Methods and systems for gas-lift anomaly detection using integrated unsupervised learning models

Abstract

A method for determining an anomaly in a gas-lift system. The method includes obtaining gas-lift data from a gas-lift system and associated well, where the gas-lift system injects a gas into a fluid mixture of the well. The method further includes obtaining a set of operation parameters including an injected gas rate and an injected gas pressure. The method further includes determining, with a first machine learned model and a second machine learned model, a first and second anomaly metric each indicative of an anomaly in the gas-lift system or a flow of a production fluid from the well, respectively, based on the gas-lift data. The method further includes forming an aggregate anomaly prediction from the first anomaly metric and the second anomaly metric and adjusting, with a controller, the set of operation parameters based on, at least, the aggregate anomaly prediction.

Description

BACKGROUND

[0001] Hydrocarbon fluids can be localized in porous rock formations of the Earth's subsurface forming hydrocarbon reservoirs. Wells may be drilled for the purpose of extracting the hydrocarbon fluids from the hydrocarbon reservoirs. In some instances, hydrocarbon fluids are able to flow naturally through producing wells (i.e., wells configured for the production of hydrocarbons) because the pressure within the reservoir is high enough to propel the hydrocarbons to the surface. However, in other instances, for example, when a reservoir approaches depletion or is naturally a reservoir with low pressure, gas-lift may be utilized to produce the hydrocarbon fluids. In general, gas-lift uses a source of high-pressure gas to lower the bulk density of a fluid mixture with hydrocarbons and “lift” the mixture to the surface. In some examples, gas-lift systems use an external source of gas which is injected into production tubing located in the production well. In these examples, the injected gas mixes with the fluid mixture in the production tubing reducing the density of the fluid mixture, now entrained with injected gas, such that the overall fluid and injected gas mixture becomes light enough to flow to the surface using the available reservoir pressure. Gas-lift systems and their associated operation are instrumental in augmenting hydrocarbon recovery rates but can encounter issues. Issues during operation of a gas-lift system or process can diminish operational efficiency and, in severe cases, result in considerable damage to infrastructure (e.g., the well). As such, there exists a need to quickly determine whether an issue or operational fault has occurred in a gas-lift system or process.SUMMARY

[0002] This summary is provided to introduce a selection of concepts that are further described below in the detailed description. This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used as an aid in limiting the scope of the claimed subject matter.

[0003] Embodiments disclosed herein generally relate to a method for detecting an anomaly in a gas-lift system. The method includes obtaining gas-lift data from a gas-lift system and associated well, where the gas-lift system injects a gas into a fluid mixture of the well. The method further includes obtaining a set of operation parameters including an injected gas rate and an injected gas pressure. The method further includes determining, with a first machine learned model and a second machine learned model, a first and second anomaly metric each indicative of an anomaly in the gas-lift system or a flow of a production fluid from the well, respectively, based on the gas-lift data, where the production fluid includes the fluid mixture and the injected gas. The method further includes forming an aggregate anomaly prediction from the first anomaly metric and the second anomaly metric and adjusting, with a controller, the set of operation parameters based on, at least, the aggregate anomaly prediction.

[0004] Embodiments disclosed herein generally relate to a system that includes a gas-lift system that injects a gas into a fluid mixture of a well and a controller that can configure one or more configurable parameters of the gas-lift system and well. The one or more configurable parameters are included in a set of operation parameters. The controller is configured to: obtain gas-lift data from the gas-lift system and the well; determine, with a first machine learned model and a second machine learned model, a first and second anomaly metric each indicative of a presence of an anomaly in the gas-lift system or a flow of a production fluid from the well, respectively, based on the gas-lift data, where the production fluid includes the fluid mixture and the injected gas; form an aggregate anomaly prediction from the first anomaly metric and the second anomaly metric; and adjust the set of operation parameters based on, at least, the aggregate anomaly prediction.

[0005] Embodiments disclosed herein generally relate to a non-transitory computer-readable memory with computer-executable instructions stored thereon that, when executed on a processor, cause the processor to perform certain steps. The steps include obtaining gas-lift data from a gas-lift system and associated well and obtaining a set of operation parameters including an injected gas rate and an injected gas pressure. The steps further include determining, with a first machine learned model and a second machine learned model, a first and second anomaly prediction metric each indicative of an anomaly in the gas-lift system or flow of a production fluid from the well, respectively, based on the gas-lift data. The steps further include forming an aggregate anomaly prediction from the first predicted anomaly metric and the second anomaly metric and adjusting, with a controller, the set of operation parameters based on, at least, the aggregate anomaly prediction.

[0006] Other aspects and advantages of the claimed subject matter will be apparent from the following description and the appended claims.BRIEF DESCRIPTION OF DRAWINGS

[0007] Specific embodiments of the disclosed technology will now be described in detail with reference to the accompanying figures. Like elements in the various figures are denoted by like reference numerals for consistency. The sizes and relative positions of elements in the drawings are not necessarily drawn to scale. For example, the shapes of various elements and angles are not necessarily drawn to scale, and some of these elements may be arbitrarily enlarged and positioned to improve drawing legibility. Further, the particular shapes of the elements as drawn are not necessarily intended to convey any information regarding the actual shape of the particular elements and have been solely selected for ease of recognition in the drawing.

[0008] FIG. 1 depicts a well environment with a gas-lift system in accordance with one or more embodiments.

[0009] FIG. 2 depicts an input-output functional relationship through a pair of machine-learned models.

[0010] FIG. 3 depicts a system in accordance with one or more embodiments.

[0011] FIG. 4 depicts an optimization system in accordance with one or more embodiments.

[0012] FIG. 5 depicts a flowchart in accordance with one or more embodiments.

[0013] FIG. 6 depicts a flowchart in accordance with one or more embodiments.

[0014] FIG. 7 depicts a system in accordance with one or more embodiments.DETAILED DESCRIPTION

[0015] In the following detailed description of embodiments of the disclosure, numerous specific details are set forth in order to provide a more thorough understanding of the disclosure. However, it will be apparent to one of ordinary skill in the art that the disclosure may be practiced without these specific details. In other instances, well-known features have not been described in detail to avoid unnecessarily complicating the description.

[0016] Throughout the application, ordinal numbers (e.g., first, second, third, etc.) may be used as an adjective for an element (i.e., any noun in the application). The use of ordinal numbers is not to imply or create any particular ordering of the elements nor to limit any element to being only a single element unless expressly disclosed, such as using the terms “before,”“after,”“single,” and other such terminology. Rather, the use of ordinal numbers is to distinguish between the elements. By way of an example, a first element is distinct from a second element, and the first element may encompass more than one element and succeed (or precede) the second element in an ordering of elements.

[0017] It is to be understood that the singular forms “a,”“an,” and “the” include plural referents unless the context clearly dictates otherwise. For example, a “sensor” may include any number of “sensors” without limitation.

[0018] Terms such as “approximately,”“substantially,” etc., mean that the recited characteristic, parameter, or value need not be achieved exactly, but that deviations or variations, including for example, tolerances, measurement error, measurement accuracy limitations and other factors known to those of skill in the art, may occur in amounts that do not preclude the effect the characteristic was intended to provide.

[0019] It is to be understood that one or more of the steps shown in the flowcharts may be omitted, repeated, and / or performed in a different order than the order shown. Accordingly, the scope disclosed herein should not be considered limited to the specific arrangement of steps shown in the flowcharts.

[0020] Although multiple dependent claims are not introduced, it would be apparent to one of ordinary skill that the subject matter of the dependent claims of one or more embodiments may be combined with other dependent claims.

[0021] In the following description of FIGS. 1-7, any component described with regard to a figure, in various embodiments disclosed herein, may be equivalent to one or more like-named components described with regard to any other figure. For brevity, descriptions of these components will not be repeated with regard to each figure. Thus, each and every embodiment of the components of each figure is incorporated by reference and assumed to be optionally present within every other figure having one or more like-named components. Additionally, in accordance with various embodiments disclosed herein, any description of the components of a figure is to be interpreted as an optional embodiment which may be implemented in addition to, in conjunction with, or in place of the embodiments described with regard to a corresponding like-named component in any other figure.

[0022] In general, embodiments of the disclosure include systems and methods for determining, at least, an anomaly in a gas-lift system or process. In some embodiments, a detected anomaly further undergoes a root cause analysis and can be assigned a class representative of a type of anomaly or issue in the gas-lift system or process. In one or more embodiments, anomaly detecting is performed on the gas-lift system or process in real time given one or more measurements of the gas-lift system or process such as gas injection flow rate, temperature, pressure, etc., where such measurements can be collected as gas-lift data.

[0023] In one or more embodiments, an anomaly in the gas-lift system or process is detected using two machine learning (ML) models based on the gas-lift data. In particular, the two ML models need not be trained using labelled data. That is, borrowing language commonly employed in the literature, the two ML models may be said to be “unsupervised” learning models. In one or more embodiments, the two ML models are an isolation forest (IF) and a one-class support vector machine (1CSVM). These ML models will be described in greater detail later in the instant disclosure.

[0024] In one or more embodiments, the two ML models each detect an anomaly in a gas-lift system or process based on the gas-lift data, resulting in two anomaly predictions that are aggregated according to an aggregation function to form an aggregated anomaly prediction (“aggregate anomaly prediction”). As will be demonstrated later, each of the ML models is capable of outputting quantitative information, such as a metric (e.g., a score) that can represent, for example, a level of divergence of a data point from those data points considered “normal” or a confidence level in a determination that a data point is anomalous. That is, the ML models need not only output a binary classification of “anomalous” and “non-anomalous” (i.e., “normal”). As a result, and as will be described in greater detail later in the instant disclosure, the two ML models can be uniquely integrated, for example, by forming an aggregate anomaly prediction dependent on their respective quantitative outputs in view of the gas-lift data.

