• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A method includes performing a first wellbore treatment operation, determining an operational attribute of the well in response to the first wellbore treatment operation, and determining a predicted response using a recurrent neural network and on the basis of the operational attribute. The method also includes defining a controllable wellbore treatment attribute based on the predicted response, and performing a second wellbore wellbore processing operation on the basis of the predicted response. the wellbore treatment attribute being controllable.
    公开号:FR3075434A1
    申请号:FR1859363
    申请日:2018-10-09
    公开日:2019-06-21
    发明作者:Srinath Madasu；Yogendra Narayan PANDEY；Keshava Prasad Rangarajan
    申请人:Landmark Graphics Corp；
    IPC主号:
    专利说明:

    CONTEXT
    The present description relates generally to wellbore operations and, more particularly, to the modeling of neural networks for wellbore operations.
    Underground processing operations may include various drilling well processing operations and drilling operations. In some applications, processing operations may include hydraulic fracturing. In a hydraulic fracturing treatment, a fracturing fluid is introduced into the formation at a rate high enough to exert sufficient pressure on the formation to create and / or extend fractures therein. The fracturing fluid suspends the proppant particles which must be placed in the fractures to prevent the fractures from closing completely when the hydraulic pressure is released, thus forming conductive channels inside the formation through which hydrocarbons can flow to the wellbore for production. The hydraulic fracturing treatment can take place in several stages, in which the fracturing fluid is injected into the well at each stage.
    In certain circumstances, several factors may interfere with the accurate and rapid prediction of responses to processing or other wellbore operations. Imprecise predictions can increase the difficulty of defining wellbore processing attributes that can be controlled to optimize operational performance, while slow predictions may be impractical due to the time constraints of a wellbore operation . Systems that increase accuracy and speed of prediction can be used to improve the definition of wellbore processing attributes that can be controlled based on response predictions.
    BRIEF DESCRIPTION OF THE DRAWINGS
    Examples of the description can be better understood with reference to the accompanying drawings.
    Figure 1 shows a diagram of a wellbore system and the underlying formation, according to some embodiments.
    FIG. 2 represents a diagram of a long-short-term memory cell (LSTM), according to certain embodiments.
    2017-IPM-101099-U1 -FR
    FIG. 3 represents a diagram of LSTM cells stacked in a neural network, according to certain embodiments.
    FIG. 4 represents a flow diagram of operations for forming stacked LSTM cells, according to certain embodiments.
    FIG. 5 represents a flow diagram of operations for predicting values using a recurrent neural network (RNN) based on operational attributes of a wellbore, according to certain embodiments.
    FIG. 6 represents an example of a graph of the surface pressure as a function of time, according to certain embodiments.
    FIG. 7 represents an example of a graph of the fluid flow as a function of time, according to certain embodiments.
    FIG. 8 represents an example of a graph of the flow rate of support agent as a function of time, according to certain embodiments.
    FIG. 9 represents an example of a graph of predicted surface pressure relative to a graph of surface pressure as a function of time, according to certain embodiments.
    FIG. 10 represents an example of a treatment operation carried out in an underground formation, according to certain embodiments.
    FIG. 11 represents an example of a drilling operation carried out in an underground formation, according to certain embodiments.
    FIG. 12 represents an example of a computer device, according to certain embodiments.
    DESCRIPTION OF EMBODIMENTS
    The following description includes examples of systems, methods, techniques, and program flows that represent embodiments of the description. However, it will be understood that this description can be implemented without these specific details. For example, this description refers to long-term memory neural networks (LSTM) in illustrative examples. Examples of this description can also be applied to other types of architectures of recurrent neural networks (RNN), such as gateway recurrent neural networks (GRU). Other well-known instances, instructions, protocols, structures and techniques have not been presented in detail so as not to obscure the description.
    2017-IPM-101099-U1 -FR
    Various embodiments include predicting one or more responses to various underground processing operations to resolve a variation in time and space of the predicted response. Resolving a time and space variation of the predicted response may include determining the response value, a time or time interval during which the response occurs, and / or a location in which the response occurs. Underground processing operations may include various drilling well processing operations and drilling operations. As used herein, the terms "treat", "treatment", "treating", etc. relate to any underground operation that uses a fluid in conjunction with the performance of a desired function and / or for a desired purpose. The use of these terms does not imply any particular action on the part of the treatment fluid. Illustrative processing operations may include, for example, fracturing operations, gravel compacting operations, acidification operations, dissolution and removal of scale, consolidation operations and the like.
    Some embodiments include the use of RNN to predict responses from various wellbore processing operations, such as fracturing, deflection, acidification, etc. applications. along a borehole to improve oil recovery. An RNN is a neural network in which connections between cells can form a directed cycle and can use their internal memory to store information from previous operations, thereby increasing the speed and accuracy of prediction. An RNN can be used in real time during these wellbore processing operations, allowing adjustments and control in real time. An RNN can predict a response based on a set (i.e. one or more) of operational attributes. An operational attribute can be any type of measurement or approximation linked to the well system carried out before or during a borehole treatment. One or more controllable wellbore processing attributes may be defined based on the predicted pressure response, where a controllable wellbore processing attribute is an attribute controllable by a user or processor (for example, surface pump pressure, sand composition, selected particle sizes, viscosity of the stimulation fluid, etc.).
    In some embodiments, the RNN may include a stacked long-term short-term memory neural network (LSTM). After iterating treatment inputs at a time interval, cells in an LSTM neural network may contain an internal cell state that can be used to respond more precisely to discontinuities
    2017-IPM-101099-U1 -FR and to non-linearities of a multivariate data set. In some embodiments, these discontinuities and non-linearities may include responses resulting from the presence of faults, unexpected formation changes, drilling anomalies, drilling accidents or unforeseen well treatment incidents. . An RNN can provide fast, accurate, and high-resolution predictions by including operations that take advantage of the temporal nature of multivariate wellbore data during multistage wellbore operations. These predictions can be used to define a wellbore processing attribute that can be controlled such as a fluid flow during a processing step.
    Example of a well bore representation
    Figure 1 shows a diagram of a wellbore system and the underlying formation, according to some embodiments. A wellbore system 100 shown in FIG. 1 includes a wellbore 104 penetrating at least a portion of an underground formation 102. The underground formation 102 may include any underground geological formation suitable for drilling, fracturing (by shale), acidification (e.g. carbonate), etc. The underground formation 102 may include pores initially saturated with reservoir fluids (eg, oil, gas and / or water). The wellbore 104 includes one or more injection points 114 into which one or more fluids can be injected from the wellbore 104 into the underground formation 102. In some embodiments, the wellbore system 100 can be stimulated by the injection of a fracturing fluid at one or more injection points 114 into the wellbore 104. In certain embodiments, the injection point or points 114 may correspond to the points d injection into a well casing 104.
    When fluid enters the underground formation 102 at the injection points 114, one or more fractures 116 can be opened and the pressure difference between the solid stress and the fracture 116 causes a flow in the fracture 116. As shown in FIG. 1, the underground formation 102 comprises at least one network of fractures 108 connected to the wellbore 104. The network of fractures 108 can comprise a plurality of junctions and a plurality of fractures 116. The number of junctions and fractures 116 can vary according to the specific characteristics of the underground formation 102. For example, the network of fractures 108 may include thousands of fractures 116 or hundreds of thousands of fractures 116.
