USING DISTRIBUTED SENSOR DATA TO CONTROL CLUSTER EFFICIENCY DOWNHOLE

Abstract

A system for determining real time cluster efficiency for a pumping operation in a wellbore includes a pump, a surface sensor, a downhole sensor system, and a computing device. The pump can pump slurry or diverter material in the wellbore. The surface sensor can be positioned at a surface of the wellbore to detect surface data about the pump. The downhole sensor system can be positioned in the wellbore to detect downhole data about an environment of the wellbore. The computing device can receive the surface data from the surface sensor, receive the downhole data from the downhole sensor system, apply the surface data and the downhole data to a long short-term memory (LSTM) neural network to produce a predicted cluster efficiency associated with operational settings of the pump, and control the pump using the operational settings to achieve the predicted cluster efficiency.

Description

TECHNICAL FIELD

The present disclosure relates generally to devices for use in wellbores. More specifically, but not by way of limitation, this disclosure relates to using distributed sensor data to determine and control cluster efficiency downhole.

BACKGROUND

A well such as an oil or gas well can include a wellbore drilled through a subterranean formation. The wellbore can include perforations. Fluid can be injected through the perforations to create fractures in the subterranean formation in a process referred to as hydraulic fracturing. The fractures can enable hydrocarbons to flow from the subterranean formation into the wellbore, from which the hydrocarbons can be extracted. Cluster efficiency can refer to how uniformly slurry fluid is distributed among perforation clusters. An inefficient cluster efficiency may involve one perforation cluster receiving most of the slurry fluid while another perforation cluster receives very little. Measuring or predicting cluster efficiency can be challenging as direct real time measurement may not be possible and predictive models may involve many variables with several assumptions that may not apply to a particular wellbore.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a cross-sectional view of an example of well that includes a system for pumping slurry or diverter material in a wellbore according to one example of the present disclosure.

FIG. 2 is a schematic diagram of a system for pumping slurry or diverter material in a wellbore according to one example of the present disclosure.

FIG. 3 is an example of a flowchart of a process for pumping slurry or diverter material in a wellbore according to one example of the present disclosure.

FIG. 5 is a chart of slurry flow rate with respect to different depth ranges detected by a distributed acoustic sensing system according to one example of the present disclosure.

FIG. 5 is a graphical illustration of a model that can be used in a system for pumping material in a wellbore according to one example of the present disclosure.

FIG. 6 is a graphical illustration of another model that can be used in a system for pumping material in a wellbore according to one example of the present disclosure.

FIG. 7 is a further graphical illustration of the model of FIG. 6 being used in the system for pumping material in a wellbore according to one example of the present disclosure.

DETAILED DESCRIPTION

Certain aspects and features of the present disclosure relate to determining a cluster efficiency of a hydraulic fracturing operation in a wellbore using distributed acoustic sensing (DAS) data and/or distributed temperature sensing (DTS) data, surface variable sensors to detect surface data, and a machine-learning subsystem to use the data to predict the cluster efficiency. Based on the predicted cluster efficiency, the hydraulic fracturing operation can be modified or otherwise controlled or planned.

A DAS or DTS can include distributed sensors in a wellbore or a fiber optic cable that is capable of sensing conditions downhole at different points along the fiber optic cable. The sensed data can be transmitted to a subsystem at the surface of the wellbore for analysis and interpretation. Examples of sensed data include temperature, pressure, fluid flow rate, and other characteristics of the environment in the wellbore. The data can be sensed and communicated to the subsystem substantially in real time such that the present conditions in a wellbore can be determined. Surface variable sensors can include one or more sensors positioned at or proximate to the surface of the wellbore to detect data at the surface, such as surface pressure, surface fluid flow rate, proppant rate, cluster spacing, and stage location.

An example of the machine-learning subsystem is a deep recurrent neural network (DRNN) or other long short-term memory (LSTM) machine-learning algorithm that can use the data from the distributed sensors and the surface variable sensors to predict cluster efficiency for the wellbore. In one example, training data is created from a subset of the received data and a temporal model is generated for the DRNN. Temporal values of surface data and distributed sensed data is inputted into the model and a cluster efficiency is outputted. The cluster efficiency can be used to control a slurry pumping subsystem that has flow control and directional control by determining how much slurry to pump and how fast to pump the slurry into the wellbore for the fracturing operation during stimulation. By combining surface data observations and downhole data, cluster efficiency can be better predicted and stimulation operations can be successfully controlled using real time decisions to change the completion design or to pump diverter material and the amount of diverter material to pump downhole.