[0025] In one or more embodiments, the two ML models may further be informed by a set of operation parameters that define configurable aspects of a well and accompanying gas-lift system. For example, the set of operation parameters may include well control parameters and gas-lift parameters, where the well control parameters define the state of hardware governing fluid flow in the well system or along any portion of the pipeline (e.g., a valve configured to be opened or closed, or partially closed) and the gas-lift parameters define the state of hardware governing aspects of a gas-lift process (e.g., gas injection rate). The collection of well control parameters and gas-lift parameters are referred to herein as a set of operation parameters. In one or more embodiments, based on the aggregate anomaly prediction, a command may be sent to update the set of operation parameters to new values or states to achieve a particular goal. Goals include, but are not limited to: optimizing the production of hydrocarbons from a well; operating the gas-lift system at a most economical setting in view of a desired hydrocarbon production rate; and restoring operation of the gas-lift system to a “normal” state in response to detecting an anomaly. With respect to restoring operation of a gas-lift system to a normal state in response to detecting an anomaly, embodiments disclosed herein may further produce a recommendation, or a recommended “treatment” (e.g., an action or sequence of actions), where application of the recommendation is expected to restore the operation of a gas-lift system to the normal state.

[0026] In general, a gas-lift system or process works to aid or increase hydrocarbon production in hydrocarbon production wells by injecting high-pressure gas, from the surface, down a casing annulus or coil tubing into fluids disposed in the production tubing. The fluid density and hydrostatic pressure of the fluid are reduced by the introduction of the injected gas thereby allowing the in-situ reservoir pressure to lift the lightened fluids. In one or more embodiments, a gas-lift system includes a source of gas and a gas pump. In other embodiments, a source of gas and gas pump (or other means of supplying pressurized gas) are external to a gas-lift system and the gas-lift system receives a high-pressure gas. Hereafter, a gas-lift system refers to the set of components and mechanical or electrical devices used in a gas-lifting operation.

[0027] In accordance with one or more embodiments, systems of the instant disclosure include a well site with a gas-lift system, including an oil and gas well (“well”) with access to a subsurface formation containing hydrocarbons or reservoir (“reservoir”) and a source of gas to be used for gas-lifting. In one or more embodiments, an external source of gas may be accessed by a gas-lift system including a gas pump. In one or more embodiments, the injected gas is nitrogen. The operation of the gas-lift system, including the source of gas and gas injection rate, is defined by one or more gas-lift parameters. The result of injecting gas into the well is sensitive to a number of factors, including the physical properties or characteristics of the well, the reservoir, and the various fluids that may be present in either the well or reservoir. The result of injecting gas into the well is further sensitive to the values or states of the parameters of the gas-lift parameters. In one or more embodiments, the set of gas-lift parameters is adjusted, automatically and in real time, to optimize gas-lifting via injection of gas into the well (e.g., using a controller and / or gas-lift optimization system described below).

[0028] FIG. 1 shows a schematic diagram in accordance with one or more embodiments. More specifically, FIG. 1 illustrates a well site with gas-lift (101) including a well (115). A gas-lift system “lifts” production fluids (103), such as oil, gas, and / or water, to a well exit (105) by lowering the density of the production fluids (103) with high-pressure gas (107). In one or more embodiments, the gas (107), commonly nitrogen, is pumped from a gas source (131) using a gas pump (133) as part of a gas-lift system and is supplied into the well (115) through a surface gas injection valve (109). In some embodiments, a gas-lift system may simply include one gas pump, though in other embodiments the gas-lift system may include several gas pumps. A gas-lift system further includes electronic controllers (not shown) which may be operated by a computer, such as the computer of FIG. 7 and its accompanying description. The gas source (131) can be of many shapes and sizes. In one or more embodiments, the gas source (131) and gas-lift system including a gas pump (133) are located near the well site (101), as illustrated in FIG. 1. In other examples, the gas source (131) and gas pump (133) may be located at some distance from the well site (101). A surface gas injection valve (109), disposed at a surface (111) and connected to an annulus (113) of the well (115), controls the flow of the gas (107) and injects the gas (107) into the annulus (113) with a user-defined pressure, which can be considered as high (e.g., 20 MPa), according to one or more embodiments. The annulus (113) consists of the space between a casing (117) and a main inner pipe (119) and is configured to isolate the gas (107), allowing the gas (107) to flow the depth of the well (115) without mixing with other fluids. In one or more embodiments, the gas may instead be injected using coiled tubing threaded into the main inner pipe (119) instead of injecting the gas through the annulus (113). Gas injection may involve defining the depth at which the gas is injected, e.g., into the main inner pipe (119), to lift the fluids. For example, injecting gas with coiled tubing may involve defining the depth of the coil tubing landing for gas injection. The depth at with gas is injected in a gas-lift operation can affect the efficiency of the gas-lift operations and the quantity and rate of hydrocarbons extracted from the well. In some instances, depth(s) of injection are considered configurable gas-lift parameters. In other instances, depth(s) of injection are fixed and temporally invariant.

[0029] In the depiction of FIG. 1, the casing (117), disposed in the well (115) against a wellbore (121) and typically formed of a durable material such as steel, extends to a depth above a reservoir (123). The casing (117) isolates the subsurface fluids and supports a wellbore (121), a drilled hole of the well (115), up to the depth above the reservoir (123). At the other end, the wellbore (121) extends through the reservoir (123) beneath the casing (117). The reservoir (123) is disposed below the surface (111) of the Earth in porous rock formations and is the source of the production fluids (103). The main inner pipe (119), disposed in the wellbore (121), extends from a well exit (105) to a depth proximate or intersecting the reservoir (123) and is a conduit for production fluids (103) to exit the well. In one or more embodiments, the main inner pipe (119) may be formed of tempered steel or equivalent. For the embodiment depicted in FIG. 1, the production fluids (103) may be oil and gas. The oil and gas flows from the reservoir (123) into the wellbore (121) and into the main inner pipe (119). If the reservoir (123) has enough pressure, the oil travels upwards in the main inner pipe (119) to the surface (111). Conversely, if the reservoir (123) does not have enough pressure to lift the oil and gas by itself, in a gas-lift system, external gas (107)—where here “external” is used to distinguish the to-be-injected gas from gases originating from the wellbore—is injected into the well (115).

[0030] In one or more implementations, at a given pressure that may be considered as high (e.g., 20 MPa), the external gas (107) enters the main inner pipe (119) through a downhole gas injection valve (127) and mixes with the production fluids (103) disposed in the main inner pipe (119). The downhole gas injection valve (127) forms a hydraulic connection between the annulus (113) and the main inner pipe (119) and is configured to allow the flow of the external gas (107) from the annulus (113) into the main inner pipe (119). In one or more embodiments, the downhole gas injection valve (127) forms a hydraulic connection between coil tubing which is disposed within the main inner pipe (119) and is configured to allow the flow of external gas (107) from the coil tubing into the main inner pipe (119). In one or more embodiments, the downhole gas injection valve (127) is composed of stainless steel or the equivalent. The pressure of the injected gas (107), combined with the lighter weight of the gas (107), lowers the density of the production fluids (103) until the mixture becomes light enough to flow towards the well exit (105) and into a production tank (129). The production tank (129) is a storage tank, disposed at the surface (111), that collects and stores the production fluids (103) after the production fluids (103) exit the well (115). In some instances, the production fluids (103) may be directly transported to (e.g., by a pipeline), or otherwise processed by, an oil and gas processing facility.

[0031] In accordance with one or more embodiments, one or more downhole sensors (135) are disposed within the well (115). The one or more downhole sensors (135) measure downhole fluid and process properties such as downhole gas injection rate, downhole production rate (which may be multiphase; i.e., flow rates for water, oil, and gas), downhole pressure, downhole temperature, and downhole gas concentration. In one or more embodiments, the downhole sensors (135) are grouped together and located on the side of the wellbore (121), as illustrated in FIG. 1. In other examples, the downhole sensors (135) may be separated and located anywhere below the surface (111). Downhole sensors (135) may further be disposed within the reservoir (123) and measure properties of the reservoir such as reservoir temperature and reservoir pressure.

[0032] In accordance with one or more embodiments, one or more surface sensors (137) are disposed at, or above, the surface (111) of the well (115). The one or more surface sensors (137) measure fluid and process properties at the surface (111) such as surface pressure, gas injection pressure, gas injection rate, and gas injection temperature. In one or more embodiments, the surface sensors (137) are grouped together and located in the vicinity of the well exit (105), as illustrated in FIG. 1. In other examples, the surface sensors (137) may be separated and located anywhere at or above the surface (111). In one or more embodiments, the downhole sensors (135) or surface sensors (137) further include an acoustic sensor, such as a geophone, acquiring acoustic data that is added to the well data. In one or more embodiments, the downhole sensors (135) or surface sensors (137) further include at least one contamination sensor, configured to detect a presence, and possibly a concentration of contaminant in a produced hydrocarbon. The data measured by the downhole sensors (135) and surface sensors (137) are recorded and / or stored using a data collection unit (139). In one or more embodiments, the data collection unit (139) is located in the vicinity of the well exit (105), as illustrated in FIG. 1. In other examples, data collection unit (139) may be located separate from the well but in communication using either wired relays or a remote signal. One or more measurements acquired using the sensors, as described above, can be collected as well data.

[0033] Configuring a gas-lift system (including, for example, the operating conditions of a gas pump (133)) for optimal gas-lift is a difficult and laborious task. Further, the state and behavior of a well site with gas-lift (101) may be transient according to conditions at the environment of the site (e.g., changing weather conditions in an offshore well) or within the reservoir (e.g., seismic activity and changing temperature and pressure). Gas-lift stability, in particular, is challenging to maintain. Gas-lift stability refers to the ability of a gas-lift system to maintain consistent and efficient operations over time without experiencing issues that could disrupt production. Instabilities in gas-lift operations can lead to reduced production rates, increased operational costs, and potential damage to equipment. To provide gas-lift stability, issues must be quickly detected and addressed. As such, embodiments disclosed herein, among other things, detect anomalies representative of issues and provide recommendations for restoring the operation of the gas-lift system to, at least, a normal or stable state.

