welds, distributions or proportions of geological features (e.g., lithotype proportions, facies relationships, distribution of petrophysical properties within a defined depositional facies), etc.). In this regard, geological features may include one or more subsurface features, such as subsurface fluid features, that may be hydrocarbon indicators (e.g., Direct Hydrocarbon Indicator (DHI)). Examples of geological features include, without limitation salt, fault, channel, environment of deposition (EoD), facies, carbonate, rock types (e.g., sand and shale), horizon, stratigraphy, or geological time, and are disclosed in US Patent Application Publication No. 2010/0186950 A1, incorporated by reference herein in its entirety.

The terms “velocity model,” “density model,” “physical property model,” or other similar terms as used herein refer to a numerical representation of parameters for subsurface regions. Generally, the numerical representation includes an array of numbers, typically a 2-D or 3-D array, where each number, which may be called a “model parameter,” is a value of velocity, density, or another physical property in a cell, where a subsurface region has been conceptually divided into discrete cells for computational purposes. For example, the spatial distribution of velocity may be modeled using constant-velocity units (layers) through which ray paths obeying Snell's law can be traced. A 3-D geologic model (particularly a model represented in image form) may be represented in volume elements (voxels), in a similar way that a photograph (or 2-D geologic model) may be represented by picture elements (pixels). Such numerical representations may be shape-based or functional forms in addition to, or in lieu of, cell-based numerical representations.

The term “subsurface model” as used herein refer to a numerical, spatial representation of a specified region or properties in the subsurface.

The term “geologic model” as used herein refer to a subsurface model that is aligned with specified geological feature such as faults and specified horizons.

The term “reservoir model” as used herein refer to a geologic model where a plurality of locations have assigned properties including any one, any combination, or all of rock type, EoD, subtypes of EoD (sub-EoD), porosity, clay volume, permeability, fluid saturations, etc.

For the purpose of the present disclosure, subsurface model, geologic model, and reservoir model are used interchangeably unless denoted otherwise.

Stratigraphic model is a spatial representation of the sequences of sediment, formations and rocks (rock types) in the subsurface. Stratigraphic model may also describe the depositional time or age of formations.

Structural model or framework results from structural analysis of reservoir or geobody based on the interpretation of 2D or 3D seismic images. For examples, the reservoir framework comprises horizons, faults and surfaces inferred from seismic at a reservoir section.

The term “hydrocarbon” refers to an organic compound that includes primarily, if not exclusively, the elements hydrogen and carbon. Examples of hydrocarbons include any form of natural gas, oil, coal, and bitumen that can be used as a fuel or upgraded into a fuel.

The term “hydrocarbon fluids” refers to a hydrocarbon or mixtures of hydrocarbons that are gases or liquids. For example, hydrocarbon fluids may include a hydrocarbon or mixtures of hydrocarbons that are gases or liquids at formation conditions, at processing conditions, or at ambient conditions (e.g., 20° C. and 1 atm pressure). Hydrocarbon fluids may include, for example, oil, natural gas, gas condensates, coal bed methane, shale oil, shale gas, and other hydrocarbons that are in a gaseous or liquid state.

The term “sensor” includes any electrical sensing device or gauge. The sensor may be capable of monitoring or detecting pressure, temperature, fluid flow, vibration, resistivity, or other formation data. Alternatively, the sensor may be a position sensor.

The term “subsurface” refers to geologic strata occurring below the earth's surface.

The term “wellbore” refers to a hole in the subsurface made by drilling or insertion of a conduit into the subsurface. A wellbore may have a substantially circular cross section, or other cross-sectional shape. As used herein, the term “well,” when referring to an opening in the formation, may be used interchangeably with the term “wellbore.”

The terms “zone” or “zone of interest” refer to a portion of a subsurface formation containing hydrocarbons. The term “hydrocarbon-bearing formation” may alternatively be used.

The terms “hydrocarbon management” or “managing hydrocarbons” include any one, any combination, or all of the following: hydrocarbon extraction; hydrocarbon production, (e.g., drilling a well and prospecting for, and/or producing, hydrocarbons using the well; and/or, causing a well to be drilled, e.g., to prospect for hydrocarbons); hydrocarbon exploration; identifying potential hydrocarbon-bearing formations; characterizing hydrocarbon-bearing formations; identifying well locations; determining well injection rates; determining well extraction rates; identifying reservoir connectivity; acquiring, disposing of, and/or abandoning hydrocarbon resources; reviewing prior hydrocarbon management decisions; and any other hydrocarbon-related acts or activities, such activities typically taking place with respect to a subsurface formation. The aforementioned broadly include not only the acts themselves (e.g., extraction, production, drilling a well, etc.), but also or instead the direction and/or causation of such acts (e.g., causing hydrocarbons to be extracted, causing hydrocarbons to be produced, causing a well to be drilled, causing the prospecting of hydrocarbons, etc.). Hydrocarbon management may include reservoir surveillance and/or geophysical optimization. For example, reservoir surveillance data may include, well production rates (how much water, oil, or gas is extracted over time), well injection rates (how much water or CO2 is injected over time), well pressure history, and time-lapse geophysical data. As another example, geophysical optimization may include a variety of methods geared to find an optimum model (and/or a series of models which orbit the optimum model) that is consistent with observed/measured geophysical data and geologic experience, process, and/or observation.