Illustrative examples are given to introduce the reader to the general subject matter discussed here and are not intended to limit the scope of the disclosed concepts. The following sections describe various additional features and examples with reference to the drawings in which like numerals indicate like elements, and directional descriptions are used to describe the illustrative aspects but, like the illustrative aspects, should not be used to limit the present disclosure.

FIG. 1 is a partial cross-sectional diagram of a well 100 having layers 104a, 104b, 104c, and 104d in a production zone. The layers 104a-d can be adjacent portions of a subterranean formation through which a wellbore 102 is formed. The layers 104a-d can each have a different composition. Each layer may be treated differently with respect to material placement, and to that extent, a material placement process for well 100 may be said to be a “multi-stage” process in that each time a change in fracturing characteristics occurs during material placement due to the multiple layers, another stage in the material placement process can be reached. The well 100 can also include a computing device 110, a pump 120, and downhole sensors 130. The downhole sensors 130 can be part of a DAS or DTS that can collect, in real time, data about the environment within the well 100. Also included is a surface sensor 140, which can detect pressure, flow rates, and other data at the surface of the well 100. Although one surface sensor 140 is shown, more than one surface sensor 140 can be used.

The computing device 110 can dynamically control a pumping schedule for slurry to be pumped in well 100 by the pump 120, such as for stimulating the well 100. The computing device 110 can determine the specific pressure and the specific pump rate of slurry to pump into the wellbore 102. The computing device 110 can receive data from the downhole sensors 130 and the surface sensor 140 and use the data to determine a cluster efficiency for the well 100 that is used to control the pump 120 in how much slurry is pumped and how fast the slurry is pumped.

In some examples, the computing device 110 can be used to control the placement of a diverter. A diverter can be a fluid (e.g., polylactic acid) for temporarily reducing permeability in a layer. The diverter material injected into the subsurface formation may be, for example, a degradable polymer. Examples of different degradable polymer materials that may be used include, but are not limited to, polysaccharides; lignosulfonates; chitins; chitosans; proteins; proteinous materials; fatty alcohols; fatty esters; fatty acid salts; aliphatic polyesters; poly(lactides); poly(glycolides); poly(ϵ-caprolactones); polyoxymethylene; polyurethanes; poly(hydroxybutyrates); poly(anhydrides); aliphatic polycarbonates; polyvinyl polymers; acrylic-based polymers; poly(amino acids); poly(aspartic acid); poly(alkylene oxides); poly(ethylene oxides); polyphosphazenes; poly(orthoesters); poly(hydroxy ester ethers); polyether esters; polyester amides; polyamides; polyhydroxyalkanoates; polyethyleneterephthalates; polybutyleneterephthalates; polyethylenenaphthalenates, and copolymers, blends, derivatives, or combinations thereof. But various examples of the present disclosure are not intended to be limited thereto and that other types of diverter materials may also be used. At a certain stage, the amount of diverter placed in the wellbore may be greater than or less than at other stages. In some examples, the computing device 110 can be used to similarly control the placement of hydraulic fracturing fluid. In additional or alternative examples, the computing device 110 can optimize the pumping schedule so that less time is needed or less material is needed to achieve a desired result.

In some aspects, the pump 120 can be positioned at the surface of the well 100 for pumping a fluid into the wellbore 102. The pump 120 can be communicatively coupled to the computing device 110 for receiving instructions from the computing device 110. In additional or alternative aspects, the well 100 can include one or more pumps.

The downhole sensors 130 can be positioned in the wellbore 102 for measuring average pressures and flow rates at each stage, and communicating this data to the surface. The well 100 can include a multilateral wellbore having any number of lateral bores, each passing through any number of layers. In some examples, the wellbore can include a cement casing. The wellbore can be in any phase, including installation, completion, stimulation, and production. In some aspects, a wellbore can have a single downhole sensor.