[0034] A notable example of gas-lift instability is a slug flow, in which large pockets of gas and liquid alternate in the production tubing, causing rapid changes in pressure and flow rates. A slug flow may lead to inefficient lifting, decreased production, and wear on well equipment. The gas injection rate is typically balanced in order to create sufficient buoyancy to lift the fluids without causing excessive gas breakout at the surface. Over-injecting gas might lead to inefficient gas usage, while under-injecting gas could result in poor production rates. The gas-to-liquid ratio may also play a significant role in gas-lifted hydrocarbon production, as injecting too much gas may lead to inefficient fluid separation at the surface, while injecting too little gas may result in poor lifting performance. The opening and closing pressures of the surface gas injection valve (109) and the downhole gas injection valve (127) determine when gas injection starts and stops. Properly setting these pressures may help maintain an effective gas-lift cycle. Further, it may be advantageous, in some embodiments, for the hydrocarbon production rate itself to be adjusted according to different circumstances. For instance, the hydrocarbon production rate may be intentionally reduced while a maintenance action is performed. As another example, the hydrocarbon production rate may be desired to change based on current demand.

[0035] Though FIG. 1 depicts a well that may be considered typical for onshore wells, the elements shown and described are also generally applicable to offshore (and “rigless”) wells. Consequently, FIG. 1 should not be considered as limiting as embodiments disclosed herein can be readily applied to offshore and rigless well sites although not depicted.

[0036] Regarding the components of a well site with gas-lift, the various systems and facilities may each have associated data that describe the status of the given system or facility. Additionally, the systems and / or facilities of a well site with gas-lift can have associated parameters that define and control one or more aspects of their operation. For example, features of the well site with gas-lift may be recorded in measurements collectively referred to as “well data.” In one or more embodiments, the well data may include “gas-lift data” from the gas-lift system (e.g., gas injection rates) and “reservoir data” from the reservoir. The well data may describe both static properties of the well (i.e., properties generally invariant to the passage of time), such as its diameter, end of tubing depth, perforation depth, and geographical location (among others not listed) and dynamic properties of the well (i.e., properties that can fluctuate / change temporally) such as the wellhead pressure, gas-to-oil ratio, fluid density inside the well, fluid temperature inside the well, fluid pressure inside the well, water cut, and liquid flowing rate (among others not listed). The reservoir data may describe both static properties of the reservoir, such as its depth, extent, and geographical location (among others not listed) and dynamic properties of the reservoir such as its temperature, pressure, salinity, porosity, resistivity, radioactivity, and fluid density inside the reservoir (among others not listed).

[0037] In one or more embodiments, the external gas source—where “external” is used to distinguish the to-be-injected gas from any gas originating in the wellbore—and gas pump (and other elements, components, and or facilities of a gas-lift system) have an associated set of gas-lift parameters. The set of gas-lift parameters may include gas injection parameters that define the aspects of the gas injection into the well. The gas injection parameters may include defining a rate of gas injection, a fluid conduit medium (e.g., coil tubing or tubing annulus), and a depth at which gas is to be injected. To clarify, the selection or identification of the element(s) through which gas is injected into the well are referenced herein as a “fluid conduit medium,” and examples of fluid conduit mediums include coiled tubing and using an annular volume in the wellbore around the main pipe, i.e., a tubing annulus. A volume of injected gas may be defined by the injection parameters as well, though it is noted that a volume may be considered equivalent with a rate through integration of the rate over time. In accordance with one or more embodiments, the gas-lift parameters may include configurable parameters more commonly associated with the well rather than the gas-lift system such as valve states that control the rate of production fluids from the well.

[0038] In one aspect, embodiments disclosed herein relate to a system for determining the set of gas-lift parameters that optimize the gas-lifting at a well site with gas-lift. Optimizing the gas-lifting may refer to several goals, depending on the circumstances of the well site with gas-lift. For example, in one or more embodiments, optimizing gas-lifting may include detecting and mitigating against issues that result in decreased production or operational states that can be damaging to well equipment.

[0039] In accordance with one or more embodiments, gas-lift data from a well environment with a gas-lift system is processed with two machine learning (ML) models to, at least, detect an anomaly in the gas-lift system or process. As will be described below, these ML models are capable of operating on high-dimensional data without labelled training data. Consequently, these models can provide quick and accurate determinations allowing for real-time, or near-real time, recommendations and adjustments to be made to the operation of a gas-lift system and / or well operation to optimize or restore the gas-lift processes.

[0040] Machine learning, broadly defined, is the extraction of patterns and insights from data. The phrases “artificial intelligence,”“machine learning,”“deep learning,” and “pattern recognition” are often convoluted, interchanged, and used synonymously throughout the literature. This ambiguity arises because the field of “extracting patterns and insights from data” was developed simultaneously and disjointedly among a number of classical arts like mathematics, statistics, and computer science. For consistency, the term machine learning (ML) will be adopted herein, however, one skilled in the art will recognize that the concepts and methods detailed hereafter are not limited by this choice of nomenclature.

[0041] In accordance with one or more embodiments, the ML models are unsupervised anomaly detection models. Sometimes, in the literature, the subject of anomaly detection is further partitioned into the subgroups of outlier detection novelty detection. The primary distinction between outlier detection and novelty detection is that the training data for outlier detection is thought to contain some outliers (although these outliers are not labelled or otherwise distinguished from any other data points) and the training data for novelty detection is considered to be free of outliers (or anomalies). Herein, no distinction between outlier detection and novelty detection is made and only the more general term of anomaly detection is used. This is because, in part, the anomaly detection models specified herein are readily applicable to both tasks of outlier detection and novelty detection. For example, and as will be described later in the instant disclosure, a one-class support vector machine (1CSVM) can be defined to accept, as a hyperparameter during training, an expected proportion of outliers in a given set of training data. As such, this model generalizes to both outlier detection and novelty detection where setting such a training hyperparameter to 0 indicates the specific case of novelty detection. In accordance with one or more embodiments, the first ML model is an isolation forest (IF) and the second ML model is a one-class support vector machine (1CSVM). More details regarding these anomaly detection models are provided later in the disclosure. In general, these ML models are configured according to one or more “hyperparameters” which further describe the models. For example, hyperparameters providing further detail about the IF may include, but are not limited to, the number of decision trees in the forest and the configuration of a bootstrapping process, if used. The selection of hyperparameters may be informed through evaluation of a model performance metric (e.g., precision, recall, etc.) if a portion of labelled data is provided, however, labelled data is not a requirement to train and use these models. The processes of training, validating, and testing the ML models are described in greater detail later in the disclosure.

[0042] In accordance with one or more embodiments, FIG. 2 depicts the determination of an aggregate anomaly prediction (260) for a gas-lift system or process. As depicted in FIG. 2, two machine learning (ML) models, namely, a first ML model (240) and a second ML model (250) are each used to determine an anomaly metric. Specifically, the first ML model (240) is an isolation forest (IF) and the second ML model (250) is a one-class support vector machine (1CSVM). In one or more embodiments, the IF and 1CSVM models are specifically designated selected herein for the task of anomaly detection in gas-lift systems or processes because of their distinct operational strengths in view of the described gas-lift data. In particular, the IF is effective in identifying anomalies by isolating data points. The IF works on the principle that anomalies are few and different, such that they can be isolated more quickly than normal points. In other words, and as will be described alter, anomalies are expected to have shorter average path lengths when averaged over the decision trees of the forest. Further, the IF is particularly good at handling large datasets and can efficiently detect anomalies without the need for extensive data labelling (or any data labelling). Continuing, the 1CSVM creates a boundary in a high-dimensional space such that data points that fall outside this boundary are considered anomalies. The IF and 1CSVM can each process large amounts of data more quickly and efficiently than conventional methods, making them suitable for real-time anomaly detection tasks. By providing accurate and timely detections of anomalies, these ML models can support real-time recommendations leading to more efficient gas-lift operations and optimized hydrocarbon production. Additionally, once trained, these ML models can be continually updated and improved as new data is collected, allowing them to adapt to changing conditions in the well environment, reservoir, and gas-lift system.

[0043] As seen in FIG. 2, the first ML model (240) and the second ML model (250) process, as inputs, gas-lift data (220). In accordance with one or more embodiments, the gas-lift data (220) includes temperature data (222), pressure data (224), an injected gas rate (226), a hydrocarbon production rate (228), and in some instances, one or more derived quantities (222). The temperature data (222) can include temperature measurements acquired from one or more sensors disposed on the well (e.g., subsurface sensor) or as part of the gas-lift system. The pressure data (224) can include pressure measurements acquired from one or more sensors disposed on the well or as part of the gas-lift system. The injected gas rate (226) indicates the rate at which the gas-lift system injects the external gas into the well, as described above with respect to FIG. 1. The hydrocarbon production rate (228) represents a measurement of the hydrocarbons or production fluid produced from the well. In some instances, the gas-lift data (220) further includes one or more derived quantities (230). A derived quantity is a quantity or value developed using at least one measured value. For example, a derived quantity may include a ratio of the hydrocarbon production rate (228) to the injected gas rate (226). Another example consists of a polynomial expansion (e.g., taking the square), up to a given order, of one or more measured values. Derived quantities may be developed, tested, and / or selected as part of a feature engineering process when training the ML models. Feature engineering and training are described in greater detail later in the instant disclosure.

[0044] In accordance with one or more embodiments, and as seen in FIG. 2, each of the first ML model (240) and the second ML model (250) process the gas-lift data (220) and output an anomaly metric. Specifically, the first ML model (240) returns a first anomaly metric (245) and the second ML model (250) returns a second anomaly metric (255). In accordance with one or more embodiments, the first and second anomaly metrics (245, 255) are aggregated to form an aggregate anomaly prediction (260) using an aggregation function.

