• 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 real-time drilling optimization method with training using a multilayer deep neural network (DNN) created from input drilling data. A plurality of drilling parameter characteristics are extracted using the DNN. A linear regression model is created based on the plurality extracted from drilling parameter characteristics. The linear regression model is applied to predict one or more drilling parameters. Figure to be published with the abstract: Fig. 1
    公开号:FR3081026A1
    申请号:FR1902256
    申请日:2019-03-05
    公开日:2019-11-15
    发明作者:Srinath Madasu；Keshava Rangarajan
    申请人:Landmark Graphics Corp；
    IPC主号:
    专利说明:

    Description
    Title of the invention: BAYESIAN OPTIMIZATION BASED ON LEARNING to optimize drilling parameters THAT CAN BE CONTROLLED
    Technical Field [0001] The embodiments described herein generally relate to land formation drilling operations and, more particularly, Bayesian optimization to optimize drilling parameters that can be controlled.
    BACKGROUND OF THE INVENTION In drilling operations, conventional drilling processes are relatively complex and involve considerable expense. The industry is continually striving to improve safety, cost reduction and efficiency, particularly with regard to the characterization of hydrocarbon reservoirs and optimization of drilling. However, more efficient, improved and optimized drilling processes remain necessary.
    BRIEF DESCRIPTION OF THE DIFFERENT VIEWS OF THE DRAWINGS For a more complete understanding of the embodiments described, and for other advantages thereof, reference is now made to the following description, taken in conjunction with the attached drawings, in which:
    [Fig.l] is a diagram of a drilling system according to certain embodiments of the present invention;
    [Fig.2] is a flowchart for optimizing drilling parameters which can be controlled using a combination of deep neural network and a regressor produced using the drilling system of [Fig. 1], according to an embodiment of the present invention;
    [Fig.3A] is a diagram illustrating an embodiment of a deep neural network, according to an embodiment of the present invention;
    [Fig.3B] illustrates a schematic representation of the connections in stacked LSTM cells constituting a deep recurrent neural network according to an embodiment of the present invention; and [0009] [fig.4A] [fig.4B] illustrate a comparison between the best real and predicted points with and without range constraints, respectively, according to an embodiment of the present invention.
    DETAILED DESCRIPTION OF THE DESCRIBED EMBODIMENTS The following description is presented to enable those skilled in the art to make and use the invention. Various modifications will be more apparent to those skilled in the art and the general principles described herein can be applied to embodiments and applications other than those detailed below without departing from the spirit and scope described embodiments as defined herein. The embodiments described are not intended to be limited to the particular embodiments indicated, but must be accorded full scope compatible with the principles and characteristics described herein.
    The term "top well" as used herein means along the drill string or hole from the distal end to the surface, and "bottom of the well" or "bottom of the well" as 'used herein means along the drill string or hole from the surface to the distal end.
    It will be understood that the terms "oil well drilling equipment" or "oil well drilling system" are not intended to limit the use of the equipment and processes described by these terms for drilling from an oil well. The terms also include drilling for natural gas wells or oil wells in general. In addition, these wells can be used for production, monitoring or injection as part of the recovery of hydrocarbons or other materials from the underground. This could also include geothermal wells intended to provide a source of thermal energy instead of hydrocarbons.
    As indicated above, more efficient, improved and optimized drilling processes remain necessary. Embodiments of the present invention provide apparatus and methods for characterizing hydrocarbon reservoirs and drilling optimization using new Bayesian optimization (BO) based on learning with range constraints. The BO method described with range constraints predicts the optimal controllable parameters necessary for drilling optimization. The determination of optimal drilling parameters, such as instantaneous penetration rate (ROP), drill bit weight (WOB) and optimal rotations per minute (RPM), is calculated for training during drilling using the BO process with range constraints and drilling parameters are adjusted to optimal WOB and RPM. The method described provides fast, robust and accurate prediction using discrete data as input.
    The methodology described uses a deep learning technique based on a neural network which helps to quickly and efficiently calculate the parameters that can be ordered optimal optimal and to use more parameters for an automated control in real time of the parameters ROP, RPM , WOB, etc. The deep learning technique described is rapid, in part because it does not need an objective function to provide during the prior training of the neural network. Using a deep neural network (DNN) in combination with a regressor further helps to perform a quick and efficient calculation. In some implementations, the technique described is at least three times faster than advanced Gaussian process (GP) modeling.