As used herein, “obtaining” data generally refers to any method or combination of methods of acquiring, collecting, or accessing data, including, for example, directly measuring or sensing a physical property, receiving transmitted data, selecting data from a group of physical sensors, identifying data in a data record, and retrieving data from one or more data libraries.

As used herein, terms such as “continual” and “continuous” generally refer to processes which occur repeatedly over time independent of an external trigger to instigate subsequent repetitions. In some instances, continual processes may repeat in real time, having minimal periods of inactivity between repetitions. In some instances, periods of inactivity may be inherent in the continual process.

If there is any conflict in the usages of a word or term in this specification and one or more patent or other documents that may be incorporated herein by reference, the definitions that are consistent with this specification should be adopted for the purposes of understanding this disclosure.

As discussed in the background, a monitoring system and controller, in order to optimally manage artificially lifted wells or well pads using rod pumps, assesses the state of each well and its components (including the rod pump) using all available data. In the absence of physics-based models for such wells, machine learning technology may assist in providing surveillance and optimization capabilities. However, machine learning is typically very limited in this area and suffers from serious challenges including: (1) not incorporating other information/features other than pump card itself; and (2) asset/reservoir specific model development.

Thus, in one or some embodiments, information and/or features (other than pump card data) are used to train the neural network in order to generate a machine-learned model to predict at least one aspect of the system (such as predict in near real time any one, any combination, or all of the status or condition of at least one aspect of the well, the well pad, the status or condition of at least one aspect of pump system, or at least one aspect of the pump card), Further, separate from predicting at least one aspect of the system, the machine-learned model may further include providing recommended action(s) (e.g., automated decision support) for use with hydrocarbon management. Merely by way of example, the method may select any one or any combination of the following actions: pump change; pump check; pump recalibration; pump speed (e.g., modification of the pump speed); shutting in the well; etc. Thus, various types of interventions are contemplated including: (1) well stimulation; (2) modify operation of the pump in some aspect (e.g., repair the pump, replace the pump, etc.); (3) modify the measurement tool itself that produces the pump cards (akin to a sensor recalibration).

In a particular embodiment, the neural network is trained using pump card data and additional data from any one, any combination, or all of: well data; reservoir data; flowline data; user input data (e.g., user feedback; maintenance data); or data generated from analysis of or derivative of any one, any combination, or all of well data, pump card data (e.g., geometric structure of the pump card), reservoir data, flowline data, or user input data (e.g., history or trend data).

Thus, unlike previous solutions which limited the input to the data generated by the pump card, the methodology, which may comprise machine learning, may use the different inputs for training. In this regard, the neural network trained by the machine learning may consider the different types of data in the first instance (as input for machine learning). Further, this is unlike previous machine learning methodologies, which solely used data generated by the pump card as input for training the neural network, and which used the different types of data (such as reservoir data, well data, etc.) to compare against the output of the trained neural network. In this way, the data used to train the neural network to generate the machine-learned model considers the various reasons for failure. For example, pump cards, by their very nature, are complicated and may affect operation as well. In particular, the well head controller that generates these pump cards may impact the operation (e.g., calibration on the well ID process may fundamentally change the shape of those cards). Ignoring this and simply focusing on the data generated by the card is very limiting. Thus, the neural network, based on the input data, may consider the pump card history as well and not simply the data generated by the pump card.

Thus, in one or some embodiments, separate from pump card data, supervised learning may use different types of data for training, such as any one, any combination, or all of: data generated by other sensors (e.g., sensor(s) to generate well data); data based on analysis (e.g., historical data or trend data); data from user input; or data from the pump card other than the pump card data (e.g., service data, calibration data, etc. related to the pump card). In particular, user input data may be generated from a variety of sources. For example, user input data may comprise user-generated data related to user intervention in the system, including mechanics/service/operation data related to servicing various equipment, such as servicing or modifying any one, any combination, or all of the rod pump system, the well, the pump card, or the well pad. In particular, the user input data may relate to servicing the rod pump with a service rig, replacing equipment using the service rig, and/or modifying the configuration (e.g., any activity that may affect pump activity, such as positioning the pump higher or lower in the tubing; installing a heavier rod string; dumping acid or other chemical in the well bore to clean up the perforation where the fluid enters).