FIG. 2 is a block diagram of an example of a system 200 for controlling a pump over time according to some aspects. In some examples, the components shown in FIG. 2 (e.g., the computing device 110 and power source 220 can be integrated into a single structure. For example, the components can be within a single housing. In other examples, the components shown in FIG. 2 can be distributed (e.g., in separate housings) and in electrical communication with each other.

The system 200 includes a computing device 110. The computing device 110 can include a processor 204, a memory 207, and a bus 206. The processor 204 can execute one or more operations for obtaining data associated with the wellbore and controlling a pump 120 to place material, such as slurry, into the wellbore. The processor 204 can execute instructions stored in the memory 207 to perform the operations. The processor 204 can include one processing device or multiple processing devices. Non-limiting examples of the processor 204 include a Field-Programmable Gate Array (“FPGA”), an application-specific integrated circuit (“ASIC”), a microprocessor, etc.

The processor 204 can be communicatively coupled to the memory 207 via the bus 206. The non-volatile memory 207 may include any type of memory device that retains stored information when powered off. Non-limiting examples of the memory 207 include electrically erasable and programmable read-only memory (“EEPROM”), flash memory, or any other type of non-volatile memory. In some examples, at least some of the memory 207 can include a non-transitory medium from which the processor 204 can read instructions. A non-transitory computer-readable medium can include electronic, optical, magnetic, or other storage devices capable of providing the processor 204 with computer-readable instructions or other program code. Non-limiting examples of a computer-readable medium include (but are not limited to) magnetic disk(s), memory chip(s), ROM, random-access memory (“RAM”), an ASIC, a configured processor, optical storage, or any other medium from which a computer processor can read instructions. The instructions can include processor-specific instructions generated by a compiler or an interpreter from code written in any suitable computer-programming language, including, for example, C, C++, C#, etc.

In some examples, the memory 207 can include computer program instructions 210 for executing a DRNN long-short term memory (LSTM) machine-learning module or other type of deep recurrent neural network. The instructions 210 can be usable for applying the DRNN to wellbore data and surface data associated with the wellbore and controlling the pump in response to a predicted value of a response variable. In some examples, the memory 207 can include stored variable values 212 and stored data 213, such as surface data and wellbore data from sensors 140, 130, respectively.

The system 200 can include a power source 220. The power source 220 can be in electrical communication with the computing device 110. In some examples, the power source 220 can include a battery or an electrical cable (e.g., a wireline). In some examples, the power source 220 can include an AC signal generator. System 200 receives input from downhole sensors 130 and surface sensor 140. System 200 in this example also includes input/output interface 232. Input/output interface 232 can connect to a keyboard, pointing device, display, and other computer input/output devices. An operator may provide input using the input/output interface 232. All or portions of input/output interface 232 may be located either locally or remotely relative to the rest of system 200.

During a stage of the stimulation process for a stimulated well, slurry can be pumped by the pump 120 at the top of the wellhead. The physics and engineering aspects that are involved can be complex and data can sometimes be uncertain and noisy. The use of a DRNN can resolve time and spatial non-linear variations. The DRNN can predict a cluster efficiency in a slurry pumping operation based on surface data from a surface sensor and wellbore data from one or more downhole sensors, such as a DAS or DTS. The DRNN can be trained using training data of a subset of data received from the sensors (e.g., time scaled or based on another variable) with known cluster efficiencies, or from data received from another well than the one to which the trained DRNN will be applied.

FIG. 3 is an example of a flowchart of a process 300 for predicting a cluster efficiency and controlling a pump for pumping slurry in a wellbore. Some examples can include more, fewer, or different blocks than those shown in FIG. 3. The blocks shown in FIG. 3 can be implemented using, for example, the computing device illustrated in FIG. 1 and FIG. 2.

In block 302, a computing device with a trained DRNN model receives surface data detected by a surface sensor at a surface of a wellbore. The surface data can include a pressure of slurry being pumped in to a wellbore and a flow rate of the slurry being pumped into the wellbore. The surface sensor can be a flow sensor and may include different components, such as a component to measure flow rate and another component, such as a piezoelectric component, to measure pressure. The computing device can receive the surface data by being communicatively coupled to the surface sensor. The surface data can be associated with a pumping stage at which the pump is operating.