    Referring now to [Fig. 1], a drilling system 100 includes a drilling platform 102 disposed at the top of a borehole 104. A logging tool 106 is carried by a sub-element 108, generally a drilling collar, incorporated in a train of borehole 110 and disposed inside the borehole 104. A drill bit 112 is located at the lower end of the drill string 110 and cuts a borehole 104 through the earth formations 114. The mud of borehole 116 is pumped from a storage tank 118 near the wellhead 120, under an axial passage (not shown) through the drill string 110, out of the drill bit openings 112 and returned to the surface by the region annular 122. The metal casing 124 is positioned in the borehole 104 above the drill bit 112 to maintain the integrity of an upper part of the borehole 104.
    Referring to [Fig. 1], the annular 122 located between the drill pipe 110, the sub-element 108 and the side walls 126 of the borehole 104 constitutes the return flow path for the drilling mud. The mud is pumped from the storage basin near the wellhead 120 by the pumping system 128. The mud passes through a mud supply pipe 130 which is coupled to a central passage extending over the entire length of the train of drilling 110. In this way, the drilling mud is forced to flow down into the drilling string 110 and enters the drilling hole through openings in the drilling bit 112 to cool and lubricate the drilling bit. drilling and bringing to the surface the formation spoil produced during the drilling operation. A fluid exhaust duct 132 is connected from the annular passage 122 at the level of the wellhead to conduct the flow of mud returning from the borehole 104 to the mud basin 118. The drilling mud is generally handled and treated by various devices (not shown) such as degassing units and circulation tanks to maintain a preset viscosity and consistency of mud.
    The logging tool or instrument 106 can be any conventional logging instrument such as an acoustic (sometimes called sonic), neutronic, gamma ray, density, photoelectric, nuclear magnetic resonance logging instrument , or any other conventional logging instrument, or a combination thereof, which can be used to measure the lithology or porosity of the formations surrounding an earth borehole.
    The logging instrument being integrated in the drill string 110 of [FIG. 1], the system is considered a measurement during drilling (MWD) system, that is to say that it records while the drilling process is in progress. Logging data can be stored in a conventional downhole recorder (not shown), which can be accessed at ground level when drill string 110 is retrieved, or which can be transmitted to the ground surface at telemetry aid such as conventional pulse mud telemetry systems. In either case, the logging data from the logging instrument 106 ultimately reaches a surface measurement device processor 134 to allow processing of the data for use in accordance with the embodiments of the present invention described herein. In other words, the measurement processor 134 processes the log data appropriately for use with the embodiments of the present invention.
    In addition to MWD instrumentation, wired line logging instrumentation can also be used. In other words, wireline instrumentation can also be used to record the formations around the borehole as a function of depth. With wired line instrumentation, a wired line truck (not shown) is usually located on the surface of a wellbore. A wired line logging instrument is suspended in the borehole by a logging cable which passes over a pulley and a depth measurement sleeve. When the logging instrument crosses the borehole, it records the formations surrounding the borehole as a function of depth. The log data is transmitted by a log cable to a processor located on or near the log truck to process the log data appropriately for use with the embodiments of the present invention. As with the MWD embodiment of [Fig. 1], the wired line instrumentation can include any conventional logging instrumentation which can be used to measure the lithology and / or porosity of formations surrounding an earth borehole, for example, acoustic, neutron logging instrumentation, gamma rays, density, photoelectric, nuclear magnetic resonance, or any other conventional logging instrument, or combinations thereof, which can be used to measure lithology.
    Referring again to [Fig. 1], a drilling control system 140 is shown. The drilling control system 140 includes a prescribed set of geology and drilling mechanics. The drilling control system 140 further includes a device generally referred to herein as processor 142 and comprising any suitable commercially available computer, control device or data processing apparatus having a processor and a memory device coupled to the processor or accessible by it. The memory device, which may be part of the processor 142, contains a set of instructions for implementing the method and the apparatus as described in more detail herein. The processor 142 receives input from any suitable input device (s) 148. The input device (s) 148 may (may) include a keyboard, a numeric keypad, a pointing device, or a similar item, further including a network interface or other communications interface for receiving input information from a remote computer or database. The processor 142 transmits information signals and / or equipment control commands. The output signals can be output to a display device 150 via signal lines 144 for use in generating a display of the information contained in the output signals. The output signals can also be output to a printing device 152 for use in generating a printout 154 of the information contained in the output signals. The information and / or control signals can also be transmitted via signal lines 156 if necessary, for example, to a remote device to be used in the control of one or more drilling operating parameters of the drilling platform 102, as described in more detail herein. In other words, a suitable device or means is provided on the drilling system which responds to a predicted drilling machine output signal to control a parameter in an actual drilling of a wellbore (or of a interval) with the drilling system. For example, a drilling system may include equipment such as one of the following types of controllable motors selected from a downhole motor 160, an upper drive motor 162 or a rotary table motor 164, in which a rotation per given minute of a respective motor can also be controlled remotely. The parameter may also include one or more of the following elements selected from the group consisting of drill bit weight, rotations per minute, flow rate of the mud pump, hydraulic systems or any other control system parameter suitable drilling.