Further, the training of the neural network may be tailored in one or more aspects to the specific asset/reservoir. By way of example, training the neural network with well data and/or reservoir data resulting in tailoring the machine-learned model output to the specific well/reservoir.

In addition to (or instead of) training the neural network with the additional data discussed above, the machine-learned model may likewise be trained to generate one or more outputs including any one, any combination, or all of: predictive status/conditions (e.g., near real-time status/condition associated with one or both of the well or any associated equipment (e.g., any part of the reciprocating system and/or the pump card)); one or more recommended actions (e.g., prescriptive modeling); or associated loss/gain (predicted loss and/or gain associated with prescribed actions or inactions). In this regard, the output of the neural network may include one or both of predictive and prescriptive outputs. This is in contrast to typical outputs, which may limit the output to predictive status/conditions and fail to recommend a course of action and/or associated loss/gain.

For example, in one or some embodiments, the pump arrangement may include a rod string having a series of downhole rods coupling a downhole piston/chamber pump to an oil or gas production rod pump. The oil or gas production rod pump may have a rocking beam with one end connected to the downhole rods and an opposite end connected by eccentric linkages to a rotating counterweight member. The counterweight member may be rotated by an electric motor, being coupled by a belt or chain drive, and/or coupled to the motor through a gear train. In operation, as the motor turns the counterweight member, the beam may be rocked to raise and lower the downhole rods, operating the pump in a periodic manner at a relatively low frequency. The recommended course of action may comprise repairing or replacing one or more parts of the pump arrangement.

Further, the output may be generated in near real-time. For example, the data, which is generated periodically (such as hourly, daily, weekly, etc.) in real time or near real-time, may act as a trigger for using the trained neural network by immediately inputting the data to the trained neural network. In turn, the neural network may generate an output, such as the predictive status/conditions, the recommended action(s) or associated loss/gain. In this way, the method may predict near-real-time well and pump status/conditions using some or all available well, reservoir, flowline and pump card data from a given well. Specifically, the data-driven method of predicting well and pump conditions may combine historical pump card and well test data with real-time measurement data, with stroke-by-stroke predictions being useful for well and/or field surveillance and optimization.

In addition, in one or some embodiments, the training and/or re-training of the neural network may be automated at least in part. For example, as additional data is obtained, the neural network may be automatically re-trained so that the neural network may evolve using the additional data.

Further, in one or some embodiments, a user interface may be configured to assist in data labeling. As discussed above, the methodology may comprise three separate parts including data labeling, feature selection (via machine learning), and automation. With regard to data labeling, the user interface may assist in tagging/labeling data used in combination with the pump card data to train the neural network. In one or some embodiments, the pump card data is displayed in combination with (e.g., alongside or on top of) other types of data used to train the neural network, such as any one, any combination, or all of data generated by other sensors; data based on analysis; data from user input; or data from the pump card other than the pump card data. In one or some embodiments, the data may be displayed in a time series, such as different types of data simultaneously displayed on the user interface with a same reference time scale. In this way, the different types of data may be displayed, either in reference to one another, or in reference to the pump card data, on the same user interface simultaneously.

Thus, in one or some embodiments, the methodology may assist in one or more aspects of hydrocarbon management may be performed. In particular, the methodology may divide various tasks into separate modeled components. As one example, discrete tasks may include: (1) determining whether the pump cards are providing accurate measurements (e.g., is there a defect in the pump card, leading to unreliable data); (2) if the pump card is determined to be providing accurate measurement, determining at least one aspect of performance of the well (e.g., estimating fillage), and (3) given (1) and (2), determining one or more actions to improve operations (e.g., one or more recommendations to improve hydrocarbon management). This division of tasks (and the machine-learned models associated with the tasks that feed to one another) may better determine how to improve operations as opposed to a single general task of determining how to improve operations.

FIG. 1 is a block diagram 100 of the different types of data as input to the pump status diagnostics 120. As discussed above, various types of data, in addition to pump card data generated by the pump card, may be input to the pump status diagnostics 120. Merely by way of example, any one, any combination, or all of well/reservoir data 110, pump data 112, flowline data 114, history/trend data 116, or user input data 118 may be input to the pump status diagnostics 120.

Well data may comprise data associated with the well, such as sensor data that is associated with the well (e.g., casing pressure in the well; temperature; liquid level), data indicative of the well (e.g., well type (such as vertical well or horizontal well); geometry of the well; etc.), well test history, etc.

Reservoir data may comprise data indicative of the reservoir, such as steam history, pressure, volume and/or temperature. Pressure, volume, and/or temperature may change over time, which may be accounted for when building the machine-learned model for the pump status diagnostics 120.

