Underground wireless hydrogen sensing framework for reservoir optimization

Abstract

A method for optimizing a wireless sensor network (230) for monitoring fireflooding hydrogen production involves training a machine learning model to generate an estimate of communication performance for each of a plurality of sensors (232). The sensors are components of a wireless sensor network (230) arranged in a subterranean hydrogen reservoir (202), wherein each of the plurality of sensors (232) is configured to obtain a measurement of an environmental variable of the hydrogen reservoir (202). The method also involves using a machine learning model to minimize the cardinality of the plurality of sensors (232).

Description

Background Art

[0001] Fireflooding or in-situ combustion is a method of injecting air or oxygen into a well followed by ignition. The combustion zone propagates through the reservoir from one or more injection wells to one or more production wells. Fireflooding creates a steam and / or gas band from the combustion process and vaporized hydrocarbons that drives reservoir oil into production wells. There are three types of fireflooding processes: dry forward, dry reverse, and wet forward. In a dry forward process, air is injected and the combustion front moves from the injection well to the production well. In a dry reverse process, the combustion front moves in the opposite direction. Wet forward injection is similar to dry forward injection, where air and water are injected simultaneously or alternately.

[0002] Subsurface reservoir sensing plays a vital role in assessing the effectiveness of in-situ hydrogen generation, maximizing hydrogen recovery from fire-driven oil and gas reservoirs, and minimizing CO2. A deep understanding of in-situ reservoir conditions, such as temperature and gas content, plays a vital role in optimizing different types of subsurface fire-driven hydrogen recovery processes. Summary of the invention

[0003] In general, in one aspect, an embodiment relates to a method for optimizing a wireless sensor network for monitoring fire-driven hydrogen production, the method comprising: training a machine learning model to generate an estimate of communication performance of each of a plurality of sensors, wherein the sensors are components of a wireless sensor network disposed in an underground hydrogen reservoir, wherein each of the plurality of sensors is configured to obtain a measurement of an environmental variable of the hydrogen reservoir; and using the machine learning model to minimize a cardinality of the plurality of sensors.

[0004] In general, in one aspect, an embodiment relates to a system for monitoring fire-driven hydrogen production, the system comprising: a wireless sensor network arranged in an underground hydrogen reservoir, the sensor network comprising a plurality of sensors and at least one base station, the plurality of sensors being configured to obtain measurements of environmental variables of the hydrogen reservoir, the at least one base station being configured to wirelessly receive measurements from the sensors and forward the measurements to a ground gateway; and a processing system configured to minimize the cardinality of the plurality of sensors using a machine learning model.

[0005] In general, in one aspect, an embodiment relates to a non-transitory machine-readable medium comprising a plurality of machine-readable instructions executed by one or more processors, the plurality of machine-readable instructions causing the one or more processors to perform operations comprising: training a machine learning model to generate an estimate of communication performance of each of a plurality of sensors, wherein the sensors are components of a wireless sensor network disposed in an underground hydrogen reservoir, wherein each of the plurality of sensors is configured to obtain a measurement of an environmental variable of the hydrogen reservoir; and using the machine learning model to minimize a cardinality of the plurality of sensors.

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

[0007] Specific embodiments of the disclosed technology will now be described in detail with reference to the accompanying drawings. For consistency, similar elements in the various drawings are represented by similar reference numerals.

[0008] Figure 1 A well environment is shown according to one or more embodiments.

[0009] Figure 2 An example of a fire flooding scenario with a wireless subsurface hydrogen sensing system in accordance with one or more embodiments is shown.

[0010] Figure 3 A flow chart of a method according to one or more embodiments is shown.

[0011] Figure 4 A flow chart of a method according to one or more embodiments is shown.

[0012] Figure 5 A computer system is shown in accordance with one or more embodiments. DETAILED DESCRIPTION

[0013] In the following detailed description of the embodiments of the present disclosure, many specific details are set forth to provide a more thorough understanding of the present disclosure. However, it is obvious to those skilled in the art that the present disclosure can be practiced without these specific details. In other cases, well-known features are not described in detail to avoid unnecessarily complicating the description.

[0014] Throughout the application, ordinal numbers (e.g., first, second, third, etc.) may be used as adjectives for elements (i.e., any noun in the application). Unless explicitly disclosed, such as using the terms "before," "after," "single," and other such terms, the use of ordinal numbers does not imply or create any particular order of elements, nor does it limit any element to only a single element. Instead, the use of ordinal numbers is to distinguish between elements. As an example, a first element is different from a second element, and a first element may contain more than one element and be ranked after (or before) a second element in the ordering of elements.

[0015] In general, embodiments of the present disclosure include systems and methods for recovering hydrogen from oil and gas reservoirs using fire drive. More specifically, embodiments disclosed in this specification relate to underground wireless hydrogen sensing for hydrogen reservoir optimization. In-situ reservoir sensing via wireless hydrogen sensors can be used to enhance the recovery rate in depleted oil and gas reservoirs (e.g., depleted gas reservoirs, heavy oil reservoirs, and ultra-sour reservoirs) driven by fire. Embodiments of the present disclosure include wireless sensing devices and an optimization framework for reliable and optimized wireless sensing. The main purpose is to generate sufficient heat so that hydrogen atoms and oxygen atoms can be split and recovered from water. The remaining hydrocarbons are used as a fuel source for the process. A detailed description is provided subsequently.