    The processor 142 is programmed to execute functions as described herein, using programming techniques known in the art. In one embodiment, computer readable medium is included, the computer readable medium having a computer program stored thereon. The computer program to be executed by the processor 142 is intended to optimize the drilling. The computer program includes instructions for creating a multilayer DNN from input drilling data. The computer program also includes instructions for extracting a plurality of drilling parameter characteristics from geological data using the DNN. The computer program further includes instructions for creating a linear regression model based on the extracted plurality of characteristics of drilling parameters. Finally, the computer program includes instructions for applying the linear regression model to predict one or more drilling parameters. The programming of the computer program to be executed by the processor 142 can also be carried out using known programming techniques to implement the embodiments described and discussed herein. In addition, associated with the knowledge of optimized drilling parameters that can be controlled, the drilling operation can be advantageously optimized, as discussed in more detail below.
    In a preferred embodiment, the geological data include at least the resistance of the rock. In another embodiment, the geological data may further include any of the following: logging data, lithology, porosity and plasticity of the shale.
    The input device 148 can be used to enter the specifications of the proposed drilling equipment intended for use in drilling the wellbore (or the interval of the wellbore). In a preferred embodiment, the specifications include at least one drill bit specification of a recommended drill bit. In another embodiment, the specifications may also include one or more specifications of the following equipment which may include a downhole motor, a top drive motor, a rotary table motor, a mud system and a pump with mud. Corresponding specifications may include maximum torque output, type of mud, or mud pump output power, for example, if particular drilling equipment requires it.
    In a preferred embodiment, the predicted drilling mechanics can include wear of the drill bit, mechanical efficiency, power and operating parameters. In another embodiment, the operating parameters may include weight on drill bit (WOB), revolutions per minute of rotation (revolutions per minute), cost, penetration rate and torque. The penetration rate also includes an instantaneous penetration rate (ROP) and an average penetration rate (ROP-A VG).
    [Lig. 2] is a flowchart for optimizing the drilling carried out by the drilling system of [Pig. 1], according to an embodiment of the present invention. Before going on to the description of [Lig. 2], it should be noted that the flowchart in this figure shows an example in which the operational steps are carried out in [0027] [0029] [0030] [0031] [0032] a particular order, as indicated by the lines connecting the blocks, but the different stages represented in this flowchart can be carried out in any order, or in any combination or sub-combination. It should be noted that, in certain embodiments, some of the steps described below can be combined in a single step. In some embodiments, one or more additional steps can be performed.
    The drilling control system 140 starts the process described in step 202 by receiving discrete drilling data related to the designed engineering constraints. In some embodiments, this data may be stored in a database (not shown), which may be part of the drilling control system 140. Non-limiting embodiments of the discrete drilling data include the WOB, the RPM and the flow rate of the drilling fluid. These drilling parameters are generally known and can be constant.
    In step 204, the discrete data received can then be entered by the drilling control system 140 into a neural network module, which can be executed on site (for example, in processor 142) or at a remote location. The neural network module can include any of a deep neural network (DNN), a convolutional neural network (CNN), a long and short term memory block (LSTM), a chronological convolutional neural network (TCNN), a time-frequency CNN (TLCNN) and a fused CNN (fCNN), some of which will be discussed below.
    The neural network module can then be used to extract characteristics of drilling parameters from the input data of a regressor in step 206. Non-limiting embodiments of the regressor include a linear regressor, a machine support vectors (SVM) with a radial basic function nucleus (RBL) or a polynomial. Support vector machines are generally used for machine learning classification and a wide range of kernel functions are available for specific problem classes. SVMs are relatively robust training devices and are numerically stable for the most common kernel functions. In some embodiments, the drilling control system 140 uses an SVM with a core defined by a radial base function in the form [Math 1]:
    l t a, λ) - αχμ - 2 2σ I where x, x 'are the vectors of characteristics and σ constitutes a free parameter.