Pump data 112 may include data generated by the pump card, and data associated with the pump card although not generated by the pump card (e.g., maintenance and/or calibration data associated with the pump card). Merely by way of example, pump card data may include any one, any combination, or all of: motor power; pump card history; fillage; pump speed; pump change history; or green pump pull history.

Flowline data 114 may comprise various aspects of the flowline, including temperature and/or pressure. History/trend data 116 may include any one, any combination, or all of: historical data (e.g., well servicing history) and/or trend data (e.g., trends of data including trends of well data, reservoir data; pump card data (e.g., changes of pump card data between pump cards from one time slice to another)); or difference data (e.g., difference or mapping between the surface pump and the bottomhole pump, such as at a given time slice). In this regard, the history/trend data 116 may indicate the progression of the data. User input data 118 may include any data indicative of user input to the system, such as user feedback and/or maintenance.

FIG. 2 is a flow chart 200 for training and retraining. As discussed above, the method of diagnosing rod pump performance and condition may include supervised machine learning to generate a data-driven machine-learned model. In one or some embodiments, supervised machine learning comprises a machine learning task of learning a function that maps an input to an output based on example input-output pairs, examples of which are disclosed in US Patent Application Publication No. 2020/0183032 A1, US Patent Application Publication No. 2020/0309978 A1, and US Patent Application Publication No. 2021/0041596 A1, all of which are incorporated by reference herein in their entirety.

At 210, data labeling is performed. In one or some embodiments, as part of supervised machine learning, data labeling is performed in order to have a sufficient amount of representative training set tied to specific asset/field. As discussed in more detail below (see FIG. 8), a labeling tool may be used to efficiently account for the potentially enormous amount pump cards. In this way, the trained model may be equipped with asset/field specific knowledge. The table 300 in FIG. 8 illustrates example pump card label categories and a column for counts associated with the labels.

At 220, feature selection and engineering are performed. As part of the machine learning process, various features may be selected as a subset of the available features. For example, while a pump card may carry important information, a well/pump may include other measurements or parameters available that are probative to assessing pump condition. As discussed above, these other measurements/parameters may include any one, any combination, or all of casing pressure, wellhead temperature, well type (e.g., horizontal or deviated), well size, bridle rod load, production rate, pump speed and previous/historical pump cards. To include these data as features of data-driven model, series of model training and evaluation may be performed. Via the machine learning process, the features that enhance model performance may be selected. Table 400 in FIG. 4 illustrates an example of features that may be included in a model. In particular, surface and downhole cards in table 400 may refer to raw card data (e.g., card position given in x and y domain) or any, geometric or otherwise, that may be derived from card information.

At 230, results may be generated. Various types of results are contemplated. Merely by way of example, FIGS. 5A-D illustrate various aspects of a rod pump diagnostics results with FIG. 5A including illustration 500 of pump stroke inferred rate over time with darker dots 510 indicative of the rate associated with a bad (or insufficiently functioning) pump card and lighter dots 512 indicative of the rate associated with a good (or sufficiently functioning) pump card, and with FIGS. 5B-D including illustrations 520, 530, 540 of surface 522 and downhole 524 cards at different points, illustrating how card shapes changed over time due to beam transducer (BT) wire failure. In this regard, the machine-learned model may accurately identify the pump condition.

FIGS. 6A-F illustrate example prediction results in different label categories, including FIG. 6A illustration 600 of gas interference (with dots 602 associated with a bad pump card and dots 604 associated with a good pump card), FIG. 6B illustration 610 of beam transducer (BT) wire (with dots 612 associated with a bad pump card and dots 614 associated with a good pump card); FIG. 6C illustration 620 of the traveling valve or plunger leak (with dots 622 associated with a bad pump card and dots 624 associated with a good pump card), FIG. 6D illustration 630 of the hang up (with dots 632 associated with a bad pump card and dots 634 associated with a good pump card), FIG. 6E illustration 640 of the SAM load calibration (with dots 642 associated with a bad pump card and dots 644 associated with a good pump card), and FIG. 6F illustration 650 of the reference error (with dots 652 associated with a bad pump card and dots 654 associated with a good pump card).

FIGS. 7A-F illustrate another example various aspects of a rod pump diagnostics results with FIG. 7A showing an illustration 700 of pump stroke inferred rate over time with darker dots 710 indicative of the rate associated with a bad pump card and light dots 712 indicative of the rate associated with a good pump card, and with FIGS. 7B-D illustrates a surface card 722, 732, 742 and a downhole card 724, 734, 744 at different points including FIG. 7B showing an illustration 720 a bad pump card indicative of hang up, FIG. 7C showing an illustration 730 a good pump card, and FIG. 7D showing an illustration 740 a bad pump card indicative of flumping. FIG. 7E showing an illustration 750 a bad pump card indicative of gas interference (with data from the surface card 752 and data from the downhole card 754) and FIG. 7F showing an illustration 760 a bad pump card indicative of plunger leak (with data from the surface card 762 and data from the downhole card 764).