[0016] Steering Figure 1 , Figure 1 Schematic diagrams are shown in accordance with one or more embodiments. Figure 1 A well environment 100 is shown, which includes an oil and gas reservoir ("reservoir") 102 located in a subterranean hydrocarbon-bearing formation 104 and a well system 106. The hydrocarbon-bearing formation 104 may include a porous or fractured rock formation located underground, below the earth's surface ("ground") 108. In the case where the well system 106 is a hydrocarbon well, the reservoir 102 may include a portion of the hydrocarbon-bearing formation 104. The hydrocarbon-bearing formation 104 and the reservoir 102 may include different rock layers with different properties (e.g., different degrees of permeability, porosity, and resistivity). In the case where the well system 106 is operated as a production well, the well system 106 may facilitate the extraction of hydrocarbons (or "products") from the reservoir 102. In the case where the well system 106 is operated as an injection well, the well system 106 may be used in a tertiary recovery method to displace produced hydrocarbons and / or maintain a pressure profile in the reservoir 102.

[0017] In some embodiments, the well system 106 includes a wellbore 120, a well subsurface system 122, a well surface system 124, and a data acquisition system 126. The data acquisition system 126 can monitor and / or control various operations of the well system 106, such as well production operations, well completion operations, well maintenance operations, and reservoir monitoring, evaluation, and development operations. In some embodiments, the data acquisition system 126 includes the following: Figure 5 and a computer system that is the same as or similar to the computer system 502 described in the accompanying description.

[0018] Wellbore 120 may include a borehole extending from the surface 108 into a target area of ​​hydrocarbon-bearing formation 104 (e.g., reservoir 102). The upper end of wellbore 120 terminating at or near the surface 108 may be referred to as the "uphole" end of wellbore 120, while the lower end of the wellbore terminating in hydrocarbon-bearing formation 104 may be referred to as the "downhole" end of wellbore 120. Wellbore 120 may facilitate the circulation of drilling fluids during drilling operations, the flow of hydrocarbon products ("products") 121 (e.g., oil and natural gas) from reservoir 102 to the surface 108 during production operations, the injection of substances (e.g., water) into hydrocarbon-bearing formation 104 or reservoir 102 during injection operations, or the communication of monitoring equipment (e.g., logging tools) with hydrocarbon-bearing formation 104 or reservoir 102 during monitoring operations (e.g., during in situ logging operations).

[0019] In some embodiments, during the operation (production / injection) of the well system 106, the data acquisition system 126 collects and records wellhead data 140 of the well system 106 and other data about downhole equipment and downhole sensors. The wellhead data 140 may include, for example, wellhead pressure (P) (e.g., including flowing wellhead pressure (FWHP)), wellhead temperature (T) (e.g., including flowing wellhead temperature), wellhead productivity (R) and / or water cut data during part or all of the life of the well 106. In some embodiments, the measurements are recorded in real time and are available for viewing or use within seconds, minutes, or hours of sensing the condition (e.g., the measurements are available within 1 hour of sensing the condition). In such embodiments, the wellhead data 140 may be referred to as "real-time" wellhead data 140. Real-time wellhead data 140 may enable a well operator to assess the relatively current state of well system 106 and make real-time decisions regarding the development of well system 106 and reservoir 102 , such as adjusting production flows from or injection flows to a well as needed.

[0020] In some embodiments, the well surface system 124 includes a wellhead 130. The wellhead 130 may include a rigid structure mounted at the "uphole" end of the wellbore 120, where the wellbore 120 terminates at or near the earth's surface 108. The wellhead 130 may include a structure for supporting (or "hanging") casing and production tubing extending into the wellbore 120. The product 121 may flow through the wellhead 130 after leaving the wellbore 120 and the well subsurface system 122 (including, for example, casing and production tubing). In some embodiments, the well surface system 124 includes a flow regulating device that can be operated to control the flow of substances into and out of the wellbore 120. For example, the well surface system 124 may include a production valve 132 that can be operated to control the flow of the product 121. For example, the production valve 132 can be fully opened to allow the product 121 to flow out of the wellbore 120 without restriction, the production valve 132 can be partially opened to partially restrict (or "throttle") the flow of the product 121 out of the wellbore 120, and the production valve 132 can be fully closed to fully restrict (or "block") the flow of the product 121 out of the wellbore 120 and through the well surface system 124.

[0021] Continue to refer to Figure 1 In some embodiments, the well surface system 124 includes a surface sensing system 134. The surface sensing system 134 may include sensor devices for sensing properties of substances (including the product 121) passing through the well surface system 124 or otherwise located in the well surface system 124. These properties may include, for example, the pressure, temperature, and flow rate of the product 121 flowing through the wellhead 130 or other conduits of the well surface system 124 after flowing out of the wellbore 120.

[0022] In some embodiments, the surface sensing system 134 includes one or more surface pressure sensors 136 operable to sense the pressure of the product 121 flowing through the well surface system 124 after flowing out of the wellbore 120. In some embodiments, the surface sensing system 134 includes a surface temperature sensor 138 operable to sense the temperature of the product 121 flowing through the well surface system 124 after flowing out of the wellbore 120. The surface temperature sensor 138 may include, for example, a wellhead temperature sensor that senses the temperature of the product 121 flowing through or otherwise located in the wellhead 130, referred to as the "wellhead temperature" (T). In some embodiments, the surface sensing system 134 includes a flow sensor 139 operable to sense the flow rate of the product 121 flowing through the well surface system 124 after flowing out of the wellbore 120.

[0023] Continue to refer to Figure 1In the case of a well completion, one or more completion operations may be performed before the well is delivered to the party responsible for production or injection. Completion operations may include casing operations, cementing operations, perforating the well, gravel packing, directional drilling, hydraulic stimulation of the reservoir area, and / or installation of a Christmas tree or wellhead assembly at the wellbore 120. Likewise, well operations may include open hole completions or cased completions. For example, an open hole completion may refer to a well that is drilled to the top of a reservoir. Thus, the well is casingd at the top of the reservoir, while the bottom of the wellbore remains open. Conversely, a cased completion may include lowering casing into the reservoir area.