In block 304, the computing device receives downhole data from one or more downhole sensors, such as sensors that are part of a DAS or DTS. For example, the downhole sensors may be part of or implemented by a fiber optic cable that is positioned in the wellbore and capable of detecting data about the environment downhole at different positions in the wellbore. The downhole data can include flow rates and pressures of flow of slurry at different positions in the wellbore, along with the locations of fractures in the wellbore. The downhole data can include other types of data, such as acoustically sensed data representing a physical structure of the wellbore environment at different positions in the wellbore. The computing device can be communicatively coupled to the downhole sensors, such as via a fiber optic coupling to the DAS or DTS fiber optic cable.

In block 306, the computing device uses the surface data and the downhole data as inputs to the trained DRNN and determines a cluster efficiency for the wellbore that is associated with certain pump operational settings, such as pressure and flow rate of slurry pumping. The cluster efficiency can be how uniformly the slurry flow will be distributed among perforation clusters in the wellbore and the outputted cluster efficiency can represent an achievable uniform and distributed slurry flow among the factures in the wellbore to allow for more efficiency production if slurry or diverter material is pumped according to the operational settings.

In block 308, the computing device outputs a command to the pump to operate according to the operational settings that the computing device predicts will result in a high cluster efficiency. The pump can be communicatively coupled to the computing device to receive the command and can change operational settings in accordance with the command. In response to the pump changing operational settings, the pump can be considered to be operating in a subsequent pumping stage and the process can return to block 302 to repeat the process for the subsequent pumping stage.

FIG. 4 is a chart 400 of slurry rate flow percentages per depth range as detected by downhole sensors that are part of a DAS system. The chart 400 indicates that slurry flow rate percentage is highest at depth range 1 and depth range 3 and lower at other depth ranges. For cluster efficiency, the percentage of slurry flow detected at different depth ranges would ideally be the same or similar. The detected slurry flow rate can be used with surface data by a computing device with a DRNN to determine operational settings of the pump to achieve a better cluster efficiency than is reflected in FIG. 4.

FIG. 5 is a graphical illustration of an example of a DRNN that is an LSTM 500 that can be used to control pumping operations for slurry or diverter material in a wellbore in a wellbore. LSTM 500 does not make use of convolutional layers. The model can provide for a multi-step prediction method for spatiotemporal data provided by the surface sensor and downhole sensors. LSTM 500 can provide a network structure for spatiotemporal data that accounts for the spatiotemporal characteristics of the data by providing strong correlation between local neighbors. The structure of the LSTM is configured by the surface data and the downhole data to form a structure based on state-to-state transitions. The network can be regularized by specifying a number of hidden units in the LSTM to avoid over-fitting or under-fitting the surface data. Over-fitting of the data occurs when a model fits the data almost perfectly and the model is complicated. This typically means the model is fitting the noise of the data. The model has low bias and high variance. On the contrary, under-fitting occurs when the model is too simple to fit the data hence it has high bias and low variance. The predictive model should balance between over-fitting and under-fitting the data. This can be determined by the performance of the model on the training and test data.

The network at each of points 502, 504, and 506 can be a learned representation that accounts for the characteristics of the wellbore as related to slurry placement. The network at future points in time 508, 510, and 512 is a predicted representation that takes into account these characteristics.

FIG. 6 is a graphical illustration of a DRNN that is an LSTM network 600 with convolutional layers. Convolutions are used for both input-to-state and state-to-state connections. Using convolutional layers, the final state, represented by layer 602, can have a large receptive field. FIG. 7 shows how convolutional layers 700 are used to predict wellbore properties. An input 702 is used in an encoding network including layers 704 and 706. A prediction network configured with the real-time surface data and including convolutional layers 708 and 710 produces a prediction, 712.

In some aspects, systems, devices, and methods for multi-stage placement of material in a wellbore are provided according to one or more of the following examples:

Example 1 is a system comprising: a pump in operable communication with a wellbore having multiple stages, to pump slurry or diverter material into the wellbore; a surface sensor positionable at a surface of the wellbore to detect surface data about the pump; a downhole sensor system positionable in the wellbore to detect downhole data about an environment of the wellbore; and a computing device to communicate with the pump, the surface sensor, and the downhole sensor system, the computing device being operable to: receive the surface data from the surface sensor; receive the downhole data from the downhole sensor system; apply the surface data and the downhole data to a long short-term memory (LSTM) neural network to produce a predicted cluster efficiency associated with operational settings of the pump; and control the pump using the operational settings to achieve the predicted cluster efficiency.