At 240, input is received. The input may comprise any one, any combination, or all of: user input; sensor input (e.g., measurements generated by sensor or calculations derived from the measurements), or other input. At 250, it is determined whether to retrain the machine-learned model (e.g., responsive to receipt of the input at 240). If so, flow chart 200 loops back to 220. If not, flow chart 200 ends. In one or some embodiments, the machine-learned model may be purely data-driven and the labeling tool (discussed further below) may be implemented. Model re-training and the improvement process may be fully automated with the interaction of users. The users, who may comprise domain experts or field operators, may override the predicted label with theirs if they disagree, which is an example of user input. Then, the machine-learned model may be automatically re-trained with the new label to adapt and incorporate user feedback. This automated process allows the machine-learned model to evolve by adapting user feedback while minimizing the cost of maintenance.

FIG. 8 is an example user interface 800 illustrating various data. As discussed above, various types of data, separate from data generated by the pump card, may be used to train the network. User interface 800 is an example of a tool in which to label the various types of data to assist in the supervised machine learning. Specifically, FIG. 8 illustrates three sections, including section 810 (graphing rate versus time), section 850 (graphing load versus position), and section 870. Section 870 includes graphing variable options 872, pump card labels 874, and additional features 876. In practice, a user may designate variable options in variable options 872, which may then be graphed in section 810. As shown, inferred hourly 812, inferred hourly accepted 814, inferred hourly card 816, exception 818, baseline 820, reconciled inferred rate 822, alpha rate 824, presto daily 826, WTrate 828, Group 830, XSPOC™ Fillbase/10 (832) (from XSPOC™ Production Optimization Software from ChampionX), XSPOC™ Fillage/10 (834), and XSPOC™ SPM (836). Section 850 further includes a graph of load versus position for surface 852, downhole 854, and fillbase 856.

FIG. 9 are examples of data being displayed in time-series including 900, 930, 950. In particular, the time-series data is integrated (such as overlayed) with the pump card data, and may comprise history/trend 116 data. Thus, the pump card data may be integrated with the various other data available (e.g., any one, any combination, or all of: rate history; pump efficiency history; pump speed history; or temperature history) for use in machine learning to generate the machine-learned model.

In particular, time-series 900 illustrates rate information including data from rate-accepted/warning well tests 920 and data from temperature inferred rate-rejected well test 922 along with dashed line 910, which represents a change in at least one aspect of the pump (e.g., pump replacement; pump recalibration; modification of pump speed; etc.). For example, 906 represents a relative changes of rates of the pump (e.g., a relative change before and after a pump failure).

Time-series 930 illustrates pump efficiency (PE) 940 and Delta QT (the total rate) 942. Time-series 950 illustrates separator temperature 960 and speed of the pump (SPM) 962. A separator may separate oil and water and may perform various measurements associated with the separation, such as the rate of separation or the temperature of separation. Temperature of separation may be indicative of or correlated with performance of the well (and in turn may be used to assess effectiveness of the pump change using the temperature.

Further, time-series 900, 930 include columns 902, 904 representing a time-window before or after the change in the at least one aspect of the pump, with a change indicative of an “event” of the pump. For example, column 902 indicates a time-period before the event and column 904 indicates a time-period after the event. As discussed above, pump card data and one or more of other data (e.g., any one, any combination or all of well data, reservoir data, flowline data, user input data, or data generated from analysis of one or more of the well data, the reservoir data, the flowline data, or the user input data) may be used for machine learning. The other data may be selectively used, such as depending on the amount of data available in the respective time window. By way of example, time-series 930 indicates “True” and “False” associated with columns 902, with “True” indicating that sufficient data is included in the respective time window to use for machine learning and with “False” indicating that insufficient data is included in the respective time window to use for machine learning (so that the data associated with a respective event that is tagged as “False” is not used for machine learning).

In all practical applications, the present technological advancement must be used in conjunction with a computer, programmed in accordance with the disclosures herein. For example, FIG. 10 is a diagram of an exemplary computer system 1000 that may be utilized to implement methods described herein. A central processing unit (CPU) 1002 is coupled to system bus 1004. The CPU 1002 may be any general-purpose CPU, although other types of architectures of CPU 1002 (or other components of exemplary computer system 1000) may be used as long as CPU 1002 (and other components of computer system 1000) supports the operations as described herein. Those of ordinary skill in the art will appreciate that, while only a single CPU 1002 is shown in FIG. 10, additional CPUs may be present. Moreover, the computer system 1000 may comprise a networked, multi-processor computer system that may include a hybrid parallel CPU/GPU system. The CPU 1002 may execute the various logical instructions according to various teachings disclosed herein. For example, the CPU 1002 may execute machine-level instructions for performing processing according to the operational flow described.