[0024] Some embodiments include perforating operations. More specifically, perforating operations may include perforating the casing and cement at different locations in the wellbore 120 to allow hydrocarbons to enter the well flow from the holes formed. For example, some perforating operations include using perforating guns at different reservoir levels to create hole sections through the casing, cement, and the sides of the wellbore 120. Hydrocarbons can then enter the well flow through these hole sections. In some embodiments, the perforating operation is performed using a discharge jet or a shaped charge to penetrate the casing surrounding the wellbore 120.

[0025] In one well completion example, the sides of the wellbore 120 may need support, so casing may be run into the wellbore 120 to provide this support. After drilling, the casing ensures that the wellbore 120 does not close on itself, while also protecting the well flow from external contaminants such as water or sand. Likewise, if the formation is strong, the casing may comprise a solid steel pipe that is run into the well and will remain that way for the life of the well.

[0026] In another completion, a gravel packing operation may also be performed using a gravel packing slurry of coarse sand or gravel of appropriate particle size. In this way, the gravel packing slurry may be pumped into the wellbore 120 between the slotted liner of the casing and the side of the wellbore 120. The sand screen 110 and the gravel pack may filter sand and other debris that may enter the well flow with the hydrocarbons.

[0027] In another well operation example, the space between the casing and the untreated side of the wellbore 120 can be filled with cement to fix the casing in place. The well operation may include pumping cement slurry into the wellbore 120 to replace the existing drilling fluid and fill the space between the untreated side of the casing and the wellbore 120. The cement slurry may include a mixture of various additives and cement. After the cement slurry hardens, the cement can seal the wellbore 120 and prevent non-hydrocarbon substances from invading the well flow. In some embodiments, the cement slurry is forced to pass through the lower end of the casing and enter the annulus between the casing and the wall of the wellbore 120. More specifically, the cement plug can be used to push the cement slurry out of the casing. For example, the cement plug can be a rubber plug for separating the cement slurry from other fluids, thereby reducing pollution and maintaining predictable cement slurry performance. A displacement fluid (e.g., water) or a properly weighted drilling fluid can be pumped into the casing above the cement plug. The displacement fluid can be a pressurized fluid that is used to push the cement plug downward through the casing to squeeze the cement out of the casing outlet and return upward to the annulus.

[0028] In another well completion, a wellhead assembly may be installed at the wellhead of the wellbore 120. The wellhead assembly may be a Christmas tree (also known as a Christmas tree), which includes valves, instruments, and other components to provide surface control of the subsurface conditions of the well.

[0029] In some embodiments, the wellbore 120 includes one or more casing centralizers. For example, a casing centralizer can be a mechanical device that fixes the casing at various positions in the wellbore to prevent the casing from contacting the wellbore wall. Therefore, the casing centralizer can produce a continuous annular gap around the casing, so that the casing can be completely sealed to the wellbore wall using cement. Without casing centralizer, the cementing operation may suffer from mud channeling and poor interlayer isolation. Examples of casing centralizers may include bow spring centralizers, rigid centralizers, semi-rigid centralizers, and mold-on centralizers. In particular, the bow spring can be slightly larger than a specific wellbore to provide complete centralization in a straight well or a low-angle well. On the other hand, a rigid centralizer can be made of solid steel bars or cast iron with a fixed blade height to fit a specific casing or wellbore size. Rigid centralizers can perform well even in deviated wells, regardless of any specific lateral forces. Semi-rigid centralizers can be made from double crown bows and used as a hybrid centralizer that includes features of bow spring centralizers and rigid centralizers. The spring properties of bow spring centralizers can allow semi-rigid centralizers to be compressed for deployment in tight locations in the wellbore. Molded centralizers can have blades made from carbon fiber ceramic materials that can be applied directly to the casing surface.

[0030] In some embodiments, well intervention operations may also be performed at the well site. For example, well intervention operations may include various operations (e.g., fracturing operations, CT, flowback, separators, pumping, wellhead and Christmas tree maintenance, slickline, braided cable, coiled tubing, forced tripping, well repair, subsea well intervention, etc.) performed by one or more service entities for an oil well or gas well during the production life of the oil well or gas well. For example, well intervention activities may be similar to completion operations, well delivery operations, and / or drilling operations in order to modify the state of a well or the geometry of a well. In some embodiments, well intervention operations are used to provide well diagnostics and / or manage the production of a well. With respect to service entities, a service entity may be a company or other actor that performs one or more types of oilfield services (such as well operations) at a well site. For example, one or more service entities may be responsible for performing cementing operations in the wellbore 120 before delivering the well to a production entity.

[0031] Although Figure 1 Various configurations of hardware components and / or software components are shown, but other configurations may be used without departing from the scope of the present disclosure. Figure 1 The various components in the embodiment of the present invention can be combined to create a single component. As another example, the function performed by a single component can be performed by two or more components.

[0032] Steering Figure 2 , shows a fire flooding scenario 200 according to one or more embodiments. Fire flooding can be used to enhance hydrocarbon recovery from a reservoir 202. Air or oxygen can be injected into the oil and gas reservoir via an O2 injection well 220. A mixture of air / oxygen, hydrogen, and other gases in the reservoir 202 can be ignited, and the combustion zone can propagate through the reservoir from the injection well 220 to a production well (not shown). Fire flooding produces a large amount of steam, gases produced by the combustion process, and vaporized hydrocarbons, driving reservoir oil into the production well.

[0033] Embodiments of the present disclosure provide a system for underground wireless hydrogen sensing to enhance the recovery in depleted reservoirs of fire drive, including depleted gas reservoirs, heavy oil reservoirs, and high sulfur reservoirs. In one or more embodiments, underground sensors are used to make in-situ measurements in the reservoir and allow for accurate sensing of the presence of hydrogen. By using underground sensing, reservoir monitoring operations can be improved and facilitated.