Example 2 is the system of example 1, wherein the LSTM neural network is a deep recurrent neural network (DRNN) that is trained using a subset of the surface data and of the downhole data.

Example 3 is the system of example 1, wherein the surface data includes a pump pressure from the pump and flow rate of slurry or diverter material.

Example 4 is the system of example 1, wherein the downhole sensor system is a distributed acoustic sensing system or a distributed temperature sensing system implemented by a fiber optic cable.

Example 5 is the system of example 1, wherein the downhole data includes flow rate percentage at different depth ranges in the wellbore.

Example 6 is the system of example 1, wherein the cluster efficiency represents a measurement of how uniformly that slurry or diverter material is distributed among perforation clusters in the wellbore.

Example 7 is the system of example 1, wherein the computing device is operable to control the pump using the operational settings to achieve the predicted cluster efficiency substantially in real time with respect to receiving the surface data and the downhole data.

Example 8 is a method comprising: receiving surface data from a surface sensor positioned at a surface of a wellbore to detect surface data about a pump; receiving downhole data from a downhole sensor system disposed in the wellbore to detect the downhole data about an environment of the wellbore; applying the surface data and the downhole data to a long short-term memory (LSTM) neural network to produce a predicted cluster efficiency associated with operational settings of the pump; and controlling the pump using the operational settings to achieve the predicted cluster efficiency.

Example 9 is the method of example 8, wherein the LSTM neural network is a deep recurrent neural network (DRNN) that is trained using a subset of the surface data and of the downhole data.

Example 10 is the method of example 8, wherein the surface data includes a pump pressure from the pump and flow rate of slurry or diverter material.

Example 11 is the method of example 8, wherein the downhole sensor system is a distributed acoustic sensing system or a distributed temperature sensing system implemented by a fiber optic cable.

Example 12 is the method of example 8, wherein the downhole data includes flow rate percentage at different depth ranges in the wellbore.

Example 13 is the method of example 8, wherein the cluster efficiency represents a measurement of how uniformly that slurry or diverter material is distributed among perforation clusters in the wellbore.

Example 14 is the method of example 8, wherein controlling the pump using the operational settings to achieve the predicted cluster efficiency comprises controlling the pump substantially in real time with respect to receiving the surface data and the downhole data.

Example 15 is a non-transitory computer-readable medium that includes instructions that are executable by a processing device for causing the processing device to perform operations comprising: receiving surface data from a surface sensor positioned at a surface of a wellbore to detect surface data about a pump; receiving downhole data from a downhole sensor system disposed in the wellbore to detect the downhole data about an environment of the wellbore; applying the surface data and the downhole data to a long short-term memory (LSTM) neural network to produce a predicted cluster efficiency associated with operational settings of the pump; and controlling the pump using the operational settings to achieve the predicted cluster efficiency.

Example 16 is the non-transitory computer-readable medium of example 15, wherein the LSTM neural network is a deep recurrent neural network (DRNN) that is trained using a subset of the surface data and of the downhole data.

Example 17 is the non-transitory computer-readable medium of example 15, wherein the surface data includes a pump pressure from the pump and flow rate of slurry or diverter material, wherein the downhole data includes flow rate percentage at different depth ranges in the wellbore.

Example 18 is the non-transitory computer-readable medium of example 15, wherein the downhole sensor system is a distributed acoustic sensing system or a distributed temperature sensing system implemented by a fiber optic cable.

Example 19 is the non-transitory computer-readable medium of example 15, wherein the cluster efficiency represents a measurement of how uniformly that slurry or diverter material is distributed among perforation clusters in the wellbore.

Example 20 is the non-transitory computer-readable medium of example 15, wherein the operation of controlling the pump using the operational settings to achieve the predicted cluster efficiency comprises controlling the pump substantially in real time with respect to receiving the surface data and the downhole data.

The foregoing description of certain examples, including illustrated examples, has been presented only for the purpose of illustration and description and is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Numerous modifications, adaptations, and uses thereof will be apparent to those skilled in the art without departing from the scope of the disclosure.