The computer system 1000 may also include computer components such as non-transitory, computer-readable media. Examples of computer-readable media include computer-readable non-transitory storage media, such as a random access memory (RAM) 1006, which may be SRAM, DRAM, SDRAM, or the like. The computer system 1000 may also include additional non-transitory, computer-readable storage media such as a read-only memory (ROM) 1008, which may be PROM, EPROM, EEPROM, or the like. RAM 1006 and ROM 1008 hold user and system data and programs, as is known in the art. The computer system 1000 may also include an input/output (I/O) adapter 1010, a graphics processing unit (GPU) 1014, a communications adapter 1022, a user interface adapter 1024, a display driver 1016, and a display adapter 1018.

The I/O adapter 1010 may connect additional non-transitory, computer-readable media such as storage device(s) 1012, including, for example, a hard drive, a compact disc (CD) drive, a floppy disk drive, a tape drive, and the like to computer system 1000. The storage device(s) may be used when RAM 1006 is insufficient for the memory requirements associated with storing data for operations of the present techniques. The data storage of the computer system 1000 may be used for storing information and/or other data used or generated as disclosed herein. For example, storage device(s) 1012 may be used to store configuration information or additional plug-ins in accordance with the present techniques. Further, user interface adapter 1024 couples user input devices, such as a keyboard 1028, a pointing device 1026 and/or output devices to the computer system 1000. The display adapter 1018 is driven by the CPU 1002 to control the display on a display device 1020 to, for example, present information to the user such as subsurface images generated according to methods described herein.

The architecture of computer system 1000 may be varied as desired. For example, any suitable processor-based device may be used, including without limitation personal computers, laptop computers, computer workstations, and multi-processor servers. Moreover, the present technological advancement may be implemented on application specific integrated circuits (ASICs) or very large scale integrated (VLSI) circuits. In fact, persons of ordinary skill in the art may use any number of suitable hardware structures capable of executing logical operations according to the present technological advancement. The term “processing circuit” encompasses a hardware processor (such as those found in the hardware devices noted above), ASICs, and VLSI circuits. Input data to the computer system 1000 may include various plug-ins and library files. Input data may additionally include configuration information.

Thus, the computer is a high-performance computer (HPC), known to those skilled in the art. Such high-performance computers typically involve clusters of nodes, each node having multiple CPU's and computer memory that allow parallel computation. The models may be visualized and edited using any interactive visualization programs and associated hardware, such as monitors and projectors. The architecture of system may vary and may be composed of any number of suitable hardware structures capable of executing logical operations and displaying the output according to the present technological advancement. Those of ordinary skill in the art are aware of suitable supercomputers available from Cray or IBM or other cloud computing based vendors such as Microsoft, Amazon. In this regard, although the full training-prediction algorithm may be executable on a pure CPU-based machine, in one or some embodiments, such ML-based algorithms may be executed in one or more GPUs, which are expected to have tens or even hundreds times of acceleration.

In one or some embodiments, a current GPU implementation, the neural network model may be stored in GPU caches, and the training/prediction process may be roughly as follows: (1) using CPUs (in HPC) to read in and preprocess seismic data, including rescaling, generating masks, chopping into patches, shuffling, grouping into batches; (2) passing batches data from RAM to GPU caches; (3) computing neural network updates (in training) or output data patches (in prediction) on GPU; and (4) for prediction, post-process output patches and write to files.

The above-described techniques, and/or systems implementing such techniques, can further include hydrocarbon management based at least in part upon the above techniques, including using the one or more generated geological models in one or more aspects of hydrocarbon management. For instance, methods according to various embodiments may include managing hydrocarbons based at least in part upon the one or more generated geological models and data representations (e.g., seismic images, feature probability maps, feature objects, etc.) constructed according to the above-described methods. In particular, such methods may include drilling a well, and/or causing a well to be drilled, based at least in part upon the one or more generated geological models and data representations discussed herein (e.g., such that the well is located based at least in part upon a location determined from the models and/or data representations, which location may optionally be informed by other inputs, data, and/or analyses, as well) and further prospecting for and/or producing hydrocarbons using the well.

It is intended that the foregoing detailed description be understood as an illustration of selected forms that the invention can take and not as a definition of the invention. It is only the following claims, including all equivalents which are intended to define the scope of the claimed invention. Further, it should be noted that any aspect of any of the preferred embodiments described herein may be used alone or in combination with one another. Finally, persons skilled in the art will readily recognize that in preferred implementation, some or all of the steps in the disclosed method are performed using a computer so that the methodology is computer implemented. In such cases, the resulting physical properties model may be downloaded or saved to computer storage.