[0034] In one or more embodiments, in-situ reservoir sensing is performed using a wireless sensor network 230 that provides data to a processing system 260 for further processing. The wireless sensor network includes sensors 232 for in-situ monitoring of hydrogen movement within the reservoir. A sufficient number of sensors 232 may be deployed within the O2 injection wellbore 220 and further deployed in an observation wellbore 210 drilled above the seal of the reservoir 202 to monitor operational integrity and conditions in the reservoir 202. The wireless sensor network also includes a base station 234 in each well.

[0035] After the sensors 232 and base stations 234 have been placed in the well, a sensor-to-sensor wireless connection can be established to transmit sensor data in a multi-hop manner between the sensors 232 until the sensor data reaches one of the base stations 234. The data collected by the sensors 232 of the wireless sensor network 230 can be transmitted to the processing system 260 via the base stations 234 and the ground gateway (not shown) for further processing. These data can be used to monitor environmental variables such as hydrogen (or more generally, chemical composition), pressure, and temperature within the reservoir 202. The collected pressure and temperature data can also be used to generate lateral profiles that can help determine anomalies.

[0036] In one or more embodiments, the sensor 232 is equipped with various sensing modalities. For example, the sensor 232 can sense ambient temperature, pressure, chemical substances (e.g., hydrogen), etc. The sensor 232 can include a microcontroller and a sensor interface to obtain sensor data. The sensor 232 can also include a near field communication (NFC) transceiver, for example, a dual-band NFC transceiver with active and passive interfaces. Other communication interfaces can be used without departing from the present disclosure.

[0037] In one or more embodiments, the sensor 232 obtains power from a base station 234 in the wellbore. The sensor 232 includes an energy management unit configured to operate using a protocol that enables the sensor 232 to obtain energy radiated from the base station 234, temporarily store the energy, and use the energy when needed. Rechargeable batteries and / or supercapacitors can be used as buffers.

[0038] Sensor 232 may facilitate use of a positioning protocol to generate location coordinates. Sensor 232 may generate a received signal strength indication (RSSI). RSSI data may be sent to base station 234. RSSI data may then be used to estimate the location of the sensor.

[0039] Once the sensors 232 have been placed in the reservoir 202, the sensors 232 can establish node-to-node connections to generate and exchange data within the reservoir 202 without human intervention. Thus, the sensors can begin to collect information such as temperature, pressure, chemical composition, and other variables in the reservoir 202.

[0040] In one or more embodiments, the base station 234 includes an antenna, a signal processing unit, a microcontroller, and an amplifier. The base station 234 can receive communications from the sensors 232 in the borehole in a multi-hop configuration via magnetic induction (MI) technology, and can transmit data and commands to the sensors via electromagnetic wave (EM) technology. The base station can communicate with a ground gateway using, for example, wired communication.

[0041] In one or more embodiments, processing system 260 is a computer system, such as that described in reference Figure 5 The processing system 260 may execute software instructions in the form of computer-readable program code to perform operations according to the embodiments, and the computer-readable program code may be stored in whole or in part, temporarily or permanently on a non-transitory computer-readable medium. The software instructions may represent reference Figure 3 and Figure 4 Describe the method.

[0042] In one or more embodiments, a machine learning framework is used to optimize the wireless sensor network 230. More specifically, the wireless sensor network 230 can be optimized to provide sufficient reservoir sensing coverage so that hydrogen recovery from fire flooding can be optimized. The deployment of sensors 232 can be optimized for the signal quality and transmission reliability of sensors 232 and the degree of power utilization. In one or more embodiments, the machine learning framework efficiently and reliably associates various influencing factors such as temperature, distance, and overall operating environment with signal and transmission quality and expected power utilization, which is different from other conventional deterministic models. In one or more embodiments, the machine learning framework is integrated into the sensor selection optimization framework to optimize reservoir coverage while minimizing the number of sensors used simultaneously. Therefore, the minimum optimal number of sensors can be selected in real time or near real time.

[0043] In one or more embodiments, the machine learning framework is based on a long short-term memory (LSTM) deep learning architecture. Other machine learning models may be used without departing from the present disclosure. After training the machine learning model, the machine learning model may be used to optimize the number of sensors to maximize the coverage area or volume in the reservoir. Figure 3 and Figure 4 A flowchart is shown to discuss the training of the machine learning model and its use after integration into the sensor selection optimization framework.

[0044] Although Figure 2 A fire flooding scenario is shown, but the described sensor network can also be used for other applications. For example, the sensor arrangement and framework can also be applied to hydrogen storage, CO2 storage and gas reservoirs.

[0045] Figure 3 and Figure 4 A flow chart is shown according to one or more embodiments. Figure 3 One or more boxes in may be composed of one or more components (e.g., Figure 2 Although Figure 3 The various blocks in the embodiment are presented and described in sequence, but those skilled in the art will appreciate that some or all of these blocks may be executed in a different order, may be combined or omitted, and may be executed in parallel. In addition, these blocks may be executed actively or passively.

[0046] Steering Figure 3 , a method 300 for training a machine learning model according to one or more embodiments is shown.

[0047] In step 302, as previously described with reference to Figure 2 As described, measurements of environmental variables are obtained by sensors of a wireless sensor network arranged in a reservoir. The measured environmental variables may include hydrogen, pressure, and temperature. Measurements may be obtained from at least some or all of the sensors of the wireless sensor network. As an alternative to using underground data, laboratory data may be used. The sensors may also provide location information. Alternatively, the sensor location may be known based on the original placement of the sensor.