    Operational attributes can be determined before or during a wellbore processing operation. In some embodiments, the attributes
    2017-IPM-101099-U1 -EN operational may include one or more measurements acquired by a sensor, one or more predicted results (for example, average fracture length) and / or one or more properties of the well (for example, radius of well, casing radius, length of well). For example, an operational attribute can characterize a processing operation for a wellbore 104 penetrating at least part of an underground formation 102. In certain embodiments, the operational attribute or attributes can include measurements in real time. For example, real-time measurements can include pressure measurements, flow measurements, and fluid temperature. In some embodiments, real-time measurements can be obtained from one or more sources of well location data. Well location data sources may include, but are not limited to, flow sensors, pressure sensors, thermocouples, and any other suitable measuring device. For example, well location data sources can be positioned on the surface, on a downhole tool, in wellbore 104 or in a fracture 116. Pressure measurements can, for example, be obtained from a pressure sensor on a surface of borehole 104.
    The values of the operational attributes can be used by an RNN to determine values for one or more wellbore processing attributes that can be controlled. In some embodiments, one or more controllable wellbore treatment attributes may include, but are not limited to, an amount of treatment fluid pumped into the wellbore system 100, the wellbore pressure drilling at the injection points 114, the flow rate at the wellbore inlet 110, the pressure at the wellbore inlet 110, an acid flow rate, a propellant flow rate, a concentration of propping agent, a distance chosen between puncturing groups, a diameter of propping particles and any combination thereof.
    In some embodiments, the well inlet pressure 110 predicted by an RNN can be used, at least in part, to determine whether or not to use a proppant, to determine the amount of the propellant. support to use, to develop a stimulation pumping program or any combination thereof. For example, in some embodiments, flow and / or pressure sensors may be positioned at the well bore inlet 110 of the well bore 104 to measure the flow and pressure in real time. Measured inlet flow and pressure data can be used as operational attributes. In some embodiments, the formation attribute (s) may characterize the underground formation 102. In some embodiments, the formation attribute (s) may include properties of the underground formation
    2017-IPM-101099-U1 -FR
    102 such as the geometry of the underground formation 102, the stress field, the pore pressure, the formation temperature, the porosity, the resistivity, the water saturation, the composition of the hydrocarbons and any combination thereof.
    Example of recurrent neural networks and recurrent neural network systems
    FIG. 2 represents a diagram of a long-short-term memory cell (LSTM), according to certain embodiments. The LSTM 200 cell can be part of an RNN. The LSTM 200 cell at the time interval t can receive and store various information. One type of information storable by the LSTM cell 200 is the cell state C 1 ti 202, which is the cell state of the LSTM cell determined at the previous time interval L1. In some embodiments, a time interval is a simulated representation of real time. Alternatively, a time interval may be an arbitrary unit which represents different stages of operations, such as a stage during a well treatment operation.
    One type of information that can be received by the LSTM cell 200 is the output p ^ -i 204, which is the output determined at the previous time interval L1. Another type of information that can be received by the LSTM cell 200 is the input xt 206, which can be an input with one or more variables at the current time interval t. The input xt 206 can include operational attributes such as a fluid flow rate ry t and / or a support agent flow rate r p , t from the time interval t in the predefined search window of the LSTM 200 cell, which can be expressed as shown in Equation 1.
    xi = r ft , r p , t (1)
    In some embodiments, the input xt 206 may include other operational attributes and may be determined before starting the processing depending on the design of the processing. Examples of such inputs may include the properties of the proppant, the properties of the fluid, the surface pressure, the diameter of the borehole, the temperature, the acid concentration, etc.
    The LSTM cell can use four doors to process information, each of which can be associated with weights and weights. These weights and weights can be calibrated during a training process to provide accurate predictions of an exit in a time series. In some embodiments, the forget gate 222 can be used to determine an intermediate set of forget gate values f. The oblivion gate 222 can be modeled as shown in equation 2 below, where σ is a sigmoid function, pt-i is an output of the LSTM 200 cell from the time interval
    2017-IPM-101099-U1 -EN previous Al, Wf is a weight associated with the forgetting gate, and bf is a weighting of the forgetting gate.
    ft = σ (Wf. [ t -i, xt] + bf) (2)
    Entry door 224 can be used to determine an intermediate set of input values it. Entrance door 224 can be modeled as shown in equation 3 below, where PL is a weight associated with entry door it 224 and bf is a weighting of entry door it 224.
    it = σ (Wi. [ t -i, xt] + bt) (3)
    In addition to the forgetting door and the front door, a candidate door 226 can be used to generate candidate cell state values Ct. In some embodiments, the candidate door can be modeled as shown in equation 4 below, in which W c is a weight associated with the candidate gate and b c is a weight associated with the candidate gate.
    Ct = tanh (W c . [Pt-i, xt] + b c ) (4)
    Once the values for forget gate 222, entrance gate 224 and candidate gate 226 have been determined, cell state values Ct at the current time interval t can be determined on the basis of cell state values, forgetting gate results, and previous gateway results. For example, the state of cell C 7/254 can be determined based on the state of cell C ^ -i 202 from a previous time interval, values of the state of cell candidates, the values of oblivion gate 222 calculated using equation 2, candidate gate 226 values calculated from equation 3 and candidate cell state values Ct. This determination can be modeled as shown in equation 5 below, in which O represents the product operator element by element:
    C ^ ftOC ^ -i + itOG (5)
    The output gate 228 can be used to determine a set of intermediate output values o t based on the set of input values xt. In some embodiments, a sigmoid function can be applied to a result based on the set of input values xt and previous cell output pt-i 204. This determination can be modeled as indicated in equation 6 ci - below, in which o t is the set of values of the intermediate exit door, W o is a weight associated with the exit door and b 0 is a weighting of the exit door:
    ot = σ (W 0. [t-i, xt] + b 0 ) (6)
    2017-IPM-101099-U1 -FR
    The final exit door 230 can be used to determine the output p 2 t 254 based on the result of the exit door o t to keep it within a particular range depending on the cell state. In some embodiments, the final exit door 230 can be modeled as shown in Equation 7.
    pt = OtOtanh (Ct) (7)
    FIG. 3 represents a diagram of LSTM cells stacked in a neural network, according to certain embodiments. The stacked LSTM cells 300 include the LSTM cell 200 and the LSTM cell 399 operating at a first time interval t-1 and at a second time interval t. The two cells operating at the first time interval A1 are presented in the first column and the two cells operating at the second time interval t are presented in the second column 381.
    At time interval A1, the cell of LSTM 200 can determine a cell state Cdt-i 202 (its cell state at time interval Al) and an output p ^ -i 204 (the output of the cell of LSTM 200 at the time interval A1) based on the cell state C '1-2 201 (the cell state of the cell of LSTM 200 at the time interval A2), the output p ! 1-2 203 (the output of the LSTM 200 cell at the time interval A2), and the multivariate input xt-i 205 (the multivariate input at the time interval Al). With reference to FIG. 2, the LSTM cell 200 can use the same operations as those described above for the forgetting door 222, the entry door 224, the candidate door 226, the exit door 228, the normalized exit door 230 and equations 2 to 7 to determine the cell state Cdt-i 202 and the output p ^ -i 204.