The following example embodiments of the invention are also disclosed.

Embodiment 1. A computer-implemented method for managing one or more of a well, a rod pump, or a well pad for hydrocarbon extraction, the method comprising:

• • performing machine learning in order to generate a machine-learned model using pump card data and at least one of well data, reservoir data, flowline data, user input data, or data generated from analysis of one or more of the well data, the pump card data, the reservoir data, the flowline data, or the user input data;
  • predicting, by the machine-learned model, at least one aspect of the well, the rod pump, or the well pad; and
  • using the predicted at least one aspect for hydrocarbon management.

Embodiment 2: The method of embodiment 1:

• • wherein using the machine-learned model in order to predict the at least one aspect of the well or the rod pump comprising generating a recommended action regarding the one or both of the well, the rod pump, or the well pad.

Embodiment 3: The method of embodiments 1 or 2:

• • wherein the recommended action regarding the rod pump comprises at least one of pump change, pump check, pump recalibration, modification of pump speed, or shutting in the well.

Embodiment 4: The method of embodiments 1-3:

• • wherein the recommended action comprises at least one of well stimulation, modifying operation of the rod pump in at least one aspect, or modifying a measurement tool that generates the pump card data.

Embodiment 5: The method of embodiments 1-4:

• • wherein the at least one aspect of the well and of the rod pump are predicted; and
  • wherein the at least one aspect of the well and of the rod pump are used for hydrocarbon management.

Embodiment 6: The method of embodiments 1-5:

• • wherein the user input data comprises user-generated data related to servicing or modifying one or more of the rod pump, the well, or a pump card.

Embodiment 7: The method of embodiments 1-6:

• • wherein the user input data comprises one or more of servicing the rod pump with a service rig, replacing equipment using the service rig, modifying position of the rod pump, or cleaning a wellbore.

Embodiment 8: The method of embodiments 1-7:

• • wherein the machine learning uses pump card data and the data generated from analysis of one or more of the well data, the reservoir data, the flowline data, or the user input data; and
  • wherein the data generated from analysis of the one or more of the well data, the reservoir data, the flowline data, or the user input data comprises one or both of trend data or difference data.

Embodiment 9: The method of embodiments 1-8:

• • wherein the machine learning uses pump card data and the data generated from analysis of two or more of the well data, the reservoir data, the flowline data, or the user input data; and
  • wherein the data generated from analysis of the two or more of the well data, the reservoir data, the flowline data, or the user input data comprises trend data.

Embodiment 10: The method of embodiments 1-9:

• • wherein the machine learning uses pump card data and the data generated from analysis of three or more of the well data, the reservoir data, the flowline data, or the user input data; and
  • wherein the data generated from analysis of the three or more of the well data, the reservoir data, the flowline data, or the user input data comprises trend data.

Embodiment 11: The method of embodiments 1-10:

• • wherein the machine learning is performed using the well data and the reservoir data;
  • wherein the well data comprises sensor data associated with the well, well type, geometry of the well, and well test history; and
  • wherein the reservoir data comprises at least one of pressure, volume or temperature.

Embodiment 12: The method of embodiments 1-11:

• • wherein the machine learning is performed using the pump card data generated by a pump card, maintenance data for the pump card, and calibration data for the pump card.

Embodiment 13: The method of embodiments 1-12:

• • wherein the machine-learned model generates:
  • predictive status of one or both of the well or a pump;
  • one or more recommended actions associated with the well or the pump; and
  • associated loss or gain in response to performing the one or more recommended actions.

Embodiment 14: The method of embodiments 1-13:

• • wherein the machine-learned model is retrained responsive to receipt of the user input data indicative of user input or sensor input regarding the predicted at least one aspect of the well or the rod pump.

Embodiment 15: The method of embodiments 1-14:

• • further comprising generating a user interface in order to label one or more of the well data, the reservoir data, the flowline data, or the user input data.

Embodiment 16: The method of embodiments 1-15:

• • wherein the user interface displays the pump card data relative to the one or more of the well data, the reservoir data, the flowline data, or the user input data.

Embodiment 17: The method of embodiments 1-16:

• • wherein the one or more of the well data, the reservoir data, the flowline data, or the user input data is displayed along with the pump card data in a time series with a same reference time scale.

Embodiment 18: The method of embodiments 1-17:

• • wherein the one or more of the well data, the reservoir data, the flowline data, or the user input data and the pump card data are superimposed on one another and displayed in a time series with a same reference time scale.

Embodiment 19: The method of embodiments 1-18:

• • wherein the machine-learned model is configured to perform at a series of discrete tasks including at least: (1) determining whether pump cards are providing accurate measurements; and (2) responsive to determining that the pump card is providing accurate measurement, determining at least one aspect of performance of the well.