[0048] In step 304, the measurements of the environmental variables are preprocessed. Outliers in the measurements may be detected and removed. Outlier detection may be performed using moving windows and / or z-score techniques.

[0049] In step 306, measurements of the communication performance of the sensor may be obtained under the circumstances of the environmental conditions prevailing during the acquisition of the environmental variables. The communication performance may be characterized based on signal quality, transmission reliability, and power utilization. The signal quality may be monitored inside the sensor, for example, by the wireless communication circuit of the sensor. The transmission reliability may be evaluated based on transmission errors, retransmissions, etc., and may also be determined inside the sensor, for example, by software executed on the sensor. Broadly speaking, the transmission reliability may be evaluated based on the continuity of data being transmitted and / or successfully received and the operational reliability of the sensor. Similarly, the power utilization may also be monitored inside the sensor, for example, by a power control module. The signal quality, transmission reliability, and power utilization may be digitally characterized, for example, in a specified range such as 1 to 10 or any other range.

[0050] In step 308, the measurements of communication performance are analyzed and processed, for example, to remove inconsistent data. Examples of inconsistent data include extremely high signal quality that is highly unlikely or abnormally high power utilization that may indicate a power control module failure.

[0051] In step 310, a machine learning model is built for predicting communication performance based on the measured values ​​of the environmental variables. Building the machine learning model may involve selecting a particular type of machine learning model.

[0052] In one embodiment, an LSTM deep network is selected to estimate signal quality, reliability, and power utilization based on measurements of environmental variables. The LSTM deep network may be selected because of its strong performance for network estimation problems covering both time series and single feature data, which can efficiently integrate these two data types. Any other type of machine learning model may be used without departing from the present disclosure.

[0053] The machine learning model can be any type of machine learning model. Examples of machine learning models that can be used include, but are not limited to, perceptrons, convolutional neural networks, deep neural networks, recurrent neural networks, support vector machines, regression trees, random forests, extreme learning machines, type I and type II fuzzy logic (T1FL / T2FL), decision trees, inductive learning models, deductive learning models, supervised learning models, unsupervised learning models, reinforcement learning models, etc. In some embodiments, two or more different types of machine learning models are integrated into a single machine learning architecture, for example, a machine learning model may include a support vector machine and a neural network.

[0054] Regarding neural networks, for example, a neural network may include one or more hidden layers, wherein the hidden layer includes one or more neurons. A neuron may be a modeling node or object roughly mimicking a neuron of a human brain. In particular, a neuron may combine a data input with a set of coefficients (i.e., a set of network weights for adjusting the data input). These network weights may amplify or reduce the value of a particular data input, thereby assigning an amount of importance to various data inputs for the task being modeled. Through machine learning, a neural network may determine which data inputs should receive a higher priority during determination of one or more specified outputs of a neural network. Likewise, these weighted data inputs may be summed so that the sum is transmitted to other hidden layers within the neural network through the activation function of the neuron. In this way, the activation function may determine whether the output of a neuron progresses to other neurons and to what extent, wherein the output may be weighted again to serve as the input of the next hidden layer.

[0055] Turning to recurrent neural networks, a recurrent neural network (RNN) can repeatedly perform a specific task for multiple data elements in an input sequence (e.g., a sequence of maintenance data or inspection data), where the output of the recurrent neural network depends on past calculations (e.g., failure to perform maintenance or resolve an unsafe condition may generate one or more dangerous events). In this way, the recurrent neural network can operate with memory or hidden unit states, which provide information used by the current unit calculation relative to the current data input. For example, a recurrent neural network can be similar to a chain structure of RNN units, where different types of recurrent neural networks can have different types of repeating RNN units. Likewise, the input sequence can be time series data, where the hidden unit state can have different values ​​at different time steps during prediction or training operations. For example, where a deep neural network can use different parameters in each hidden layer, a recurrent neural network can have common parameters in the RNN unit, which can be performed between multiple time steps. To train a recurrent neural network, supervised learning algorithms such as a backpropagation algorithm can also be used. In some embodiments, the backpropagation algorithm is a backpropagation through time (BPTT) algorithm. Likewise, the BPTT algorithm can determine gradients to update multiple hidden layers and neurons within the recurrent neural network in a manner similar to that used to train various deep neural networks. In some embodiments, a reinforcement learning algorithm such as a deep reinforcement learning algorithm is used to train the recurrent neural network. For more information on reinforcement learning algorithms, see the discussion below.

[0056] Embodiments with different types of RNNs are contemplated. For example, a classic RNN, a long short-term memory (LSTM) network, a gated recurrent unit (GRU), a stacked LSTM including multiple hidden LSTM layers (i.e., each LSTM layer includes multiple RNN units), a recurrent neural network with an attention mechanism (i.e., a machine learning model can focus attention on specific elements in an input sequence), a bidirectional recurrent neural network (e.g., a machine learning model that can be trained in two time directions simultaneously, with separate hidden layers, such as a forward layer and a reverse layer), and a multi-dimensional LSTM network, a graph recurrent neural network, a grid recurrent neural network, etc. can be used. With respect to LSTM networks, an LSTM unit can include multiple output lines that carry information vectors, such as from the output of one LSTM unit to the input of another LSTM unit. Thus, an LSTM unit can include multiple hidden layers and various point-by-point operation units that perform calculations such as vector addition.

[0057] In some embodiments, one or more ensemble learning methods may be used in conjunction with a machine learning model. For example, an ensemble learning method may use multiple types of machine learning models to obtain better predictive performance than that obtainable using a single machine learning model. In some embodiments, for example, an integrated architecture may combine multiple base models to produce a single machine learning model. An example of an ensemble learning method is a BAGGing model (i.e., BAGGing refers to a model that performs self-service sampling and aggregation operations), which combines predictions from multiple neural networks to add biases, thereby reducing the variance of a single trained neural network model. Another ensemble learning method includes a stacking method, which may involve fitting many different model types to the same data and combining the various predictions using another machine learning model.