[0045] In accordance with one or more embodiments, the first ML model (240) is an isolation forest (IF) and the first anomaly metric (245) is an anomaly score based on a normalized path length of a given data point through the decision trees of the IF. The IF will be described in greater detail later in the instant disclosure, however, for now it suffices to say that an IF is composed of one or more decision trees and that a data point can traverse through each tree until, conventionally, terminating at a so-called leaf node. The number of edges (or nodes) traversed by the data point though a decision tree indicates the path length of that data point through the decision tree. The average path length across all decision trees in an isolation forest may serve as an anomaly score. However, in many instances, it is desirable to normalize the average path length in some manner because the average path length or path lengths through the individual decision trees are sensitive to the number of datapoints used to train the IF. In accordance with one or more embodiments, the anomaly score s of a data point according to a trained IF iss⁡(x,n)=2-Σt=1T⁢ht⁡(x)T⁢c⁡(n),EQ.⁢1where x is a data point being scored, n is the number of data points used while training the IF, ht(x) is the path length of data point x through decision tree t, T is the total number of decision trees in the IF, and c(n) is a function that returns the average path length of an unsuccessful search in a binary search tree given a data set of n instances. Specifically, c(n) is given asc⁡(n)=2⁢H⁡(n-1)-(2⁢(n-1) / n)EQ.⁢2where H is the harmonic number and it can be estimated by ln(i)+0.5772156649 (Euler's constant). Thus, given a data point, a trained IF (where the trained IF need not be trained using labelled training data) can return an anomaly score for that data point.In accordance with one or more embodiments, the second ML model (250) is a one-class support vector machine (1CSVM) and the second anomaly metric (255) is signed distance from a decision boundary or hyperplane that has been learned, through training, to separate normal (“inlier”) and anomalous (“outlier”) data points. Thus, depending on the sign of a distance associated with a data point, the datapoint can be classified as normal or anomalous and the absolute value of the distance represents a type of confidence that the data point is normal or is anomalous. A greater description of the 1CSVM is provided later in the instant disclosure, however, for now it suffices to say that similar to the IF, the output of the 1CSVM (used, for example, as the second anomaly metric (255)) is quantitative in accordance with one or more embodiments.With these descriptions of the first anomaly metric (245) and the second anomaly metric (255) in mind, various aggregation functions can be used to form the aggregate anomaly prediction (260). For example, in one implementation, the aggregation function forms the aggregate anomaly prediction (260) as the weighted sum of the first anomaly metric (245) and the second anomaly metric (255). In other implementations, the possible values of an instance of gas-lift data (220) are said to span a data space (e.g., hyperspace) and the data space is partitioned into two or more regimes. For example, in one or more embodiments, the data space is partitioned into a first regime and a second regime, the first and second regimes being mutually exclusive and their union exhausting the data space. In such an implementation, the first anomaly metric (245) can be selected and used as the aggregate anomaly prediction (260) when the gas-lift data (220) resides within the first regime. Similarly, in such an implementation, the second anomaly metric (255) is selected and used as the aggregate anomaly prediction (260) when the gas-lift data (220) resides within the second regime. That is, in one or more implementations, the first and second ML models (240, 250) can be said to have complementary modalities of discrimination, where one model is more apt (or more accurate) at detecting an anomaly under a certain set of conditions (i.e., encompassed and defined by a regime in the data space) relative to the other model. In other implementations, the data space is partitioned into three regimes, where, for example, the aggregate anomaly prediction (260) is set equal to the first anomaly metric (245) when the gas-lift data (220) resides within the first regime, the aggregate anomaly prediction (260) is set equal to the second anomaly metric (255) when the gas-lift data (220) resides within the second regime, and the aggregate anomaly prediction (260) is set equal to a weighted sum of the first and second anomaly metrics (245, 255) when the gas-lift data (220) resides within the third regime. In instances where the aggregation function represents a weighted sum of the first anomaly metric (245) and the second anomaly metric (255), the weights assigned to each model may be given according to a regime in which the input resides. In another example, a confidence level (e.g., the magnitude of the signed distance output by the 1CSVM) is associated with at least one of the first anomaly metric (245) and the second anomaly metric (255). In this example, the aggregation function can consist of a weighted average of the first anomaly metric (245) and the second anomaly metric (255), where the weights correspond to the confidence level of at least one of the predictions. For example, in one or more embodiments, the weight assigned to the first anomaly metric (245) and the weight assigned to the second anomaly metric (255) when the aggregation function is a weighted average is w1 and w2, respectively. Further, the confidence level associated with the first anomaly metric (245) and the confidence level associated with the second anomaly metric (255) is c1 and c2, respectively, where c1 and c2 need not sum to 1 and, in some instances, either c1 and c2 can be based on the other. Using this notation, in one or more embodiments, the weights used in the aggregation function are determined using the softmax function asw1=exp⁡(c1)exp⁡(c1)+exp⁡(c2)⁢⁢and⁢⁢w2=exp⁡(c2)exp⁡(c1)+exp⁡(c2).In one or more embodiments, the first ML model (240) and the second ML model (250) operate synergistically. That is, the first ML model (240) and the second ML model (250) do not necessarily operate independently. For example, in one or more embodiments, the first ML model (240) and the second ML model (250) operate in a hierarchical manner where one model's output informs the focus or parameter settings of the other. For example, in some embodiments, the first ML model (240) processes gas-lift data (220) to determine a first anomaly metric (245). Then, the gas-lift data (220) and the first anomaly metric (245) are processed, as inputs, by the second ML model (250) to produce the second anomaly metric (255). In this case, the second anomaly metric (255) can be directly taken as the aggregate anomaly prediction (260). In other embodiments, the first ML model (240) processes gas-lift data (220) to determine a first anomaly metric (245) and rather than passing the first anomaly metric (245) as an input to the second ML model (250), the first anomaly metric (245) is used to inform (or adjust) the parameters of the second ML model (250). For example, consider notation where the first ML model (240) is represented as a function ƒ1 that produces an output y1 given an input x, the function parameterized by parameters β1 (i.e., y1=ƒ1(x:β1)). Likewise, consider a notation where the second ML model (250) is represented as a function ƒ2 that produces an output y2 given an input x, the function parameterized by parameters β2 (i.e., y2=ƒ2(x:β2)). Thus, in some embodiments, the informed nature of the second ML model (250), being informed by the first ML model (240), can be represented mathematically as y2=ƒ2(x:β2(y1)). That is, the parameterization of the second ML model (250) is dependent on the output (i.e., first anomaly metric (245)) of the first ML model (240). In one or more embodiments, the input and / or parameterization of the first ML model (240) is based on the output of the second ML model (250) and, similarly, the input and / or parameterization of the second ML model (250) is based on the output of the first ML model (240). In these embodiments, an initial (or null) output can be used to initialize either of the first or second ML models (240, 250) and the subsequent output can be used to as input (and / or parameterization) to the other ML model. This may form an iterative process of interaction between the first ML model (240) and second ML model (250) that proceeds until a stopping criterion such as a number of iterations.In summary, the synergistic operation of the first ML model (240) and the second ML model (250) can be implemented in a variety of ways in accordance with one or more embodiments, such as using the outputs of one model as inputs or contextual modifiers (e.g., dependent parametrization) for the other and / or iterative processes that continually refines the model predictions. Further, the aggregate anomaly prediction (260) can be formed using one or more of the first and second anomaly metrics (245, 255). For example, in instances where the first and second ML models (240, 250) interact iteratively, the aggregate anomaly prediction (260) can be set equal to either the first or second anomaly metric (245, 255) upon determination of convergence in the predictions.

[0050] Benefits of synergistic operation are described with respect to specific ML model types. Consider the case where the first ML model (240) is an isolation forest (IF) and the second ML model (250) is a one-class support vector machine (1CSVM). In accordance with one or more embodiments, the IF and 1CSVM are specified for their complementary detection processes. For example, while isolation forests are generally efficient in quickly isolating anomalies, they may not always be precise in high-dimensional spaces where anomalies are not well-separated. One-class support vector machines, on the other hand, excel in these environments. By combining these two models, the methods and systems disclosed herein leverage the rapid detection capability of isolation forests and the high-dimensional precision of one-class support vector machines. Moreover, the synergistic integration of these models can be implemented either sequentially or in parallel. In a sequential setup, one model's output can refine the input for the other (as described above), enhancing the overall accuracy. In parallel processing, both models independently analyze the data, and their results are aggregated (e.g., a weighted sum) to make the final decision on anomalies. Additionally, the synergistic integration of these models allows for cross-validation and feedback. In particular, results from each model can be used to cross-validate the findings of the other. For instance, if both models independently identify the same data point as an anomaly, it can be considered with higher confidence. Conversely, if there's a discrepancy (e.g., one model identifies a data point as anomalous while the other model identifies it as normal), it can trigger a deeper analysis (e.g., a root cause analysis, described later in the instant disclosure). Additionally, because these models operate based on distinct processes, they can extract different insights from the same dataset. This dual extraction of insights provides a more comprehensive understanding of the data. Finally, embodiments disclosed herein can use adaptive thresholding strategies where the sensitivity of anomaly detection is adjusted based on the outputs of both models. Here, this is stated to be particularly useful in dynamic operational environments like gas-lift systems.

[0051] In some implementations, methods and systems of the instant disclosure are effectuated as a gas-lift monitoring system. Turning to FIG. 3, FIG. 3 depicts an instance of a gas-lift monitoring system (300) in accordance with one or more embodiments. As seen in FIG. 3, the gas-lift monitoring system (300) interacts with a well (e.g., Well A (302)) with access to a reservoir (e.g., Reservoir C (312)). One or more sensors (e.g., Sensors A (306)) are disposed on or within the well. The sensors may include one or more temperature sensors, pressure sensors, vibration sensors, and flow rate sensors (including multiphase flowrate devices (e.g., MPFM)), among other sensors.

[0052] Sensor data, and other data related to the well (e.g., Well A (302)) is collected as well data (e.g., Well Data A (304)). Additional data, associated with the reservoir (e.g., Reservoir C (312)) is collected as reservoir data (e.g., Reservoir Data C (314)). Reservoir data can be collected using one or more sensors associated with the well, one or more sensors associated with the reservoir, data collected as part of a drilling process (e.g., logging while drilling data), data collected apart from a drilling process (e.g., seismic survey), or any combination thereof.

[0053] The well (e.g., Well A (302)) is coupled with a gas-lift system (e.g., Gas-lift System B (308)) for effectuating a gas-lift process on the well. That is, the well (e.g., Well A (302)) is associated with or otherwise can access the gas-lift system. The gas-lift system can, at least, set the values of configurable parameters of a gas-lift process applied to the well. Configurable gas-lift parameters are defined as a set of gas-lift parameters (e.g., Set of Gas-lift Parameters B (310)). In some embodiments, the gas-lift system (e.g., Gas-lift System B (308)) includes a computer system that is the same as or similar to that of computer system depicted in FIG. 7.