    In some embodiments, an LSTM 399 cell can operate simultaneously with the LSTM 200 cell. At time interval A1, the LSTM 399 cell can determine a cell state Cdt-i 302 (its cell state at the time interval Al) and an output p ^ -i 204 (the cell output from LSTM 399 at the time interval Al) based on its cell state C 2 t-2 301 (cell state of the cell from LSTM 399 at time interval A2), outputpt-i 304 (output from cell LSTM 399 at time interval A2), and multivariate input xt 206 (multivariate input at time interval Al). With reference to FIG. 2, the forgetting door 322, the entry door 324, the candidate door 326, the exit door 328 and the door 330 can operate in operations identical or similar to those of the entry door. forgetting 222, entry door 224, candidate door 226, exit door 228 and standard exit door 230, respectively. The LSTM 399 cell can use these operations to determine cell status Ct-i 302 and outputpt-i 304.
    In some embodiments, the time interval t, the cell MTSA 200 may determine the state of cell C 7/252 and sortiep t 2 254 at the time interval t
    2017-IPM-101099-U1 -EN based on its cell state (Tt-i 202, output p 2 ti 204, and multivariate input xt 206 at time interval t. The cell of LSTM 200 can determine the state of cell C 7/252 and output p 2 t 254 to the time interval t using operations identical or similar to those described above for the cell MTSA 200 to the time interval t .
    In certain embodiments, at the time interval ζ the cell of LSTM 399 can determine its cell state G 352 and the output p 2 t 354 at the time interval t based on its cell state C 2 ti 302, the output p 2 ti 304 and the multivariable input xt 306 at the time interval t. At the time interval ζ the cell of LSTM 399 can use operations identical or similar to those described above for the cell of LSTM 399 at the time interval Al to determine the state of cell G 352 and the output p 2 t 354.
    Example of RNN operations
    Figure 4 and Figure 5 show flowcharts of operations that can be performed by software, firmware, hardware, or a combination thereof. For example, with reference to FIG. 12 (described in more detail below), a processor in a computer device located on the surface can execute instructions for performing operations of the flowchart 400.
    FIG. 4 represents a flow diagram of operations for forming stacked LSTM cells, according to certain embodiments. The operations of the flowchart 400 begin at block 402. At block 402, a set of operational attributes at a first measurement time is determined. For example, a set of operational attributes may include a flow of fluid in units of barrels per minute (BPM) and a flow of retaining agent also in units of BPM. An example of a data set for a set of time intervals recording these operational attributes can be presented in Table 1, as well as a surface pressure with units of pounds per square inch (psi).
    Table 1 Time interval Fluid flow (BPM) Support Agent Flow (BPM) Surface pressure (psi) 1 7.81 81.30 8,500.5 2 7.87 81.23 8,501.3 3 7.90 81.21 8,495.9 4 7.90 81.21 8,499.1 5 7.87 80.95 8,498.3
    2017-IPM-101099-U1 -FR
    In block 404, an initial LSTM cell and an initial time interval t are defined. The initial LSTM cell can be defined on a cell in an RNN system. For example, with reference to FIG. 3, in the case of the execution of the operations of the flowchart 400 using the stacked LSTM cells 300, the initial LSTM cell can be defined at the LSTM cell 200. The initial time interval may be the first time interval at which the input data is available, a predetermined number of time intervals before a target time interval, or an initial time interval based on an event. For example, with reference to Table 1, the initial time interval can be defined as time interval 1. Alternatively, if the target time interval is time interval 4, the initial time interval can be defined at two time intervals before the target time interval, which would result in setting the initial time interval to time interval 2. Alternatively, the initial time interval can be based on an event and defined at the first time interval after which an event such as a fracture reaching a defect occurs.
    At block 406, a set of predicted responses is determined and the cell parameters are updated based on operational attributes at the current time interval, the output of a previous time interval, the state of cell from the previous time interval, and from the cell parameter set for an LSTM cell. In some embodiments, the cell parameter set may include cell states, weights, and weights for each gate (for example, Ct, Wf bf Wi „b h W c , b c , W o , b 0 , etc.) and other parameters of a neural network cell. The set of outputs for the current time interval can be determined using the cell of LSTM 399 based on equations 1 to 7. In some embodiments, the set of cell parameters can be updated by function of the difference between a predicted response and a measured response.
    For example, with reference to Table 1, a set of operational attributes may include the fluid flow rate and the propellant flow rate corresponding to the time interval 3. Again with reference to Figure 3, the LSTM cell 399 can be used to predict a surface pressure of 8,401.5 psi based on the fluid flow rate of 7.90 BPM, the propellant flow rate of 81.21 BPM, the cell state Ci of the LSTM 399 cell and the cell parameter set. In some embodiments, this prediction can be compared to the actual surface pressure measurement 8499.1 to determine a prediction error. The prediction error can be used to update the value of the cell parameter set. For example, the set of cell parameters used to determine
    2017-IPM-101099-U1 -EN the predicted value of 8 401.5 psi could have been 0.75, 0.55, 0.57, 0.54, 0.46, 0.3, 0.76 and 0.9 for Wf bf Wt, b h W c , b c , W o and b 0 respectively . After updating the cell parameters using a back propagation method, the new cell parameters can be 0.65, 0.95, 0.50, 0.53, 0.16 respectively , 0.2, 0.76 and 0.95.
    At block 408, it is determined whether a target time interval has been reached. In some embodiments, a target time interval can be set manually. For example, the target time interval can be set to 10. Alternatively, a target time interval can also be set to the total number of available time intervals. For example, when training an RNN to calibrate its cell parameters with 20 recorded time intervals, the target time interval can be set to 20. If the target time interval is reached, then the operations of flowchart 400 continues at block 412. If the target time interval is not reached, then the operations of flowchart 400 continue at block 410.
    In block 410, the time interval is incremented. Once the time interval is incremented, the operations of flowchart 400 continue to block 406, in which an output for the incremented time interval can be determined. In addition, the set of cell parameters can be updated based on the inputs of the incremented time interval, the set of outputs of the previous time interval and the cell state of the cell. 'previous time interval, as described above.
    In block 412, it is determined whether more LSTM cells should be used. In some embodiments, more LSTM cells must be used if at least one assigned LSTM cell has not been formed and / or used to determine the set of outputs. In some embodiments, the number of LSTM cells allocated can be predetermined or manually defined before the start of operations in flowchart 400. For example, with respect to Figure 3, the number of LSTM cells allocated can be defined at 2. In some embodiments, the number of allocated LSTM cells can be determined dynamically based on the set of outputs at a previous time interval. If more LSTM cells are to be used, the operations of flowchart 400 continue to block 416. Otherwise, the operations of flowchart 400 continue to block 416.
    In block 414, the operation goes to the next LSTM cell and resets the time interval t to the initial time interval. In some embodiments, it can be determined that at least one additional available LSTM cell has not been used and that one LSTM cell is selected as the next LSTM cell. For example,
    2017-IPM-101099-U1 -EN referring to Figure 3, after using the LSTM 200 cell, the LSTM 399 cell can be selected as the next LSTM cell.
    At block 416, the efficiency of the LSTM neural network is quantified using unused data. In some embodiments, the efficiency of the LSTM neural network can be quantified based on accuracy, precision and speed of calculation using data sets that have not been used to train or validate the LSTM neural network. Based on the effectiveness of the LSTM neural network, one can decide whether or not to use the LSTM neural network formed during wellbore operations. Once the efficiency of the LSTM network is quantified, the operations of the organization chart 400 are completed.