[0058] Building a machine learning model may also involve selecting features to be considered by the machine learning model. For example, initially, the neural network may consider data for all environmental variables for training (step 312). However, the neural network may later be retrained with a reduced number of features. Specifically, after performing step 314, as discussed below, some features may be identified as being less relevant than other features. Therefore, when retraining the machine learning model, these less relevant features may be excluded.

[0059] In one or more embodiments, expert information is incorporated into the machine learning model to optimize the weighting of data in the network. The expert information may include, for example, information provided by a user, such as sensor data in which the user has detected degradation. Based on the detection of degradation, the data may be marked accordingly.

[0060] In step 312, a machine learning model is trained to estimate signal quality, reliability, and power utilization for each sensor based on environmental variables. The training is performed using previously obtained measurements of communication performance and previously obtained measurements of environmental variables as training data. Additional variables to consider for training may include oxygen injection volume throughout the injection wells and / or hydrogen production volume throughout the production wells. The location of the sensors may also be used. The training may be performed using a machine learning training algorithm.

[0061] Various types of machine learning training algorithms (e.g., back propagation algorithms) can be used to train machine learning models. In the back propagation algorithm, the gradient for each hidden layer of the neural network is calculated backwards, from the layer closest to the output layer to the layer closest to the input layer. In this way, the gradient can be calculated based on an error function (also referred to as a "loss function") using the transpose of the weights of the corresponding hidden layer. The error function can be based on various criteria, such as a mean square error function, a similarity function, etc., where the error function can be used as a feedback mechanism for adjusting the weights in the machine learning model.

[0062] Once trained, the output of the machine learning model is an estimate of the communication performance of each sensor. Signal quality, reliability, and power utilization can be estimated based on inputs including measurements of environmental variables (e.g., temperature, pressure, chemical composition) in the reservoir. The input can also include values ​​of oxygen injection volume throughout the injection well and / or hydrogen production volume throughout the production well. Sensor location can also be used as input. Signal quality can be quantified within a range, for example, using a categorical variable between 0 and 10, where 0 represents very poor quality and 10 represents the best quality. The reliability of the sensor can be quantified within a range, for example, using a categorical variable between 1 and 5, where 1 represents poor reliability and 5 represents high reliability. Other scales can be used without departing from the present disclosure. Each environmental variable is estimated at the sensor level (i.e., for each sensor of the sensor network).

[0063] In step 314, the performance of the machine learning model is evaluated. The evaluation can be performed by applying the trained machine learning model to a large amount of test data. The resulting estimated values ​​of signal quality, reliability, and power utilization can be compared with the actual values ​​of signal quality, reliability, and power utilization in the test data. In one or more embodiments, evaluating the performance of the machine learning model also includes evaluating the feature set of the machine learning model. The impact of each feature in the feature set on the estimated value generated by the machine learning model can be analyzed. For example, the Shapley value can be used to evaluate the impact.

[0064] In step 316, a determination is made as to the adequacy of the accuracy of the machine learning model. Whether the model performance is adequate can be determined based on the degree of deviation between the estimated value and the actual value. If the deviation is acceptable, the process can be terminated. Figure 3 If the deviation is unacceptable, retraining of the machine learning model may be performed beginning at step 312. Alternatively, the machine learning model itself (e.g., its hyperparameters) may be updated beginning at step 310. Alternatively, the retraining may be more comprehensive and may include the collection of new training data beginning at step 302.

[0065] Steering Figure 4 , shows a method 400 for monitoring reservoir conditions and / or determining sensor placement according to one or more embodiments. Once the Figure 3 Once you have trained a machine learning model using the method, you can execute the method at any time. For example, you can initially execute Figure 4 method to determine initial sensor placement. In addition, Figure 4 The method may be executed periodically to provide real-time feedback regarding sensor placement, for example, to detect suboptimal sensor placement as environmental conditions in the reservoir change.

[0066] In step 402, Figure 3 The acquisition of the measured values ​​in step 302 is similar, and the measured values ​​of the environmental variables are acquired by the sensors of the wireless sensor network.

[0067] In step 404, Figure 3 Similar to the preprocessing in step 402, the measured values ​​of the environmental variables are preprocessed.

[0068] Steps 402 and 404 may be performed at regular intervals and reservoir conditions may be continuously assessed by monitoring changes in environmental variables.

[0069] The detected change may trigger the execution of step 406. Alternatively, step 406 may also be executed without detecting a change.

[0070] In step 406, the selection of sensors for monitoring the reservoir is optimized. The sensor selection can be performed using a mixed integer optimization framework that minimizes the number of active sensors at each step while taking into account the power level of the sensors. The mixed integer optimization framework constrains by ensuring that the reservoir is adequately covered (e.g., full coverage of the entire reservoir or an area of ​​interest in the reservoir, for example, based on production considerations) and that the sensors covering the reservoir have sufficient signal quality to perform robustly. Throughout the optimization process, since sensors are systematically excluded from consideration, estimates of communication performance predicted using a trained machine learning model are relied upon (step 406.1). Finally, the sensor selection can be determined to minimize the number of active sensors while still providing coverage of all areas of the reservoir by at least one sensor. In other words, the cardinality of the set of sensors actively used for reservoir monitoring is minimized. Although the full set of sensors present in the reservoir may be large, if the subset is carefully selected, a much smaller subset may be sufficient to fully monitor the reservoir. To further constrain the optimization, a reliability of no less than 50% and a signal quality greater than 70% of the best possible signal can be specified. Another constraint may involve avoiding sensors with high power consumption and / or low battery status. Although it is assumed that no sensors are added and / or relocated during the current optimization cycle, additional sensors may eventually be added to the reservoir as needed, depending on the results of the sensor selection. Faulty sensors may be ignored and may be removed.