[0054] As seen in FIG. 3, gas-lift data (e.g., Gas-lift Data D (320)) can include temperature data (e.g., Temperature Data D (322)), pressure data (e.g., Pressure Data D (324)), an injected gas rate (e.g., Injected Gas Rate D (326)), a hydrocarbon production rate (e.g., Hydrocarbon Production Rate D (328)), and, in some instances, one or more derived quantities (e.g., Derived Quantities D (330)). As depicted in FIG. 3, the gas-lift data (e.g., Gas-lift Data D (320)) is received from one or more of the well (e.g., Well A (302)), the gas-lift system (e.g., Gas-lift System B (308)), and the reservoir (e.g., Reservoir C (312)). In general, the temperature data can include temperature measurements acquired from one or more sensors disposed on the well (e.g., subsurface sensor) or as part of the gas-lift system. The pressure data can include pressure measurements acquired from one or more sensors disposed on the well or as part of the gas-lift system. The injected gas rate indicates the rate at which the gas-lift system injects the external gas into the well, as described above with respect to FIG. 1. The hydrocarbon production rate represents a measurement of the hydrocarbons or production fluid produced from the well. In some instances, the gas-lift data (e.g., Gas-lift Data D (320)) further includes one or more derived quantities (e.g., Derived Quantities D (330)). A derived quantity is a quantity or value developed using at least one measured value. For example, a derived quantity may include a ratio of the hydrocarbon production rate to the injected gas rate. Another example consists of a polynomial expansion (e.g., taking the square), up to a given order, of one or more measured values. Derived quantities may be developed, tested, and / or selected as part of a feature engineering process when training the ML models.

[0055] As understood, the gas-lift data can include quantities that can vary with time (e.g., temperature, pressure, etc.). As such, an instance (or a single “input” from the perspective of the first and second ML models (240, 250)) may be composed of the measured or known properties of well, reservoir, and gas-lift system (collected as gas-lift data) at a given time.

[0056] While not depicted in FIG. 3, the well (e.g., Well A (302)) can also be associated with one or more configurable well parameters. In one or more embodiments, and as described below, the gas-lift monitoring system (300) controls various operations of the well (e.g., Well C (410)) and the gas-lift system (308) through manipulation of their configurable parameters. That is, in one or more embodiments, the gas-lift monitoring system (300) can command one or more configurable parameters of well and gas-lift system to assume a specified value or setting. As such, in some embodiments, the gas-lift monitoring system (300) includes a computer system that is the same as or similar to that of computer system depicted in FIG. 7 with its accompanying description. In accordance with one or more embodiments, the configurable parameters of the well may include valves, such as production valve and inflow control valves near the surface, controllers of a permanent downhole monitoring system and associated devices, one or more tools for logging, one or more choke assemblies, and one or more electrical submersible pumps.

[0057] Continuing with FIG. 3, and in accordance with one or more embodiments, the gas-lift data (e.g., Gas-lift Data D (320)) is passed to or otherwise provided to the gas-lift monitoring system (300). It is noted that although FIG. 3 depicts a gas-lift system (e.g., Gas-lift System B (308)) and gas-lift monitoring system (300) as separate entities, these need not be separate entities. In one or more embodiments, the gas-lift monitoring system (300) includes a set of anomaly detection models (342). The set of anomaly detection models (342) includes the first ML model (240) and the second ML model (250). In particular, and as depicted in FIG. 3, the set of anomaly detection models (342) includes an isolation forest (344) and a one-class support vector machine (1CSVM). That is, in accordance with one or more embodiments, the first ML model (240) is an isolation forest (IF) (344) and the second ML model (250) is a one-class support vector machine (1CSVM) (346). For the remaining portion of the discussion of the gas-lift monitoring system (300), the set of anomaly detection models (342) will be discussed in terms of the IF (344) and 1CSVM (346) as opposed to the more general first and second ML models (240, 250). As discussed with respect to FIG. 2, the gas-lift data (e.g., Gas-lift Data D (320)) is processed by the set of anomaly detection models to produce an aggregate anomaly prediction (260). The aggregate anomaly prediction is used by one or more other components or elements of the gas-lift monitoring system (300) as described below.

[0058] In accordance with one or more embodiments, the gas-lift monitoring system (300) includes, or has access to, historical data (350). The historical data includes, at least, historical gas-lift data that can be used, when desired, to train (or re-train, fine-tune, etc.) the set of anomaly detection models (342). Again, it is emphasized that the set of anomaly detection models (342) do not require labelled data for training. That said, in some embodiments, the historical data (350) further includes classified anomalies (352). A classified anomaly is a data point (e.g., an instance of gas-lift data) with an associated label that indicates, at least, that the data point represents an anomaly. In some instances, the anomaly is further classified according to a type of anomaly. Types of anomalies, or classes, that may be assigned to a data point include: mulitpointing (e.g., gas is channeled through multiple valves); gas injection issue (e.g., insufficient or excessive gas injection, gas leakage in the injection lines); valve failure (e.g., gas-lift valve not opening or closing properly, premature valve opening, valve erosion, etc.); pressure anomalies (e.g., abnormal casing or wellhead pressures, sudden pressure surges or drops); temperature fluctuations (e.g., unusual temperature rise in the tubing or casing or cold spots indicating potential blockages or fluid accumulation); operational errors (e.g., incorrect valve settings leading to inefficient gas lift, mismanagement of gas injection rates); and flow irregularity (e.g., slugging, or intermittent flow of oil and gas, decreased oil production rate, decreased water cut or water production). In one more embodiments, classified anomalies (352) are manually annotated in a portion of the historical data (350) by a subject matter expert upon inspection of the gas-lift system and associated well. In other embodiments, classified anomalies (352) are annotated by inducing an anomaly (or an error) of a given type or class in a gas-lift system and associated well and recording the associated gas-lift data. In some instances, the introduction of specific anomalies (i.e., classified according to a type) into a gas-lift system and / or associated well is done in a laboratory setting. As will be described later in the instant disclosure, knowledge of the presence of anomalies and, moreover, knowledge of the class of anomaly, in at least a portion of the historical data (350), while not needed for training the set of anomaly detection models (342), can be used to classify an anomaly detected by said set of models and evaluate the performance of the set of models. The performance evaluation can be refined, wherein metrics such as accuracy, precision, recall, F1, etc. can be applied to specific types of anomalies.

[0059] The isolation forest (344) is composed of one or more decision trees. A data point, when processed by the isolation forest (344), is processed by each decision tree of the isolation forest (344). In each decision tree, the data point is filtered according to a series of partitions and can be said to take a deterministic path through the decision tree. In accordance with one or more embodiments, historical data (350)—and in particular, historical data (350) with classified anomalies (352)—is processed by a trained isolation forest (344) and analyzed to form anomaly path descriptors (354). The anomaly path descriptors (354) can be summary statistics describing the path taken by a data point, relative to other data points of the historical data (350), through the decision trees of the isolation forest (344) in view of the classified anomalies (352). For example, one anomaly path descriptor (354) can be the proportion of partitions (or segregations) taken on a datapoint with respect to a specific feature of the data point, such as temperature, relative to the total number of partitions applied to the data point. In this example, the anomaly path descriptor may indicate whether a given data point is isolated mostly according to temperature data relative to other data points in the historical data (350). The anomaly path descriptors (354) can be related to classified anomalies to develop correspondence data (374), described below. In accordance with one or more embodiments, the historical data (350) further includes maintenance logs (356) where correlations between anomalies (and their class) and their associated data points can be formed in view of the maintenance history of the well and gas-lift system. In some instances, maintenance data is also provided, as an input, to the set of anomaly detection models (342).

[0060] In accordance with one or more embodiments, the gas-lift monitoring system (300) includes a root cause analysis system (360). In response to detecting an anomaly using the set of anomaly detection models (342) the root cause analysis system (360) attempts to identify the root cause of the anomaly by identifying the relevant features of the anomalous data point that resulted in the anomalous classification and further assigning the anomaly a specific type or class, if possible. For instance, if an anomaly is detected and the root cause analysis system (360) indicates that the given data point classified as anomalous was done so based mostly the temperature data of the data point, then the root cause analysis system (360) can further annotate the anomaly with this information.

[0061] In accordance with one or more embodiments, the root cause analysis system (360) performs various processes to append information to an anomalous data point. In one or more embodiments, these processes include a path analysis (362). The path analysis identifies the path taken by the anomalous data point through each decision tree in the IF (344). Further, the path analysis determines one or more path descriptors that summarize the paths taken by the data point over the decision trees in the IF (362). The path descriptors can be the same as the anomaly path descriptors (354) previously described with respect to the historical data (350). In one or more embodiments, the root cause analysis system (360) performs a path comparison (364). The path comparison (364) consists of comparing the paths taken by the anomalous data point through the IF (344) to the path(s) of classified anomalies (352) known through the historical data (350). In one or more embodiments, the path comparison (364) computes a similarity metric between the result of the path analysis (362) (e.g., path descriptors) to the labelled anomaly path descriptors (354) from the historical data (350). In one or more embodiments, the similarity metric is a norm, such as the L2 norm (also known as the Euclidean norm), between the path descriptors and the historically determined and labelled (i.e., associated with a particular anomaly) anomaly path descriptors (354). In these embodiments, the detected anomaly can be assigned the same class (or anomaly type) as the classified anomaly (352) with the most similar anomaly path descriptors (354) according to the similarity metric. In this way, the root cause analysis system (360) can determine a specific anomaly type (or class) for a detected anomaly (where the anomaly was detected using the set of anomaly detection models (342)), even when the set of anomaly detection models are unsupervised (i.e., trained without explicitly labelled data) and, by definition, do not return an anomaly class (or type). Thus, embodiments disclosed describe, at least, a method for classifying, with a specific class, a detected anomaly using unsupervised learning methods and very little labelled data. That is, in instances where labelled data is sparse, anomalies can be detected generally (i.e., a data point is determined to be “anomalous” or “normal”) with high accuracy using models trained using only datasets with labels (i.e., unsupervised) and further classified according to a specific class (or type of anomaly) using the sparse labelled data through the use of a similarity metric and decision tree-based path descriptors.