    FIG. 5 represents a flow diagram of operations for predicting values using a recurrent neural network (RNN) based on operational attributes of a wellbore, according to certain embodiments. The operations of flowchart 500 begin at block 502.
    At block 502, a time interval is advanced. In some embodiments, a time interval may be a step of operation without a unit. For example, the transition from a first time interval to a second time interval could represent an advance from a first step of operation to a second step of operation. In some embodiments, a time interval may be a constant time interval. For example, the time between each of a set of time intervals can be 6 hours. Alternatively, a time interval may be a variable time interval. For example, the duration of a variable time interval can be 1 minute if a predicted response is less than 10 psi / min or 30 minutes otherwise.
    In block 504, the operational attributes are determined in the advanced time. In certain embodiments, the operational attributes can be determined by using one or more operations which are identical or similar to the operations described above in block 402 of FIG. 4.
    At block 540, it is determined whether an abnormal wellbore event has occurred. An abnormal wellbore event is an event related to a significant change in the formation or operation of a wellbore, where a recurrent neural network formed from measurements made before the abnormal wellbore event is less accurate than a recurrent neural network formed with data that sets aside actions taken before the abnormal wellbore event. In some embodiments, determining whether or not an abnormal wellbore event occurs
    2017-IPM-101099-U1 -EN on a measured operational attribute exceeding an event threshold, where exceeding an event threshold may include an operational attribute greater than or equal to a threshold value or less than or equal to a value threshold. For example, an expected increase in pressure response may be greater than a threshold value and an abnormal wellbore event titled "large defect encountered" may be defined. If an abnormal wellbore event has not occurred, flowchart 500 operations continue at block 506. Otherwise, flowchart 400 operations continue at block 542.
    At block 542, the recurrent neural network is reformed based on the data measured after the abnormal wellbore event. The recurrent neural network can be reformed using one or more operations which are identical or similar to the operations described above in blocks 404 to 416 of Figure 4.
    In block 506, an RNN is operated based on the determined operational attributes. In some embodiments, the RNN LSTM cells can be operated using one or more operations that are the same or similar to the operations described above in block 406 of Figure 4. In some embodiments, each cell a neural network can be operated in parallel for each time interval. For example, again with reference to FIG. 2, each cell of a neural network can be used as described above for the gates 222-230 and the equations 1 to 7 in parallel to determine the outputs of the cells of the neural network after a time interval. A combined output from the LSTM neural network for the time interval can be based on the outputs from each cell at that time interval.
    At block 508, a response is predicted based on the outputs of the RNN. In some embodiments, the response may be based on an average of the outputs of each of the LSTM cells multiplied by a normalization factor. For example, flowchart 500 operations could use a total of two cells, where the average of a first cell and a second cell can be 0.60, and the normalization factor can be 10 psi. This can lead to an LSTM network response of 6.0 psi.
    At block 510, the data sets are updated based on the predicted responses. In some embodiments, the data sets include the operational attributes and the predicted responses. Updating the datasets may include inserting predicted responses into the datasets. For example, a dataset can include a known fluid flow and propellant flow at time interval 10. A surface pressure of 100 psi can be predicted based on the fluid flow and the known support agent.
    2017-IPM-101099-U1 -FR
    In block 512, a controllable wellbore processing attribute is defined based on the predicted responses. In some embodiments, the controllable wellbore processing attribute may be a throughput. For example, the LSTM neural network can predict that a process fluid flow for optimal pressure in a wellbore may be 1,500 BPM. An IT device can then set a surface pump to pump the process fluid into the wellbore at 1,500 BPM.
    At block 514, it is determined whether a target time interval has been reached. In some embodiments, the target time interval may be a time interval greater than the number of time intervals available with the data. For example, with reference to Table 1, the number of time slots available is 5 and the target time slot can be 6. Alternatively, a target time slot may depend on a response or an operational attribute predicted. For example, a target time interval should be considered reached if the pressure is above 19,000 psi and is not otherwise reached. If the target time interval is not reached, then the operations of flowchart 500 can continue at block 502. If the target time interval is reached, the operations of flowchart 500 are terminated.
    Sample data
    FIG. 6 represents an example of a graph of the surface pressure as a function of time, according to certain embodiments. Plot 600 includes a abscissa axis, a ordinate axis, a set of pressure data points 602, a first region 604 and a second region 606. The abscissa axis represents the time elapsed since the start of the measurements, measured in minutes. The ordinate axis represents a processing pressure measured in “psi” units. The set of pressure data points 602 represents the process pressure measured at various measurement times. The first region 604 represents a non-linearity in the pressure response. The second region 606 represents a second non-linearity in the pressure response. Non-linearity in a response can be any non-linear trend in a data set between a first variable and a second variable. A non-linearity of the pressure response can be due to a change in operational attribute (for example, sudden increase / decrease in flow, introduction or reduction of the propellant) or to the result of a natural discontinuity (for example example, a fracture with a defect, the pressure reaching a critical fracturing stress).
    FIG. 7 represents an example of a graph of the fluid flow as a function of time, according to certain embodiments. A plot 700 includes an abscissa axis, a
    2017-IPM-101099-U1 -EN ordinate axis, a set of data points 702, a first region 704 and a second region 706. The abscissa axis represents the time elapsed since the start of the measurements, measured in minutes. The ordinate axis represents a flow rate in "cubic feet per minute" units. The set of data points 702 represents the measured flow. With regard to FIG. 6, the first region 704 represents a non-linearity in the pressure response which corresponds in time to the first region 604 and a comparison of the two regions highlights a non-linearity represented by a fall in pressure and flow, respectively. With regard to FIG. 6, the second region 706 also represents a significant non-linearity in the pressure response which corresponds in time to the second region 606. Unlike the first region, however, the drop in flow described by the second region 706 does not correspond to a significant pressure drop, as shown by the second region 606, which shows that significant drops in flow can be independent of the drops in a pressure response.
    FIG. 8 represents an example of a graph of the flow rate of support agent as a function of time, according to certain embodiments. A plot 800 includes an abscissa axis, a ordinate axis, a set of data points 802, a first region 804 and a second region 806. The abscissa axis represents the time elapsed since the start of the measurements, measured in minutes . The ordinate axis represents a flow of support agent in “pounds per minute” units. The set of data points 802 represents the measured rate of supporting agent. As regards FIG. 6, the first region 804 represents a region with no detected change in the measurement of the propellant flow rate which corresponds in time to the first region 604. A comparison of regions 604 and 804 shows that a change in the measured treatment pressure can be independent of any change in the measured propellant flow. With regard to FIG. 6, the second region 806 represents a significant non-linearity in the measurement of the flow of supporting agent which corresponds in time to the second region 606. However, the drop in the flow of supporting agent represented by the second region 806 also does not show a proportional pressure drop, as shown by the second region 606.
    FIG. 9 represents an example of a graph of predicted surface pressure relative to a graph of surface pressure as a function of time, according to certain embodiments. Plot 900 includes a x-axis, a y-axis, the set of pressure data points 602, the first region 604, the second region 606, and a predicted pressure line 902. Each of the set of pressure points pressure data 602,
    2017-IPM-101099-U1 -EN of the first region 604 and the second region 606 can represent the same information as that represented in FIG. 6. The predicted pressure line 902 includes responses predicted by the RNN described above . In certain embodiments, the values determined by the RNN can be based on the measured flow rate described in FIG. 7 and the flow rate of support agent shown in FIG. 8.