Embodiment 20: The method of embodiments 1-19:

• • wherein the series of discrete tasks further include: (3) responsive to (1) and (2), determining one or more actions to improve operations for the hydrocarbon extraction.

Embodiment 21: A system comprising:

• • a processor; and
  • a non-transitory machine-readable medium comprising instructions that, when executed by the processor, cause a computing system to perform a method according to any of embodiments 1-20.

Embodiment 22: A non-transitory machine-readable medium comprising instructions that, when executed by a processor, cause a computing system to perform a method according to any of embodiments 1-20.

Claims

1. A computer-implemented method for managing one or more of a well, a rod pump, or a well pad for hydrocarbon extraction, the method comprising:

performing machine learning in order to generate a machine-learned model using pump card data and at least one of well data, reservoir data, flowline data, user input data, or data generated from analysis of one or more of the well data, the pump card data, the reservoir data, the flowline data, or the user input data;

predicting, by the machine-learned model, at least one aspect of the well, the rod pump, or the well pad; and

using the predicted at least one aspect for hydrocarbon management.

2. The method of claim 1, wherein using the machine-learned model in order to predict the at least one aspect of the well or the rod pump comprising generating a recommended action regarding the one or both of the well, the rod pump, or the well pad.

3. The method of claim 2, wherein the recommended action regarding the rod pump comprises at least one of pump change, pump check, pump recalibration, modification of pump speed, or shutting in the well.

4. The method of claim 3, wherein the recommended action comprises at least one of well stimulation, modifying operation of the rod pump in at least one aspect, or modifying a measurement tool that generates the pump card data.

5. The method of claim 1, wherein the at least one aspect of the well and of the rod pump are predicted; and

wherein the at least one aspect of the well and of the rod pump are used for hydrocarbon management.

6. The method of claim 1, wherein the user input data comprises user-generated data related to servicing or modifying one or more of the rod pump, the well, or a pump card.

7. The method of claim 1, wherein the user input data comprises one or more of servicing the rod pump with a service rig, replacing equipment using the service rig, modifying position of the rod pump, or cleaning a wellbore.

8. The method of claim 1, wherein the machine learning uses pump card data and the data generated from analysis of one or more of the well data, the reservoir data, the flowline data, or the user input data; and

wherein the data generated from analysis of the one or more of the well data, the reservoir data, the flowline data, or the user input data comprises one or both of trend data or difference data.

9. The method of claim 1, wherein the machine learning uses pump card data and the data generated from analysis of two or more of the well data, the reservoir data, the flowline data, or the user input data; and

wherein the data generated from analysis of the two or more of the well data, the reservoir data, the flowline data, or the user input data comprises trend data.

10. The method of claim 1, wherein the machine learning uses pump card data and the data generated from analysis of three or more of the well data, the reservoir data, the flowline data, or the user input data; and

wherein the data generated from analysis of the three or more of the well data, the reservoir data, the flowline data, or the user input data comprises trend data.

11. The method of claim 1, wherein the machine learning is performed using the well data and the reservoir data;

wherein the well data comprises sensor data associated with the well, well type, geometry of the well, and well test history; and

wherein the reservoir data comprises at least one of pressure, volume or temperature.

12. The method of claim 1, wherein the machine learning is performed using the pump card data generated by a pump card, maintenance data for the pump card, and calibration data for the pump card.

13. The method of claim 1, wherein the machine-learned model generates:

predictive status of one or both of the well or a pump;

one or more recommended actions associated with the well or the pump; and

associated loss or gain in response to performing the one or more recommended actions.

14. The method of claim 1, wherein the machine-learned model is retrained responsive to receipt of the user input data indicative of user input or sensor input regarding the predicted at least one aspect of the well or the rod pump.

15. The method of claim 1, further comprising generating a user interface in order to label one or more of the well data, the reservoir data, the flowline data, or the user input data.

16. The method of claim 15, wherein the user interface displays the pump card data relative to the one or more of the well data, the reservoir data, the flowline data, or the user input data.

17. The method of claim 16, wherein the one or more of the well data, the reservoir data, the flowline data, or the user input data is displayed along with the pump card data in a time series with a same reference time scale.

18. The method of claim 16, wherein the one or more of the well data, the reservoir data, the flowline data, or the user input data and the pump card data are superimposed on one another and displayed in a time series with a same reference time scale.

19. The method of claim 1, wherein the machine-learned model is configured to perform a series of discrete tasks including at least: (1) determining whether pump cards are providing accurate measurements; and (2) responsive to determining that the pump card is providing accurate measurement, determining at least one aspect of performance of the well.

20. The method of claim 19, wherein the series of discrete tasks further include: (3) responsive to (1) and (2), determining one or more actions to improve operations for the hydrocarbon extraction.