[0062] In accordance with one or more embodiments, the root cause analysis system (360) also includes a feature importance (368) analysis. The feature importance (368) analysis identifies the path taken by each data point through each decision tree in the IF (344) and determines which features most frequently contribute to the isolation of data points. Thus, the feature importance (367) can be used to infer the nature, or class, of an anomaly through evaluation of those features, e.g., temperature, pressure, etc. that result in its isolation. For example, if temperature readings are a common factor in the isolation paths of anomalies, it might suggest temperature-related issues.

[0063] Finally, and in accordance with one or more embodiments, the root cause analysis system (360) further includes, or performs, a boundary analysis (369). As previously discussed, the 1CSVM (346) creates a boundary in a high-dimensional space such that data points that fall outside this boundary are considered anomalies. Additionally, a distance of a data point from the boundary can be determined. In one or more embodiments, the distances is a signed distance where the sign of the distance of the data point from the boundary indicates whether the data point is determined to be an anomaly and the and the absolute value of the distance represents a type of confidence that the data point is normal or is anomalous. In one or more embodiments, the path analysis (362) and path comparison (364) is taken in view of the confidence level that a data point is (or isn't) anomalous, where the confidence level is determined using the boundary analysis (369). In this way, the 1CSVM (346) can complement the IF (344) by providing a boundary-based perspective (i.e., boundary analysis (369)) when the paths of a data point through the decision trees of the IF (344) are considered by the root cause analysis system (360).

[0064] In accordance with one or more embodiments, the gas-lift monitoring system (300) further includes a recommendation system (370). In one or more embodiments, the recommendation system (370) includes a set of operational protocols (372). The operational protocols (372) consist of a list or database of best practices or indicated procedures for addressing different types of anomalies and issues in gas-lift systems. For example, when an anomaly is detected using the set of anomaly detection models (342), in one or more embodiments, the anomaly may further be classified (i.e., determination of a type of anomaly) using the root cause analysis system (360), and a correction procedure may be referenced using the operational protocols (372) where the operational protocols (372) list an action or sequence to action to be undertaken or performed in view of the classified anomaly.

[0065] In accordance with one or more embodiments, the recommendation system (370) may further include correspondence data (374). The correspondence data (374), similar to the operational protocols (372) consists of a list, table, or database that relates features of a datapoint (e.g., anomaly path descriptors (354)) to types of anomalies (i.e., a class). In one or more embodiments, the correspondence data (374) relates anomaly path descriptors (354) to classified anomalies (352) in the historical data (350) such that path descriptors of a new (i.e., non-historical) data point can used to determine a class of an anomaly for the new data point. An example of correspondence data (374), which relates a descriptor (e.g., an anomaly path descriptor (354), determined using path analysis (362), etc.) to a class of anomaly, and operational protocols (372), which relates a class of anomaly to one or more corrective actions, are depicted in Table I, below.TABLE IExample correspondence data and operational protocols.Correspondence DataOperational ProtocolsDescriptorGas-lift Anomaly ClassProtocol / ActionUnusual patterns inGas Injection IssueInspect and repair the gas injection lines,pressure and flow rate dataadjust gas injection rates, or recalibratepressure control systemsIrregular pressure readingsValve FailureInspect and replace faulty valves, ifand flow inconsistenciespresent. Check for erosion ormechanical wear.Sudden deviations inPressure AnomalyCheck for leaks, blockages, orpressure readingsmalfunctions in pressure controlequipment and adjust operationalparameters to stabilize pressure levelsUnusual spikes or drops inTemperature AnomalyInspect for blockages in the flow path.temperature readingsCheck for insulation issues. Recalibratetemperature sensors.Patterns that deviate fromOperational ErrorReview and adjust operational settings.standard operationalProvide training to personnel orprofilesimplement automated control systemsfor more consistent operationFluctuations in flow rateFlow IrregularityCheck for blockages, leaks, ordataequipment malfunctions affecting flowand adjust valves or pumps to regulateflow ratesAnomalies in equipmentEquipment MalfunctionConduct equipment diagnostics.performance data or sensorReplace or repair faulty components.outputsRecalibrate or replace malfunctioningsensors.Changes in flow rates andFormation IssueClean wellbore. Installation of sandpressure dropsfilters. Adjust production strategies tomitigate the impact of solidsTemperature and pressureHydrate FormationImplement hydrate prevention strategiesdata indicative of hydratesuch as adjusting temperature andformation conditionspressure conditions, using hydrateinhibitors, and / or mechanical removal ofhydrates.Irregularities in equipmentMechanical FailureConduct thorough inspections. Replaceperformance andworn components. Implementmaintenance datapreventive maintenance schedules.Anomalies in safety systemSafety Valve FailureImmediate inspection and repair offunction checks or alertssafety systems. Test and recalibratesafety devices. Review safety protocols.

[0066] In accordance with one or more embodiments, the recommendation system (370) further includes, or produces, a recommendation (378). The recommendation (378) consists of one or more actions or corrective procedures to be undertaken in view of the output of the set of anomaly detection models (342) (e.g., a detected anomaly), the identified cause and / or descriptors output by the root cause analysis system (360), and other elements of the recommendation system (370) (e.g., operational protocols (372)). In one or more embodiments, the recommendation (378) is the action, procedure, and / or protocol selected from the operational protocols (372) based on a determined class (or type) of anomaly, where the class of anomaly may be related to a descriptor of the associated data point using the correspondence data (374).

[0067] In accordance with one or more embodiments, the gas-lift monitoring system (300) further includes predictive analytics (380). In one or more embodiments, the predictive analytics (380) includes a treatment simulator (382) and an optimizer (384). The treatment simulator (382) includes a digital representation of the well, gas-lift system, and their interaction including physics-based and / or phenomenological models of the well production. The treatment simulator (382) can thus simulate the effect of different actions or recommendations applied to the well and gas-lift system.

[0068] The process of using the set of anomaly detection models (342) to determine a set of optimal operation parameters (i.e., well control parameters and gas-lift parameters) that optimize the gas-lift system according to a given performance metric using the optimizer (384) is summarized in FIG. 4. The performance metric can quantify one or more aspects of the gas-lift system and associated well such as: the production of hydrocarbons from a well; and the cost of operating the gas-lift system.

[0069] In one or more embodiments, the first ML model (240) (e.g., IF (344)) and the second ML model (250) (e.g., 1CSVM (346)), once trained (where the training may be unsupervised), may each be described as a function relating the inputs (e.g., gas-lift data) and the output (i.e., first anomaly metric (245) and the second anomaly metric (255)). That is, the ML models may each be mathematically represented as M=anomaly metric=ƒ(gas-lift data), such that, given an input of gas-lift data, the trained ML models (240, 250) may produce a predicted anomaly metric. In one or more embodiments, the anomaly metrics (e.g., first anomaly metric (245), second anomaly metric (255)) are aggregated to form an aggregate anomaly prediction (260). In accordance with one or more embodiments, a performance metric is based on the aggregate anomaly prediction (260). Herein, the performance metric is represented by the variable P and it is assumed that the given performance metric is configured such that increased values of P represent better performance (i.e., optimization can be represented as a maximization of P). Thus, representing the performance metric as a function g, the performance of the gas-lift system and associated will can be given as P=g(aggregate anomaly prediction) or P=g(first ML model(gas-lift data), second ML model(gas-lift data))=g(IF(gas-lift data), 1CSVM(gas-lift data)). Thus, with trained ML models an optimization wrapper (depicted as Block 402) is used to invert the models to determine the set of optimal operation parameters that maximize the performance P according to a performance metric. Mathematically, the optimization takes the form:arg⁢⁢maxs1⁢⁢P⁢⁢subject⁢⁢to:device⁢⁢constraints⁢EQ.⁢3where the set of operation parameters is denoted as S1. Thus, the optimization wrapper (402) maximizes the performance, according to a performance metric, over the set of operation parameters. The optimization wrapper (402), when applied to trained ML models parameterized by the set of operation parameters, returns a set of optimal operation parameters. Optimization algorithms that may be employed by the optimization wrapper (402) include, but are not limited to: genetic algorithm, Newton conjugate gradient (Newton-CG), Broyden-Fletcher-Goldfarb-Shanno (BFGS), and limited-memory BFGS (L-BFGS) algorithms.One with ordinary skill in the art will appreciate that maximization and minimization may be made equivalent through simple techniques such as negation. As such, the choice to represent the optimization as a maximization as shown in EQ. 3 does not limit the scope of the present disclosure. Whether done through minimization or maximization, the optimization wrapper (402) identifies the set (or sets) of operation parameters that optimize performance of the gas-lift system and associated well according to a given performance metric (e.g., hydrocarbon production) based on the trained ML models (240, 250).

[0071] Additionally, it is recognized that a well and gas-lift system may be subject to constraints, such as safety limits imposed on various devices of the well and gas-lift system. For example, it may be determined that in order for a gas-lift system and well to operate safely, the pressure of the injected gas, as measured by a given sensor of the gas-lift system, should not exceed a prescribed value. In FIG. 4, the constraints are referenced as device constraints. In one or more embodiments, the optimization wrapper (402) cannot elect any set of operation parameters that cause any portion of the gas-lift system (and associated systems such as the well) to exceed predefined device constraints.

[0072] Returning to FIG. 3, and in accordance with one or more embodiments, at least the aggregate anomaly prediction (260) and / or recommendation (378) are transmitted automatically and in real-time over a distributed network or through a physical mechanism for data transfer such as fiber optic cables via a command system (e.g., Command System H (390)). The command system may be a controller such as a programmable logic controller (PLC). The command system may further include a computer that is the same as or similar to that of computer system depicted in FIG. 7 with its accompanying description. Based on the aggregate anomaly prediction (260) and other outputs or products of the gas-lift monitoring system (300) (e.g., recommendation (378)), the command system (e.g., Command System H (390)) transmits a signal or command to update one or more parameter values, the one or more parameters belonging to the set of operation parameters. The command to update or modify the state of the set of operation parameters is represented by Command X (395) in FIG. 3.