    In some embodiments, the RNN system can be formed on data similar to or different from the values shown in Figure 7 and Figure 8. For example, the RNN system used to generate the predicted pressure line 902 can be formed using a plurality of data sets comprising measurements of time, treatment pressure, flow rate and support agent flow rate, none of which is identical to the data described in FIGS. 6 to 8 Once formed, this RNN formed can generate the predicted pressure line 902 based on the data shown in Figure 7 and Figure 8.
    The flow charts described above are provided to assist in understanding the illustrations and should not be used to limit the scope of the claims. The flow charts illustrate examples of operations which may vary within the scope of the claims. Additional operations can be performed; fewer operations can be performed; operations can be carried out in parallel; and operations can be performed in a different order. For example, the operations described in blocks 406 for each LSTM cell can be performed in parallel or simultaneously. With respect to Figure 500, updating the dataset is not necessary. It will be understood that each block of illustrations in the form of a flowchart and / or functional diagrams, and combinations of blocks in the illustrations in the form of a flowchart and / or functional diagrams, can be implemented by a program code. The program code may be supplied to a processor of a general purpose computer, a specialized computer or another machine or programmable device.
    Examples of well operations
    FIG. 10 represents an example of a treatment operation carried out in an underground formation, according to certain embodiments. FIG. 10 shows a well 1060 during a treatment operation in a part of an underground formation 1002 surrounding a wellbore 1004. The wellbore 1004 extends from a surface 1006 and a treatment fluid 1008 is applied to a part of the underground formation 1002 surrounding the horizontal part of the wellbore 1004. Although represented as being vertical relative to the horizontal, the wellbore 1004 may include geometries and orientations that are horizontal, vertical, inclined, curved and d other types, and the processing operation may
    2017-IPM-101099-U1 -EN be applied to an underground area surrounding any part of wellbore 1004. Wellbore 1004 may include casing 1010 which is cemented or otherwise attached to the wall of the wellbore . Well 1004 may be uncased or include uncased sections. Perforations can be formed in tubing 1010 to allow process fluids and / or other materials (e.g., proppant, acid, deflecting agent, etc.) to flow into the formation underground 1002. In cased wells, perforations can be formed using shaped fillers, a perforating gun, a hydrojet and / or other tools.
    The well 1060 is shown with a working rod train 1012 depending on the surface 1006 in the well bore 1004. The pump and mixer system 1048 can be coupled to the working rod train 1012 to pump the treatment fluid 1008 in borehole 1004 and be in communication with a computer device. The drill string 1012 may include a coiled tube, a pipe joint, and / or other structures that allow fluid to flow into the wellbore 1004. The drill string 1012 may include devices flow control, bypass valves, ports, and / or other well tools or devices that control fluid flow from within the working drill string 1012 in the underground formation 1002. For example, the working string 1012 may include ports adjacent to the wellbore wall to communicate processing fluid 1008 directly into the underground formation 1002 and / or the working rod string 1012 may include ports spaced from the wall wellbore for communicating process fluid 1008 in a ring in the wellbore between the working rod string 1012 and the wall of the wellbore.
    The drill string 1012 and / or the wellbore 1004 may include one or more sets of seals 1014 which seal the ring between the drill string 1012 and the wellbore 1004 to define an interval from the wellbore 1004 into which the treatment fluid 1008 will be pumped. Figure 10 shows the seals 1014, one defining an upstream border of the gap and the other defining a downhole end of the gap. When the process fluid 1008 is introduced into the wellbore 1004 (for example, the area of the wellbore 1004 between the seals 1014) at sufficient hydraulic pressure, one or more fractures 1016 can be created in the formation underground 1002.
    In some embodiments, the processing fluid 1008 may include proppant particles. For example, treatment fluid 1008 may contain proppant particles which can penetrate into fractures 1016
    2017-IPM-101099-U1 -EN as shown, or may plug or seal fractures 1016 to reduce or prevent the flow of additional fluid in these areas. A controllable wellbore processing attribute such as the retaining agent flow can be defined, wherein the retaining agent flow to be defined is based on the result of the RNN operations described above. RNN operations can be used to predict a change in pressure, and controllable wellbore processing attributes can be changed in response to the predicted pressure change. For example, the RNN operation can predict an increase in processing pressure from 10,000 to 15,000 psi based on an existing set of operational attributes, which may exceed a pressure threshold. In response, a propellant flow can be reduced to reduce the predicted and measured treatment pressure. Alternatively, the RNN operation can directly predict an optimally controllable wellbore processing attribute. For example, the RNN operation can predict an optimal support agent flow of 5,000 BPM and a computing device can set the support agent flow at 5,000 BPM in response to the prediction.
    In some embodiments, the process fluid 1008 may include an acid and be pumped into the underground formation 1002. For example, the process fluid 1008 may include hydrogen fluoride and create wormholes in part of the underground formation 1002. A controllable wellbore processing attribute such as the concentration of acid to be used may be based on the result of the RNN operations described above. RNN operations can be used to predict a wormhole growth rate, and controllable wellbore processing attributes can be changed in response to the predicted pressure change. For example, the RNN operation can predict a decrease in the length of wormholes based on an existing set of operational attributes. In response, a flow rate can be reduced to reduce the predicted and measured treatment pressure.
    In some embodiments, the treatment fluid 1008 may include a deflecting agent and / or a bridging agent for plugging or partially plugging an area of a well by forming a bridge. For example, the diverting agent can block a first zone and the treatment fluid can be diverted by the bridge to a less permeable zone. A controllable wellbore processing attribute such as the concentration of diversion agent to be used may be based on the result of the RNN operations described above. RNN operations can be used to predict a maximum stress supported by a deviation agent, and the controllable wellbore processing attributes can be changed in response to the predicted maximum stress. Through
    2017-IPM-101099-U1 -EN example, the RNN operation can predict a reduced maximum stress based on an existing set of operational attributes. In response, a concentration of deflecting agent can be increased to increase the predicted maximum stress.
    FIG. 11 represents an example of a drilling operation carried out in an underground formation, according to certain embodiments. FIG. 11 represents a drilling system 1100. The drilling system 1100 comprises a drilling tower 1101 situated on the surface 1102 of a borehole 1103. The drilling system 1100 also comprises a pump 1150 which can be actuated to pump fluid through a drill string 1104. The drill string 1104 can be operated to drill the borehole 1103 through the underground formation 1132 with the BHA.
    The BHA includes a drill bit 1130 at the downhole end of the drill string 1104. The BHA and the drill bit 1130 can be coupled to a computer system 1151, which can operate the drill bit 1130 and the pump 1150. The drill bit 1130 can be operated to create the borehole 1103 by penetrating the surface 1102 and the underground formation 1132. In some embodiments, a controllable wellbore processing attribute such as the Drilling RPM or drilling fluid flow can be based on the result of RNN operations described above. RNN operations can be used to predict a drilling speed, and the controllable wellbore processing attributes can be changed depending on the predicted drilling speed. For example, the RNN operation can predict a drilling speed of 0.5 feet / minute based on an existing set of operational attributes and that this drilling speed can be increased by increasing the mud flow. In response, the computer system 1151 can operate the pump 1150 to increase the mud flow to increase the drilling speed.