[0071] Embodiments may be implemented on a computer system. Figure 5is a block diagram of a computer system 502 for providing computing functionality associated with algorithms, methods, functions, processes, flows, and programs as described in the present disclosure, according to a specific implementation. The computer 502 shown is intended to cover any computing device, such as a high performance computing (HPC) device, a server, a desktop computer, a laptop / notebook computer, a wireless data port, a smart phone, a personal data assistant (PDA), a tablet computing device, one or more processors within these devices, or any other suitable processing device, including physical or virtual instances of computing devices (or both). In addition, the computer 502 may include a computer including the following items: an input device, such as a keypad, a keyboard, a touch screen, or other device that can accept user information; and an output device that transmits information associated with the operation of the computer 502, the information including digital data, visual or audio information (or a combination of information); or a GUI.

[0072] The computer 502 may act as a client, a network component, a server, a database or other persistent storage, or any other component (or combination of multiple roles) in a computer system for performing the subject matter described in the present disclosure. The computer 502 is shown to be communicatively coupled to a network 530. In some specific implementations, one or more components of the computer 502 may be configured to operate within an environment including a cloud-based environment, a local environment, a global environment, or other environments (or combinations of environments).

[0073] At a high level, the computer 502 is an electronic computing device that can operate to receive, transmit, process, store or manage data and information associated with the described subject matter. According to some specific implementations, the computer 502 may also include an application server, an email server, a web server, a cache server, a streaming data server, a business intelligence (BI) server or other server (or a combination of servers), or be communicatively coupled to the above-mentioned various servers.

[0074] Computer 502 may receive requests from client applications (e.g., executed on another computer 502) via network 530 and respond to the requests by processing the received requests in an appropriate software application. In addition, requests may also be sent to computer 502 from internal users (e.g., from a command console or through other suitable access methods), external or third parties, other automated applications, and any other suitable entity, individual, system, or computer.

[0075] Each component of the computer 502 can communicate using the system bus 503. In some specific implementations, any or all components (hardware or software (or a combination of hardware and software)) of the computer 502 can interact with each other or with the interface 504 (or a combination of both) on the system bus 503 using an application programming interface (API) 512 or a service layer 513 (or a combination of API 512 and service layer 513). The API 512 may include descriptions of routines, data structures, and object classes. The API 512 may be independent of or dependent on the computer language and refers to a complete interface, a single function, or even a set of APIs. The service layer 513 provides software services to the computer 502 or other components (whether or not shown) that are communicatively coupled to the computer 502. The functions of the computer 502 are accessible to all service consumers using the service layer. Software services (such as those provided by the service layer 513) provide reusable, defined business functions through defined interfaces. For example, the interface may be software written in JAVA, C++, or other suitable languages ​​that provide data in an extensible markup language (XML) format or other suitable format. Although shown as an integrated component of the computer 502, alternative implementations may show the API 512 or service layer 513 as a separate component relative to or communicatively coupled to other components of the computer 502 (whether shown or not). Furthermore, any or all portions of the API 512 or service layer 513 may be implemented as a sub-module or sub-module of another software module, enterprise application, or hardware module without departing from the scope of the present disclosure.

[0076] Computer 502 includes interface 504. Although Figure 5 504, but two or more interfaces 504 may be used depending on the particular needs, desires, or particular implementations of the computer 502. The interface 504 is used by the computer 502 to communicate with other systems in a distributed environment connected to a network 530. In general, the interface 504 includes logic that is encoded in software or hardware (or a combination of software and hardware) and is capable of operating to communicate with the network 530. More specifically, the interface 504 may include software that supports one or more communication protocols associated with the communication to enable the network 530 or the hardware of the interface to operate to transmit physical signals within and outside the computer 502 shown.

[0077] Computer 502 includes at least one computer processor 505. Although Figure 5A single computer processor 505 is shown in the figure, but two or more processors may be used depending on the particular needs, desires, or specific implementations of the computer 502. In general, the computer processor 505 executes instructions and manipulates data to perform the operations of the computer 502 and any algorithms, methods, functions, procedures, processes, and programs described in this disclosure.

[0078] Computer 502 also includes memory 506 that stores data for computer 502 or other components that may be connected to network 530 (or a combination of both). For example, memory 506 may be a database that stores data consistent with the present disclosure. Figure 5 502, two or more memories may be used depending on the particular needs, desires, or specific implementations of the computer 502 and the functionality described. Although the memory 506 is shown as an integral component of the computer 502, in alternative implementations, the memory 506 may be external to the computer 502.

[0079] Application 507 is an algorithmic software engine that provides functionality, particularly with respect to the functionality described in the present disclosure, as specifically needed, desired, or in accordance with a particular implementation of computer 502. For example, application 507 may be implemented as one or more components, modules, applications, etc. Furthermore, although illustrated as a single application 507, application 507 may be implemented as multiple applications 507 on computer 502. Additionally, although illustrated as being integral to computer 502, in alternative specific implementations, application 507 may be external to computer 502.

[0080] There may be any number of computers 502 associated with or external to the computer system containing computer 502, with each computer 502 communicating via network 530. Furthermore, the terms "client," "user," and other suitable terminology may be used interchangeably where appropriate without departing from the scope of the present disclosure. Furthermore, the present disclosure contemplates that many users may use one computer 502, or that one user may use multiple computers 502.