[0073] In accordance with one or more embodiments, the trained first and second ML models (240, 250) are used to determine an optimal set of operation parameters using the optimizer (384) to optimally operate the gas-lift system and associated well. For example, the optimal set of operation parameters may maximize oil production from a well. Upon determining the set of optimal operation parameters, the set of optimal operation parameters may be applied to the gas-lift system and well automatically. In one or more embodiments, gas-lift performance (according to some performance metric such as oil production) is continuously monitored by at least one device (e.g., sensor) of the gas-lift system or well to ensure that the determined set of optimal operation parameters improve the performance metric.

[0074] FIG. 5 depicts the general process of training and selecting the hyperparameters (also known as model tuning) for the first and second ML models (240, 250), in accordance with one or more embodiments. The processes shown in FIG. 5 may be applied to obtain the trained the ML models (240, 250) individually, or in conjunction with one another. To start, as shown in Block 502, training data is received. The training data consists of example data points that need not be labelled. That is, the data points of the training data need not be marked as anomalous or normal or have an associated class or type of anomaly. In other words, the training of the first and second ML models (240, 250) can be done in an unsupervised manner. In one or more embodiments, the training data are acquired from historical operation data.

[0075] Keeping with FIG. 5, in one or more embodiments, the training data is preprocessed as depicted by Block 504. Preprocessing, at a minimum, includes altering the training data so that it is suitable for use with ML models. For example, removing or imputing data entries with missing values. Information surrounding the preprocessing steps is saved for potential later use. For example, if normalization is performed then a computed mean vector and variance vector are retained. This allows future training data to be preprocessed identically. Values computed and retained during preprocessing are referred to herein as preprocessing parameters. One with ordinary skill in the art will recognize that a myriad of preprocessing methods beyond numericalization, removal of modelling data entries with missing values, normalization, and imputation exist. Descriptions of a select few preprocessing methods herein do not impose a limitation on the preprocessing steps encompassed by this disclosure.

[0076] In Block 508, the hyperparameters for the first and second ML models (240, 250) are selected (e.g., number of trees in an isolation forest, strength of a 1CSVM boundary relaxation or regularization parameter). Once selected, the first and second ML models (240, 250) are trained using the training data according to Block 510. Common training techniques, such as early stopping, adaptive or scheduled learning rates, and cross-validation may be used during training without departing from the scope of this disclosure.

[0077] During training, or once trained, if some labelled data is provided the performance of the trained first and second ML models (240, 250) may be evaluated as depicted in Block 512. If no labelled data is provided, then the model performance evaluation of Block 512 can be skipped or omitted.

[0078] At Block 514, a determination is made as to whether the hyperparameters of the first and second ML models (240, 250) needs to be altered. This decision may be based on the evaluation of the models as determined in Block 512. Thus, if the performance of the models is not evaluated, e.g., omission of Block 512 because no labelled training data was provided, decision Block 514 may also be skipped or omitted. However, if a performance evaluation is available, Block 514 may be implemented as follows. If the performance of the trained first and second ML models (240, 250), as determined in Block 512, is suitable, then the trained first and second ML models (240, 250) are accepted for use in a production setting. As such, in Block 518, the trained first and second ML models (240, 250) are used in production. At Block 514, if the performance of the trained first and second ML models (240, 250) are not suitable, the hyperparameters may be altered (i.e., return to Block 508) and the training process is repeated. There are many ways to alter the hyperparameters in search of suitable trained ML model performances. These include, but are not limited to: selecting new sets of hyperparameters from previously defined sets; randomly perturbing or randomly selecting new hyperparameters; using a grid search over the available hyperparameters; and intelligently altering hyperparameters based on the observed performance of previous models (e.g., a Bayesian hyperparameter search). Once suitable performance is achieved, the training procedure is complete, and the models are used in production as depicted in Block 518.

[0079] As depicted in Block 518, the trained first and second ML models (240, 250) are used “in production”—which means, as previously stated, that the trained AI models are used to process a received input not previously seen or used during training (e.g., a real-time data point). It is emphasized that the inputs received in the production setting are preprocessed identically to the manner defined in Block 504 as denoted by the connection (522), represented as a dashed line in FIG. 5, between Blocks 518 and 504.

[0080] In accordance with one or more embodiments, the first and second ML models (240, 250) are periodically updated using the data acquired in the production setting. An update may include retraining the ML model, by reverting to Block 508, with the newly acquired data from the in-production recorded values appended to the training data. An update may also include recalculating any preprocessing parameters, again, after appending the newly acquired training data to the existing training data. Thus, embodiments of the instant disclosure implement continuous learning to account for evolving conditions or new patterns in the gas-lift system. In one or more embodiments, the first and second ML models (240, 250) are periodically retrained with the most recent data, ensuring they remain accurate and relevant.

[0081] While the various blocks in FIG. 5 are presented and described sequentially, one of ordinary skill in the art will appreciate that some or all of the blocks may be executed in different orders, may be combined or omitted, and some or all of the blocks may be executed in parallel. Furthermore, the blocks may be performed actively or passively.

[0082] The process of evaluating gas-lift data and determining an aggregate anomaly prediction using the first machine learned model (240) and the second machine learned model (250) is summarized in the flowchart of FIG. 6. In Block 602, gas-lift data from a gas-lift system and associated well is obtained. In accordance with one or more embodiments, the gas-lift system is configured to inject a gas into a fluid mixture of the well. The combination of the fluid mixture and the injected gas can be referred to as the production fluid of the well. In accordance with one or more embodiments, the gas-lift data describes characteristics of, at least, one or more of the fluid mixture, the injected gas, and the production fluid. In Block 802, a set of operation parameters related to the operation of the gas-lift system and the well are obtained. In one or more embodiments, the set of operation parameters includes an injected gas rate and a pressure of the injected gas. In general, the set of operation parameters govern the injection of the gas into the well using the gas-lift system and, in turn, govern a flow of the production fluid (e.g., adjustment of the operation parameters can directly influence the flow of the production fluid including its rate and composition, among other things, like presence, frequency, and amplitude of pressure fluctuations in the flow. In some embodiments, the set of operation parameters further includes a set of well control parameters.

[0083] In Block 606, a first machine learned model and a second machine learned model are used to determine a first anomaly metric and a second anomaly metric, respectively, based on the flow data. The first and second anomaly metrics are each indicative of an anomaly (i.e., at least the presence of an anomaly) in one or more of the gas-lift system and the flow of the production fluid. In one or more embodiments, the first machine learned model is an isolation forest (IF). In one or more embodiments, the second machine learned model is a one-class support vector machine (1CSVM). In accordance with one or more embodiments, the first and the second machine learned models are unsupervised models (i.e., trained without labelled data). In some implementations, the first and second machine learned models also process or are otherwise dependent on the set of operation parameters. In Block 608, an aggregate anomaly prediction is formed from the first anomaly metric and the second anomaly metric. In one or more embodiments, the aggregate anomaly prediction is determined by applying an aggregation function to the first and second anomaly metrics.

[0084] In Block 610, the set of operation parameters are adjusted based on the aggregate anomaly prediction. In one or more embodiments, adjustment of the set of operation parameters is made using a controller. In one or more embodiments, the set of operation parameters are adjusted to a set of optimal operation parameters, where the optimization of the operation parameters is performed through evaluation of a performance metric based on, at least, the aggregate anomaly metric. In one or more embodiments, the performance metric is quantified by a quantity (or rate) of hydrocarbon production. That is, in these embodiments, the set of optimal operation parameters maximize hydrocarbon production from the well.

[0085] Other example uses of embodiments of the instant disclosure are as follows. In some implementations, despite the first and the second machine learned models being unsupervised, a class or type of anomaly is assigned to an input based on a comparison of a path descriptor of the input to the path descriptors of known anomalies (where, at least some instances of training data are given labels, e.g., sparse labelled data). Further, in one or more embodiments, a recommendation consisting of a sequence of one or more actions is recommended based on the determined class. In some embodiments the recommendation is made based on correspondence data that relates a class of anomaly to one or more descriptors or features of the input.

[0086] Embodiments of the present disclosure may provide at least one of the following advantages. As noted, complex interactions between configurable aspects of a gas-lift system and well exist such that configuring the gas-lift system for optimal operation is a difficult task. Further, anomalies in gas-lift processes go undetected to the complex interaction of system parameters and gas-lift data. Additionally, the state and behavior of a fluid mixture in a well and reservoir accessed by the well can change with time, often requiring continual changes to maintain optimal operation of the well and gas-lift system. By continuously receiving and processing gas-lift data with the trained machine learned models disclosed herein, the well and gas-lift system can be operated in an optimal state, greatly reducing the cost and time required to identify optimal settings. Further, and as previously discussed, embodiments disclosed herein can quickly (i.e., in real-time) identify anomalies the gas-lift system and flow of production fluid from the well and recommend actions or protocols to correct the anomalies. The anomaly described herein uses a first and a second machine model which are computationally inexpensive compared to, for example, a complete physics-based simulation of the well and gas-lift systems. Further, the first and second machine learned models can provide predictions in real time and can be periodically re-trained using production data without the need to label the production data by a subject matter expert. In other words, the first and second machine learned models are always up-to-date. Additionally, embodiments disclosed herein describe methods to identify the root cause of an anomaly, including assigning an anomaly a class, and describing inputs with descriptors (e.g., path descriptors), even without the use of labelled training data or supervised machine learned models.

[0087] Embodiments disclosed herein may be implemented on a computer system. FIG. 7 is a block diagram of a computer system (702) used to provide computational functionalities associated with described algorithms, methods, functions, processes, flows, and procedures as described in the instant disclosure, according to one or more embodiments. The illustrated computer (702) is intended to encompass any computing device such as a server, desktop computer, laptop / notebook computer, wireless data port, smart phone, personal data assistant (PDA), tablet computing device, one or more processors within these devices, or any other suitable processing device such as an edge computing device, including both physical or virtual instances (or both) of the computing device. An edge computing device is a dedicated computing device that is, typically, physically adjacent to the process or control with which it interacts. For example, the AI model may be implemented on an edge computing device in order to quickly provide optimal sets of transceiver parameters and well operation parameters to associated devices or their controllers.

[0088] Additionally, the computer (702) may include a computer that includes an input device, such as a keypad, keyboard, touch screen, or other device that can accept user information, and an output device that conveys information associated with the operation of the computer (702), including digital data, visual, or audio information (or a combination of information), or a GUI.