    Example of IT device
    FIG. 12 represents an example of a computer device, according to certain embodiments. A computing device 1200 includes a processor 1201 (possibly comprising several processors, several cores, several nodes and / or implementing multithreading, etc.). The computing device 1200 comprises a memory 1207. The memory 1207 can be a system memory (for example, one or more among a cache memory, SRAM, DRAM, RAM without capacitor, double transistor RAM, eDRAM, EDO RAM, DDR RAM, EEPROM, NRAM, RRAM, SONOS, PRAM, etc.) or any one or more of the possible embodiments already described above of machine-readable supports. The computing device 1200 also includes a bus 1203 (for example, PCI, ISA, PCI
    2017-IPM-101099-U1 -FR
    Express, HyperTransport® bus, InfiniBand® bus, NuBus, etc.) and a 1205 network interface (for example, a Fiber Channel interface, an Ethernet interface, a small internet computer system interface, a SONET interface, a wireless interface wire, etc.).
    The computer device 1200 comprises a well control operation device 1211. The well control operation device 1211 can perform one or more well control operations described above. For example, the wellbore operations controller 1211 may define a wellbore processing attribute which can be controlled based on predicted responses from an RNN. In addition, the wellbore treatment controller 1211 can control one or more wellbore operations of a treatment operation or a drilling operation depending on the value of the well treatment attribute. can be ordered.
    Any of the previously described functionalities can be partially (or entirely) implemented in the hardware and / or on the processor 1201. For example, the functionality can be implemented with an application-specific integrated circuit, in a logic implemented in the processor 1201, in a coprocessor on a peripheral or a card, etc. In addition, the embodiments may include fewer components or additional components not shown in Figure 12 (for example, video cards, audio cards, additional network interfaces, peripherals, etc.). The processor 1201 and the network interface 1205 are coupled to the bus 1203. Although illustrated as being coupled to the bus 1203, the memory 1207 can be coupled to the processor 1201. The computing device 1200 can be a device on the surface and / or integrated into one or more components in the wellbore.
    As will be understood, aspects of the description can be realized as a system, method or program code / instruction stored in one or more machine readable media. As a consequence, the aspects can take the form of hardware, software (including firmware, resident software, microcode, etc.), or a combination of software and hardware aspects which can all be called here "circuit", "module" or "System". The functionalities presented in the form of individual modules / units in the examples illustrated can be organized differently depending on any one among the platform (operating system and / or hardware), the application ecosystem, the interfaces , programmer preferences, programming language, administrator preferences, etc.
    2017-IPM-101099-U1 -FR
    Any combination of one or more machine-readable media can be used. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable storage medium can be, for example, but not limited to, a system, apparatus or device which uses any or a combination of electronic, magnetic, optical, electromagnetic, infrared or semiconductors to store the program code. More specific examples (non-exhaustive list) of machine-readable storage media would include the following: laptop diskette, hard drive, random access memory (RAM), read only memory (ROM), programmable read only memory erasable (EPROM or flash memory), portable compact disc (CD-ROM) read-only memory, optical storage device, magnetic storage device or any suitable combination of the above. In the context of this document, a machine-readable storage medium can be any tangible medium that can contain or store a program to be used by or in connection with a system, apparatus or device for executing instructions. A machine-readable storage medium is not a machine-readable signal medium.
    A machine-readable signal carrier may include a propagated data signal with a machine-readable program code incorporated therein, for example, in baseband or as part of a carrier wave. Such propagated signal can take any of a variety of forms, including, but not limited to, electromagnetic, optical, or any suitable combination thereof. A machine-readable signal medium can be any machine-readable medium which is not a machine-readable storage medium and which can communicate, propagate or transport a program to be used by or in connection with a system, apparatus, or device for executing instructions.
    Program code embedded in machine readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wired, fiber optic cable, RF, etc., or any suitable combination of the above.
    Computer program code for performing operations on aspects of the description can be written in any combination of one or more programming languages, including an object-oriented programming language such as the Java® programming language , C ++ or the like; a dynamic programming language such as Python; a scripting language such as the Peri programming language or the PowerShell scripting language; and classical procedural programming languages, such as language
    2017-IPM-101099-U1 -EN programming "C" or similar programming languages. Program code can run entirely on a stand-alone machine, run distributed over multiple machines, and run on one machine while providing results and / or accepting input on another machine.
    The program code / instructions can also be stored in a machine-readable medium which can control a machine in a particular way, so that the instructions stored in the machine-readable medium produce an article of manufacture comprising instructions which implement the function / act specified in the flowchart and / or block (s) of the functional diagram.
    variations
    Multiple instances can be provided for components, operations, or structures described here as a single instance. Finally, the boundaries between various components, operations and data stores are somewhat arbitrary and particular operations are illustrated in the context of specific illustrative configurations. Other functions are envisaged and may come within the scope of the description. In general, the structures and functionalities presented in the form of separate components in the example configurations can be implemented in the form of a structure or of a combined component. Likewise, the structures and functionalities presented as a single component can be implemented as separate components. These variations, modifications, additions and improvements, and others, may be included in the description.
    The use of the phrase "at least one of" preceding a list with the conjunction "and" should not be considered as an exclusive list and should not be interpreted as a list of categories with an element from each category, unless otherwise specified. A clause that cites "at least one of A, B and C" can be violated with only one of the listed elements, several listed elements and one or more elements of the list and another unlisted element.
    Examples of embodiments:
    Examples of embodiments include the following:
    Embodiment 1: A method comprising: performing a first wellbore processing operation of a wellbore; determining an operational attribute of the well in response to the first wellbore processing operation; determining a predicted response using a recurrent neural network based on the operational attribute; and defining a wellbore processing attribute that can be
    2017-IPM-101099-U1 -FR ordered on the basis of the predicted response; and performing a second wellbore processing operation of the wellbore based on the controllable wellbore processing attribute.
    Embodiment 2: The method according to Embodiment 1, wherein determining the predicted response includes resolving a variation in time and space of the predicted response.
    Embodiment 3: The method according to embodiments 1 or 2, in which the resolution of the variation in time and space of the predicted response comprises the resolution of the variation in time and space between the first wellbore processing operation and the second wellbore processing operation.
    Embodiment 4: The method according to any of embodiments 1 to 3, further comprising: training, prior to determining the predicted response, of the recurrent neural network based on a first value of the operational attribute; detecting that an abnormal wellbore event has occurred; and in response to detecting that the abnormal wellbore event has occurred, re-forming the recurrent neural network based on a second value of the operational attribute and not based on the first value of l operational attribute, wherein the second value of the operational attribute is determined based on a measurement made after the abnormal wellbore event.
    Embodiment 5: The method according to any of embodiments 1 to 4, further comprising determining a training attribute, wherein determining the predicted response is further based on the training attribute .
    Embodiment 6: The method according to any of embodiments 1 to 5, wherein the controllable wellbore processing attribute comprises at least one of the surface treatment pressure, the speed pumping fluid and propellant flow.
    Embodiment 7: The method according to any of embodiments 1 to 6, wherein the recurrent neural network comprises a long-short term memory cell.
    Embodiment 8: One or more non-transient supports readable by a machine comprising a program code, the program code making it possible: to carry out a first drilling well treatment operation of a drilling well; determining an operational attribute of the well in response to the first wellbore processing operation; to determine a predicted response using a recurrent neural network and based on
    2017-IPM-101099-U1 -EN the operational attribute; defining a controllable wellbore processing attribute based on the predicted response; and performing a second wellbore processing operation of the wellbore based on the controllable wellbore processing attribute.