[0081] In some embodiments, computer 502 is implemented as part of a cloud computing system. For example, a cloud computing system may include one or more remote servers and various other cloud components, such as cloud storage units and edge servers. In particular, a cloud computing system can perform one or more computing operations without the need for direct active management of a user device or local computer system. In this way, a cloud computing system may have different functions distributed from a central server in multiple locations, which may be performed using one or more Internet connections. More specifically, a cloud computing system may operate according to one or more service models, such as infrastructure as a service (IaaS), platform as a service (PaaS), software as a service (SaaS), mobile "backend" as a service (MBaaS), serverless computing, artificial intelligence (AI) as a service (AIaaS) and / or function as a service (FaaS).

[0082] Although only a few exemplary embodiments have been described in detail above, those skilled in the art will readily appreciate that many modifications may be made in the exemplary embodiments without substantially departing from the present invention. Therefore, all such modifications are intended to be included within the scope of the present disclosure as defined by the appended claims.

Claims

1. A method for optimizing a wireless sensor network (230, 530) for monitoring fire flooding hydrogen production, the method comprising: training a machine learning model to generate an estimate of the communication performance of each sensor in a plurality of sensors (232), wherein the sensor (232) is a component of the wireless sensor network (230, 530) disposed in an underground hydrogen reservoir (102, 202), wherein each of the plurality of sensors (232) is configured to obtain a measurement of an environmental variable of the hydrogen reservoir (102, 202); and The machine learning model is used to minimize the cardinality of the plurality of sensors (232).

2. The method according to claim 1, wherein: The machine learning model is trained using measurements of the communication performance and measurements of the environmental variables of the hydrogen reservoir (102, 202).

3. The method according to claim 1 or 2, wherein: The environmental variable includes at least one selected from the group consisting of temperature, pressure, and chemical composition.

4. The method according to any one of claims 1 to 3, wherein: The communication performance includes at least one selected from the group consisting of signal quality, reliability, and power utilization.

5. The method according to any one of claims 1 to 4, wherein: The minimization of the cardinality of the plurality of sensors (232) is performed taking into account constraints, the constraints comprising: The coverage of the underground hydrogen reservoir (102, 202), At least minimum signal quality, At least minimum reliability, Avoid sensors (232) with high power consumption, and Sensors (232) with low battery status are avoided.

6. The method according to any one of claims 1 to 5, wherein: Minimization of the cardinality of the plurality of sensors (232) is performed using mixed integer optimization based on estimates of the communication performance.

7. The method according to any one of claims 1 to 6, wherein: Training the machine learning model includes: obtaining measurements of the environmental variables using the plurality of sensors (232); obtaining measurements of communication performance of the plurality of sensors (232); generating training data, the training data including the measured values ​​of the environmental variables as input and the measured values ​​of the communication performance as output; The machine learning model is trained using the training data.

8. The method according to claim 7, wherein: Generating the training data includes: The measured values ​​of the environmental variables are preprocessed to remove outliers.

9. The method according to claim 7 or 8, wherein: Generating the training data includes: Inconsistent data is removed from the measurement of the communication performance.

10. The method according to any one of claims 1 to 9, wherein: Training the machine learning model further includes: evaluating the machine learning model; and Based on the evaluation of the machine learning model, the machine learning model is retrained when performance is deemed insufficient.

11. The method according to any one of claims 1 to 10, wherein: The machine learning model is a long short-term memory (LSTM) model.

12. A system for monitoring fire flooding hydrogen production, the system comprising: A wireless sensor network (230) arranged in an underground hydrogen reservoir (102, 202), the sensor network (230) comprising: a plurality of sensors (232) configured to obtain measurements of environmental variables of the hydrogen reservoir (102, 202); At least one base station (234), the at least one base station (234) being configured to: wirelessly receiving the measurements from the sensor (232), and forwarding the measurements to a ground gateway; and A processing system (260) is configured to minimize the cardinality of the plurality of sensors (232) using a machine learning model.

13. The system according to claim 12, wherein: The sensor (232) is arranged in at least one selected from the group consisting of: an observation well (210) above the seal in the hydrogen target injection area, and O2 is injected into the wellbore (220).

14. The system according to claim 12 or 13, wherein: The plurality of sensors (232) form a multi-hop configuration to relay the measurements to the at least one base station (234).

15. A system according to any one of claims 12 to 14, wherein: Minimizing the cardinality of the plurality of sensors (234) includes: Based on the measured values ​​of the environmental variables, a machine learning model is trained to generate an estimate of the communication performance of each sensor in the plurality of sensors (232).

16. The system of claim 15, wherein: Minimizing the cardinality of the plurality of sensors (232) includes mixed integer optimization based on the estimate of the communication performance.

17. The system according to claim 15 or 16, wherein: The communication performance includes at least one selected from the group consisting of signal quality, reliability, and power utilization.

18. A system according to any one of claims 12 to 17, wherein: The environmental variable includes at least one selected from the group consisting of temperature, pressure, and chemical composition.

19. A non-transitory machine-readable medium comprising a plurality of machine-readable instructions executed by one or more processors (505), the plurality of machine-readable instructions causing the one or more processors (505) to perform operations comprising: training a machine learning model to generate an estimate of the communication performance of each sensor in a plurality of sensors (232), in, The sensor (232) is a component of a wireless sensor network (230) disposed in a subterranean hydrogen reservoir (102, 202), wherein each of the plurality of sensors (232) is configured to obtain a measurement of an environmental variable of the hydrogen reservoir (102, 202); and The machine learning model is used to minimize the cardinality of the plurality of sensors (232).

20. The non-transitory machine-readable medium of claim 19, wherein: Minimizing the cardinality of the plurality of sensors (232) is performed using mixed integer optimization based on an estimate of the communication performance.

Images