[0089] The computer (702) can serve in a role as a client, network component, a server, a database or other persistency, or any other component (or a combination of roles) of a computer system for performing the subject matter described in the instant disclosure. In some implementations, one or more components of the computer (702) may be configured to operate within environments, including cloud-computing-based, local, global, or other environment (or a combination of environments).

[0090] At a high level, the computer (702) is an electronic computing device operable to receive, transmit, process, store, or manage data and information associated with the described subject matter. According to some implementations, the computer (702) may also include or be communicably coupled with an application server, e-mail server, web server, caching server, streaming data server, business intelligence (BI) server, or other server (or a combination of servers).

[0091] The computer (702) can receive requests over network (730) from a client application (for example, executing on another computer (702) and responding to the received requests by processing the said requests in an appropriate software application. In addition, requests may also be sent to the computer (702) from internal users (for example, from a command console or by other appropriate access method), external or third-parties, other automated applications, as well as any other appropriate entities, individuals, systems, or computers.

[0092] Each of the components of the computer (702) can communicate using a system bus (703). In some implementations, any or all of the components of the computer (702), both hardware or software (or a combination of hardware and software), may interface with each other or the interface (704) (or a combination of both) over the system bus (703) using an application programming interface (API) (712) or a service layer (713) (or a combination of the API (712) and service layer (713). The API (712) may include specifications for routines, data structures, and object classes. The API (712) may be either computer-language independent or dependent and refer to a complete interface, a single function, or even a set of APIs. The service layer (713) provides software services to the computer (702) or other components (whether or not illustrated) that are communicably coupled to the computer (702). The functionality of the computer (702) may be accessible for all service consumers using this service layer. Software services, such as those provided by the service layer (713), provide reusable, defined business functionalities through a defined interface. For example, the interface may be software written in JAVA, C++, or other suitable language providing data in extensible markup language (XML) format or another suitable format. While illustrated as an integrated component of the computer (702), alternative implementations may illustrate the API (712) or the service layer (713) as stand-alone components in relation to other components of the computer (702) or other components (whether or not illustrated) that are communicably coupled to the computer (702). Moreover, any or all parts of the API (712) or the service layer (713) may be implemented as child or sub-modules of another software module, enterprise application, or hardware module without departing from the scope of this disclosure.

[0093] The computer (702) includes an interface (704). Although illustrated as a single interface (704) in FIG. 7, two or more interfaces (704) may be used according to particular needs, desires, or particular implementations of the computer (702). The interface (704) is used by the computer (702) for communicating with other systems in a distributed environment that are connected to the network (730). Generally, the interface (704) includes logic encoded in software or hardware (or a combination of software and hardware) and operable to communicate with the network (730). More specifically, the interface (704) may include software supporting one or more communication protocols associated with communications such that the network (730) or interface's hardware is operable to communicate physical signals within and outside of the illustrated computer (702).

[0094] The computer (702) includes at least one computer processor (705). Although illustrated as a single computer processor (705) in FIG. 7, two or more processors may be used according to particular needs, desires, or particular implementations of the computer (702). Generally, the computer processor (705) executes instructions and manipulates data to perform the operations of the computer (702) and any algorithms, methods, functions, processes, flows, and procedures as described in the instant disclosure.

[0095] The computer (702) also includes a memory (706) that holds data for the computer (702) or other components (or a combination of both) that can be connected to the network (730). The memory may be a non-transitory computer readable medium. For example, memory (706) can be a database storing data consistent with this disclosure. Although illustrated as a single memory (706) in FIG. 7, two or more memories may be used according to particular needs, desires, or particular implementations of the computer (702) and the described functionality. While memory (706) is illustrated as an integral component of the computer (702), in alternative implementations, memory (706) can be external to the computer (702).

[0096] The application (707) is an algorithmic software engine providing functionality according to particular needs, desires, or particular implementations of the computer (702), particularly with respect to functionality described in this disclosure. For example, application (707) can serve as one or more components, modules, applications, etc. Further, although illustrated as a single application (707), the application (707) may be implemented as multiple applications (707) on the computer (702). In addition, although illustrated as integral to the computer (702), in alternative implementations, the application (707) can be external to the computer (702).

[0097] There may be any number of computers (702) associated with, or external to, a computer system containing computer (702), wherein each computer (702) communicates over network (730). Further, the term “client,”“user,” and other appropriate terminology may be used interchangeably as appropriate without departing from the scope of this disclosure. Moreover, this disclosure contemplates that many users may use one computer (702), or that one user may use multiple computers (702).

[0098] Although only a few example embodiments have been described in detail above, those skilled in the art will readily appreciate that many modifications are possible in the example embodiments without materially departing from this invention. Accordingly, all such modifications are intended to be included within the scope of this disclosure as defined in the following claims.

Claims

1. A method, comprising:obtaining gas-lift data from a gas-lift system and associated well, wherein the gas-lift system injects a gas into a fluid mixture of the well;obtaining a set of operation parameters comprising an injected gas rate and an injected gas pressure;determining, with a first machine learned model and a second machine learned model, a first and second anomaly metric each indicative of an anomaly in the gas-lift system or a flow of a production fluid from the well, respectively, based on the gas-lift data, wherein the production fluid comprises the fluid mixture and the injected gas;forming an aggregate anomaly prediction from the first anomaly metric and the second anomaly metric; andadjusting, with a controller, the set of operation parameters based on, at least, the aggregate anomaly prediction.

2. The method of claim 1,wherein the set of operation parameters further comprises:a set of well control parameters defining an operation of the well.

3. The method of claim 1, wherein the set of operation parameters further comprises:a valve state of a valve that control the flow of the production fluid from the well.

4. The method of claim 1, wherein the gas-lift data comprises:a temperature of the production fluid;a pressure of the production fluid;the injected gas rate; anda hydrocarbon production rate.

5. The method of claim 1:wherein the first machine learned model is an isolation forest,wherein the second machine learned model is a one-class support vector machine.

6. The method of claim 1, further comprising:determining, with an optimizer, a set of optimal operation parameters based on the aggregate anomaly prediction, wherein the set of optimal operation parameters maximize a production of hydrocarbons from the well.

7. The method of claim 5, further comprising:determining a path descriptor for the gas-lift data based on a path traversed by the gas-lift data through the isolation forest;comparing, with a similarity metric, the path descriptor to anomaly path descriptors of known and classified anomalies; anddetermining a class of an anomaly for the gas-lift data based on the comparison of the path descriptor with the anomaly path descriptors.

8. The method of claim 1, further comprising:detecting, based on the aggregate anomaly prediction, an anomaly in the gas-lift system; andrecommending a sequence of one or more actions to correct the anomaly.

9. A system, comprising:a gas-lift system that injects a gas into a fluid mixture of a well; anda controller that can configure one or more configurable parameters of the gas-lift system and well, the one or more configurable parameters comprised by a set of operation parameters, the controller configured to:obtain gas-lift data from the gas-lift system and the well;determine, with a first machine learned model and a second machine learned model, a first and second anomaly metric each indicative of a presence of an anomaly in the gas-lift system or a flow of a production fluid from the well, respectively, based on the gas-lift data, wherein the production fluid comprises the fluid mixture and the injected gas;form an aggregate anomaly prediction from the first anomaly metric and the second anomaly metric; andadjust the set of operation parameters based on, at least, the aggregate anomaly prediction.

10. The system of claim 9, further comprising:wherein the set of operation parameters comprises:a gas injection rate and a pressure of the injected gas.

11. The system of claim 9, wherein the gas-lift data comprises:a temperature of the production fluid;a pressure of the production fluid;an injected gas rate; anda hydrocarbon production rate.

12. The system of claim 9:wherein the first machine learned model is an isolation forest,wherein the second machine learned model is a one-class support vector machine.

13. The system of claim 9, wherein the controller is further configured to:determine, with an optimizer, a set of optimal operation parameters based on the aggregate anomaly prediction, wherein the set of optimal operation parameters maximize a production of hydrocarbons from the well.

14. The system of claim 12, the controller further configured to:determine a path descriptor for the gas-lift data based on a path traversed by the gas-lift data through the isolation forest;compare, with a similarity metric, the path descriptor to anomaly path descriptors of known and classified anomalies;determine a class of an anomaly for the gas-lift data based on the comparison of the path descriptor with the anomaly path descriptors; andrecommend a sequence of one or more actions to correct the anomaly based on, at least, the class of the anomaly.

15. A non-transitory computer-readable memory comprising computer-executable instructions stored thereon that, when executed on a processor, cause the processor to perform steps comprising:obtaining gas-lift data from a gas-lift system and associated well;obtaining a set of operation parameters comprising an injected gas rate and an injected gas pressure;determining, with a first machine learned model and a second machine learned model, a first and second anomaly prediction metric each indicative of an anomaly in the gas-lift system or flow of a production fluid from the well, respectively, based on the gas-lift data;forming an aggregate anomaly prediction from the first predicted anomaly metric and the second anomaly metric; andadjusting, with a controller, the set of operation parameters based on, at least, the aggregate anomaly prediction.

16. The non-transitory computer-readable memory of claim 15,wherein the set of operation parameters comprises a gas injection rate and a pressure of the injected gas.

17. The non-transitory computer-readable memory of claim 15, wherein the gas-lift data comprises:a temperature of the production fluid;a pressure of the production fluid;an injected gas rate; anda hydrocarbon production rate.

18. The non-transitory computer-readable memory of claim 15:wherein the first machine learned model is an isolation forest,wherein the second machine learned model is a one-class support vector machine.

19. The non-transitory computer-readable memory of claim 15, the steps further comprising:determining, with an optimizer, a set of optimal operation parameters based on the aggregate anomaly prediction, wherein the set of optimal operation parameters maximize a production of hydrocarbons from the well.

20. The non-transitory computer-readable memory of claim 15, the steps further comprising:determining a path descriptor for the gas-lift data based on a path traversed by the gas-lift data through the isolation forest;comparing, with a similarity metric, the path descriptor to anomaly path descriptors of known and classified anomalies;determining a class of an anomaly for the gas-lift data based on the comparison of the path descriptor with the anomaly path descriptors; andrecommending a sequence of one or more actions to correct the anomaly based on, at least, the class of the anomaly.

Images