    Embodiment 9: The non-transient machine-readable medium (s) according to embodiment 8, in which the program code for determining the predicted response includes a program code for solving a variation in time and space of the predicted response.
    Embodiment 10: The non-transient support (s) readable by a machine according to embodiments 8 or 9, in which the program code for resolving the variation in time and space of the predicted response includes a program code to resolve the variation in time and space between the first wellbore processing operation and the second wellbore processing operation.
    Embodiment 11: The non-transient machine-readable medium (s) according to any one of embodiments 8 to 10, in which the program code also comprises a program code for: training, before determining the response predicted, the recurrent neural network based on a first value of the operational attribute; detecting that an abnormal wellbore event has occurred; and in response to detection that the abnormal wellbore event has occurred, reforming the recurrent neural network based on a second value of the operational attribute and not on the basis of the first value of the operational attribute, wherein the second value of the operational attribute is determined based on a measurement made after the abnormal wellbore event.
    Embodiment 12: The non-transient machine-readable medium (s) according to any one of embodiments 8 to 11, in which the program code also comprises a program code determining a training attribute, in which the determining the predicted response is further based on the training attribute.
    Embodiment 13: The non-transient machine-readable medium (s) according to any one of embodiments 8 to 12, in which the wellbore processing attribute which can be controlled comprises at least one of the surface treatment pressure, fluid pumping speed and propellant flow.
    Embodiment 14: The non-transient machine-readable medium (s) according to any one of embodiments 8 to 13, in which the recurrent neural network comprises a long-short term memory cell.
    2017-IPM-101099-U1 -FR
    Embodiment 15: A system comprising: a well pump; a processor; a machine-readable medium having a program code executable by the processor for causing the processor to perform a first wellbore processing operation of a wellbore; determining an operational attribute of the well in response to the first wellbore processing operation; determining a predicted response using a recurrent neural network based on the operational attribute; defining a controllable wellbore processing attribute based on the predicted response; and performing a second wellbore processing operation of the wellbore based on the controllable wellbore processing attribute.
    Embodiment 16: The system according to Embodiment 15, wherein the program code executable by the processor to determine the predicted response includes program code to resolve a variation in time and space of the predicted response.
    Embodiment 17: The system according to embodiments 15 or 16, wherein the program code executable by the processor to resolve the variation in time and space of the predicted response includes program code to resolve the variation in time and space between the first wellbore processing operation and the second wellbore processing operation.
    Embodiment 18: The system according to any of embodiments 15 to 17, wherein the program code executable by the processor further includes program code to cause the processor to: train, before determining the response predicted, the recurrent neural network based on a first value of the operational attribute; detecting that an abnormal wellbore event has occurred; and in response to detection that the abnormal wellbore event has occurred, reforming the recurrent neural network based on a second value of the operational attribute and not on the basis of the first value of the operational attribute, wherein the second value of the operational attribute is determined based on a measurement made after the abnormal wellbore event.
    Embodiment 19: The system according to any of embodiments 15 to 18, wherein the program code executable by the processor further comprises program code for causing the processor to determine a training attribute, in which the determination of the predicted response is further based on the training attribute.
    Embodiment 20: The system according to any of Embodiments 15 to 19, wherein the wellbore processing attribute can be controlled
    20174PM-101099-U1-FR 26 includes at least one of the surface treatment pressure, fluid pumping speed and propellant flow.
    2017-IPM-101099-U1 -FR
    权利要求:
    Claims (15)
    [1" id="c-fr-0001]
    1. Process comprising:
    performing a first wellbore processing operation of a wellbore;
    determining an operational attribute of the well in response to the first wellbore processing operation;
    determining a predicted response using a recurrent neural network based on the operational attribute; and defining a controllable wellbore processing attribute based on the predicted response; and performing a second wellbore processing operation of the wellbore based on the controllable wellbore processing attribute.
    [2" id="c-fr-0002]
    The method of claim 1, wherein determining the predicted response includes resolving a time and space variation of the predicted response, wherein resolving the time and space variation of the predicted response includes resolving the time and space variation between the first wellbore processing operation and the second wellbore processing operation.
    [3" id="c-fr-0003]
    3. Method according to claim 1, further comprising:
    training, before determining the predicted response, of the recurrent neural network based on a first value of the operational attribute;
    detecting that an abnormal wellbore event has occurred; and in response to detecting that the abnormal wellbore event has occurred, re-forming the recurrent neural network based on a second value of the operational attribute and not based on the first value of l operational attribute, wherein the second value of the operational attribute is determined based on a measurement made after the abnormal wellbore event.
    [4" id="c-fr-0004]
    The method of claim 1, further comprising determining a training attribute, wherein determining the predicted response is further based on the training attribute.
    2017-IPM-101099-U1 -FR
    [5" id="c-fr-0005]
    The method of claim 1, wherein the controllable wellbore treatment attribute comprises at least one of a surface treatment pressure, a fluid pumping speed and a support agent flow rate. .
    [6" id="c-fr-0006]
    6. The method of claim 1, wherein the recurrent neural network comprises a long-short term memory cell.
    [7" id="c-fr-0007]
    7. Support or non-transient supports readable by a machine comprising a program code, the program code allowing:
    perform a first wellbore treatment operation on a wellbore;
    determining an operational attribute of the well in response to the first wellbore processing operation;
    determining a predicted response using a recurrent neural network based on the operational attribute; and defining a controllable wellbore processing attribute based on the predicted response; and performing a second wellbore processing operation of the wellbore based on the controllable wellbore processing attribute.
    [8" id="c-fr-0008]
    8. Machine readable non-transient medium or carriers according to claim 7, wherein the program code for determining the predicted response comprises program code for resolving a variation in time and space of the predicted response, in which the program code for solving the variation in time and space of the predicted response includes a program code for solving the variation in time and space between the first wellbore processing operation and the second drilling operation drilling well treatment.
    [9" id="c-fr-0009]
    9. Support or non-transient supports readable by a machine according to claim 7, in which the program code also comprises a program code making it possible to:
    forming, before determining the predicted response, the recurrent neural network on the basis of a first value of the operational attribute;
    detecting that an abnormal wellbore event has occurred; and
    2017-IPM-101099-U1 -EN in response to the detection that the abnormal wellbore event has occurred, reform the recurrent neural network based on a second value of the operational attribute and not on the based on the first value of the operational attribute, wherein the second value of the operational attribute is determined based on a measurement made after the abnormal wellbore event.
    [10" id="c-fr-0010]
    10. Machine readable non-transient medium or carriers according to claim 7, wherein the program code further comprises a program code determining a training attribute, wherein the determination of the predicted response is further based on A training attribute, wherein the controllable wellbore treatment attribute includes at least one of the surface treatment pressure, the fluid pumping speed and the propellant flow.
    [11" id="c-fr-0011]
    11. Support or non-transient supports readable by a machine according to claim 7, in which the recurrent neural network comprises a long-short-term memory cell.