Claims

1. A system comprising: a pump in operable communication with a wellbore having multiple stages, to pump slurry or diverter material into the wellbore;a surface sensor positionable at a surface of the wellbore to detect surface data about the pump;a downhole sensor system positionable in the wellbore to detect downhole data about an environment of the wellbore; anda computing device to communicate with the pump, the surface sensor, and the downhole sensor system, the computing device being operable to:receive the surface data from the surface sensor; receive the downhole data from the downhole sensor system;apply the surface data and the downhole data to a long short-term memory (LSTM) neural network to produce a predicted cluster efficiency associated with operational settings of the pump; andcontrol the pump using the operational settings to achieve the predicted cluster efficiency.

2. The system of claim 1, wherein the LSTM neural network is a deep recurrent neural network (DRNN) that is trained using a subset of the surface data and of the downhole data.

3. The system of claim 1, wherein the surface data includes a pump pressure from the pump and flow rate of slurry or diverter material.

4. The system of claim 1, wherein the downhole sensor system is a distributed acoustic sensing system or a distributed temperature sensing system implemented by a fiber optic cable.

5. The system of claim 1, wherein the downhole data includes flow rate percentage at different depth ranges in the wellbore.

6. The system of claim 1, wherein the predicted cluster efficiency represents a measurement of how uniformly that slurry or diverter material is distributed among perforation clusters in the wellbore.

7. The system of claim 1, wherein the computing device is operable to control the pump using the operational settings to achieve the predicted cluster efficiency substantially in real time with respect to receiving the surface data and the downhole data.

8. A method comprising: receiving surface data from a surface sensor positioned at a surface of a wellbore to detect surface data about a pump;receiving downhole data from a downhole sensor system disposed in the wellbore to detect the downhole data about an environment of the wellbore;applying the surface data and the downhole data to a long short-term memory (LSTM) neural network to produce a predicted cluster efficiency associated with operational settings of the pump; andcontrolling the pump using the operational settings to achieve the predicted cluster efficiency.

9. The method of claim 8, wherein the LSTM neural network is a deep neural network (DRNN) that is trained using a subset of the surface data and of the downhole data.

10. The method of claim 8, wherein the surface data includes a pump pressure from the pump and flow rate of slurry or diverter material.

11. The method of claim 8, wherein the downhole sensor system is a distributed acoustic sensing system or a distributed temperature sensing system implemented by a fiber optic cable.

12. The method of claim 8, wherein the downhole data includes flow rate percentage at different depth ranges in the wellbore.

13. The method of claim 8, wherein the predicted cluster efficiency represents a measurement of how uniformly that slurry or diverter material is distributed among perforation clusters in the wellbore.

14. The method of claim 8, wherein controlling the pump using the operational settings to achieve the predicted cluster efficiency comprises controlling the pump substantially in real time with respect to receiving the surface data and the downhole data.

15. A non-transitory computer-readable medium that includes instructions that are executable by a processing device for causing the processing device to perform operations comprising: receiving surface data from a surface sensor positioned at a surface of a wellbore to detect surface data about a pump;receiving downhole data from a downhole sensor system disposed in the wellbore to detect the downhole data about an environment of the wellbore;applying the surface data and the downhole data to a long short-term memory (LSTM) neural network to produce a predicted cluster efficiency associated with operational settings of the pump; andcontrolling the pump using the operational settings to achieve the predicted cluster efficiency.

16. The non-transitory computer-readable medium of claim 15, wherein the LSTM neural network is a deep recurrent neural network (DRNN) that is trained using a subset of the surface data and of the downhole data.

17. The non-transitory computer-readable medium of claim 15, wherein the surface data includes a pump pressure from the pump and flow rate of slurry or diverter material, wherein the downhole data includes flow rate percentage at different depth ranges in the wellbore.

18. The non-transitory computer-readable medium of claim 15, wherein the downhole sensor system is a distributed acoustic sensing system or a distributed temperature sensing system implemented by a fiber optic cable.

19. The non-transitory computer-readable medium of claim 15, wherein the predicted cluster efficiency represents a measurement of how uniformly that slurry or diverter material is distributed among perforation clusters in the wellbore.

20. The non-transitory computer-readable medium of claim 15, wherein the operation of controlling the pump using the operational settings to achieve the predicted cluster efficiency comprises controlling the pump substantially in real time with respect to receiving the surface data and the downhole data.