    [12" id="c-fr-0012]
    12. System comprising:
    a well pump;
    a processor;
    a machine-readable medium having a program code executable by the processor to cause the processor to perform a first wellbore processing operation of a wellbore;
    determining an operational attribute of the well in response to the first wellbore processing operation;
    determining a predicted response using a recurrent neural network based on the operational attribute; and defining a controllable wellbore processing attribute based on the predicted response; and performing a second wellbore processing operation of the wellbore based on the controllable wellbore processing attribute.
    2017-IPM-101099-U1 -FR
    [13" id="c-fr-0013]
    The system of claim 12, wherein the program code executable by the processor to determine the predicted response includes program code to resolve a variation in time and space of the predicted response, wherein the program code executable by the processor to resolve the variation in time and space of the predicted response includes a program code to resolve the variation in time and space between the first wellbore processing operation and the second drilling operation drilling well treatment.
    [14" id="c-fr-0014]
    14. The system as claimed in claim 12, in which the program code executable by the processor further comprises a program code for causing the processor to:
    forming, before determining the predicted response, the recurrent neural network on the basis of a first value of the operational attribute;
    detecting that an abnormal drilling event has occurred; and in response to detection that the abnormal wellbore event has occurred, reforming the recurrent neural network based on a second value of the operational attribute and not on the basis of the first value of the operational attribute, wherein the second value of the operational attribute is determined based on a measurement made after the abnormal wellbore event.
    [15" id="c-fr-0015]
    The system of claim 12, wherein the program code executable by the processor further comprises program code for causing the processor to determine a training attribute, wherein the determination of the predicted response is further based on the The training attribute, wherein the controllable wellbore treatment attribute comprises at least one of a surface treatment pressure, a fluid pumping speed and a propellant flow.
    2017-IPM-101099-U1-EN
    类似技术:
    公开号 | 公开日 | 专利标题
    EP2951720B1|2018-05-02|Production analysis and/or forecasting methods, apparatus, and systems
    FR3035674A1|2016-11-04|SHADE GEOMECHANICS FOR HYDRAULIC MULTI-STAGE FRACTURING OPTIMIZATION IN SCHOOL RESOURCES AND GREEN GAMES
    Li et al.2019|Deep learning for well data history analysis
    EP2253797A1|2010-11-24|Method of exploitation of a porous medium by modelling of fluid flows
    FR3075434A1|2019-06-21|RECURRENT NEURONAL NETWORK MODEL FOR PUMPING IN SEVERAL STEPS
    FR3062873A1|2018-08-17|AUTOMATION OF THE HIGHER SCALE OF RELATIVE PERMEABILITY AND CAPILLARY PRESSURE IN MULTIPLE POROSITY SYSTEMS
    FR2823877A1|2002-10-25|Production, from dynamic data, of timely geological model of physical property characteristic of structure of underground layer being examined, includes parametrizing fine mesh model to obtain distribution of property
    FR2961844A1|2011-12-30|MULTIPHASE FLOW IN A WELLBORE AND CONNECTED HYDRAULIC FRACTURE
    WO2009034253A1|2009-03-19|Method, programme and computer system for scaling the modelling data for a hydrocarbon deposit
    FR3070180A1|2019-02-22|NEURON NETWORK MODELS FOR REAL-TIME OPTIMIZATION OF DRILLING PARAMETERS DURING DRILLING OPERATIONS
    US20180171769A1|2018-06-21|Structure For Fluid Flowback Control Decision Making And Optimization
    FR3038338A1|2017-01-06|CORRECTION OF DEVIATION AND DISPERSION EFFECTS ON ACOUSTICAL DIAGRAM MEASUREMENTS OF WELLS DUE TO STRATIFIED FORMATIONS
    FR3036820A1|2016-12-02|MODELING THE SATURATION AND PERMEABILITY OF PETROLEUM FIELD RESERVOIR
    FR3081026A1|2019-11-15|LEARNING-BASED BAYESIAN Optimization to Optimize CONTROLLED Drilling Parameters
    FR3054689A1|2018-02-02|METHOD FOR TEMPORALLY DETERMINING A MAXIMUM MAXIMUM SAMPLING WATER VOLUME ADMISSIBLE FROM A GROUNDWATER SOURCE
    EP2963235B1|2017-02-22|Method for exploiting an oil deposit based on a technique for positioning wells to be drilled
    FR3067746A1|2018-12-21|ESTIMATION OF MECHANICAL PROPERTIES OF TRANSVERSE ISOTROPIC MEDIA
    FR3086779A1|2020-04-03|MATCHING AN AUTOMATED PRODUCTION HISTORY USING BAYESIAN OPTIMIZATION
    FR3080210A1|2019-10-18|RECURRENT NEURONAL NETWORK MODEL FOR DOWNHOLE PRESSURE AND TEMPERATURE IN LOWERING ANALYSIS
    FR3085404A1|2020-03-06|AUTOMATED OPTIMIZATION OF THE PENETRATION RATE FOR DRILLING
    US20210054738A1|2021-02-25|Methods for predicting properties of clean formation fluid using real time downhole fluid analysis of contaminated samples
    US10746019B2|2020-08-18|Method to estimate saturation pressure of flow-line fluid with its associated uncertainty during sampling operations downhole and application thereof
    Yang et al.2020|Pattern Recognition of the Vertical Hydraulic Fracture Shapes in Coalbed Methane Reservoirs Based on Hierarchical Bi-LSTM Network
    Sun et al.2020|Application of Both Physics-Based and Data-Driven Techniques for Real-Time Screen-Out Prediction with High Frequency Data
    FR3032222A1|2016-08-05|PRIORI UNCERTAINTY PARAMETER ARCHITECTURE INTEGRATED IN SIMULATION MODEL CREATION
    同族专利:
    公开号 | 公开日
    NO20200537A1|2020-05-07|
    GB202003267D0|2020-04-22|
    WO2019125359A1|2019-06-27|
    CA3071996A1|2019-06-27|
    GB2580243A|2020-07-15|
    US20200248540A1|2020-08-06|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US9176245B2|2009-11-25|2015-11-03|Halliburton Energy Services, Inc.|Refining information on subterranean fractures|
    EP2671167A4|2011-01-31|2018-01-31|Landmark Graphics Corporation|System and method for using an artificial neural network to simulate pipe hydraulics in a reservoir simulator|
    US8805659B2|2011-02-17|2014-08-12|Chevron U.S.A. Inc.|System and method for uncertainty quantification in reservoir simulation|
    AU2013377864B2|2013-02-11|2016-09-08|Exxonmobil Upstream Research Company|Reservoir segment evaluation for well planning|
    US20170145793A1|2015-08-20|2017-05-25|FracGeo, LLC|Method For Modeling Stimulated Reservoir Properties Resulting From Hydraulic Fracturing In Naturally Fractured Reservoirs|US11268370B2|2018-03-26|2022-03-08|Baker Hughes, A Ge Company, Llc|Model-based parameter estimation for directional drilling in wellbore operations|
    CA3140567A1|2019-05-17|2020-11-26|Schlumberger Canada Limited|System and method for managing wellsite event detection|
    法律状态:
    2019-10-30| PLFP| Fee payment|Year of fee payment: 2 |
    2021-07-09| ST| Notification of lapse|Effective date: 20210605 |
    优先权:
    申请号 | 申请日 | 专利标题
    PCT/US2017/066974|WO2019125359A1|2017-12-18|2017-12-18|Recurrent neural network model for multi-stage pumping|
    USWOUS2017066974|2017-12-18|
    [返回顶部]