    In step 208, the drilling control system 140 creates or generates a thematic model from the regressor. The mathematical model generated represents the structure of the drill string and the forces acting on the drill string. It can be understood that various types of mathematical models can be used with different levels of precision or complexity in the representation of the drill string. In one or more embodiments, a mathematical model including statistical interaction terms is adapted to the data observed using Bayesian linear regression techniques, in which prior knowledge is used to determine the posterior probability distributions of the model . The term "Bayesian linear regression" refers to an approach to linear regression in which statistical analysis is undertaken in the context of Bayesian inference. The prior belief function of the linear regression model, including the prior probability distribution function of the model parameter, is combined with the data likelihood function according to Bayes' theorem to obtain the posterior probability distribution over the settings.
    In step 210, the drilling control system 140 applies a range of constrained data using engineering constraints (step 202) to predict one or more drilling parameters. For example, to avoid vibrations of the downhole assembly tool, certain ranges of revolutions per minute must be avoided for a given WOB. This and other best drilling practices can be the range constraint for optimization. Then, in step 212, the drilling control system 140 maximizes the multivariable expected improvement (IE) values for the new observations. In one embodiment, the new observations should be compared to the best current predicted value of one or more drilling parameters x A , found as a parameter value setting x A * which maximizes a new variable El multiples for Bayesian optimization, given by the following equation (1):
    [Math. 2 min . , £ Ι β0 = σ _J (f mir: -z) 0 (z) dz [0035] or
    is the cumulative distribution function, is the probability density function, [Math. 3] f '_ Lniln ^ JL 7 _ Ύ' min &'D' μ is the mean and σ is the variance.
    In step 214, the drilling control system 140 updates at least the sampling points and the observations on the basis of the maximized expected improvement determined in step 212. Next, the drilling system drill command 140 automatically updates the values of one or more drilling parameters based on the maximized expected improvement value (step 216). Examples of these controllable drilling parameters include, but are not limited to, WOB, flow of drilling fluid through the drill pipe, speed of rotation of the drill string, as well as the density and the viscosity of the drilling fluid. In summary, the drilling control system 140 performs steps 202-216 to monitor a particular feature of the downhole operation when it is performed on each of the plurality of operating intervals and adjusts one or more parameters operational in order to optimize the downhole operation in relation to the particular characteristic monitored.
    [Fig. 3A] illustrates a fully connected deep neural network (DNN) 300 model which can be implemented in accordance with the embodiments of the present invention. The DNN 300 includes a plurality of nodes 302, organized into an input layer 304, a plurality of hidden layers 306 and an output layer 308. Each of the layers 304, 306, 308 is connected by node outputs 310. On will understand that the number of nodes 302 indicated in each layer 304, 306, 308 is indicated by way of example and is in no way limiting. Consequently, the number of nodes 302 in each layer can vary between 1000 and 2000 nodes 302. Likewise, the number of hidden layers 306 illustrated is also given by way of example and can vary between four and six hidden layers 306 In addition, although the DNN 300 illustrated is shown as fully connected, the DNN 300 may have other configurations, including a partially connected configuration.
    To give an overview of the DNN 300, one or more characteristic vectors 303 can be entered in the nodes 302 of the input layer 304. Each of the nodes 302 can correspond to a mathematical function having adjustable parameters . All nodes 302 can have the same scalar function, differing only in possibly different parameter values, for example. Alternatively, the various nodes 302 could have different scalar functions depending on the location of the layer, input parameters or other discriminating characteristics. For example, mathematical functions could take the form of sigmoid functions. It will be understood that other functional forms could be used in addition or in a variant. Each of the mathematical functions can be configured to receive one input or several inputs and, from the input or several inputs, evaluate or calculate a scalar output. In the case of a sigmoid function, each node 302 can calculate a sigmoidal non-linearity of a weighted sum of its inputs.
    Therefore, the nodes 302 of the input layer 304 assimilate the vectors of characteristics 303 and then produce the outputs of node 310, which are delivered sequentially by the hidden layers 306, the outputs of node 310 of the layer input 304 being directed towards the nodes 302 of the first hidden layer 306, the node outputs 310 of the first hidden layer 306 being directed towards the nodes 302 of the second hidden layer 306, etc. Finally, the nodes 302 of the final hidden layer 306 can be delivered to the output layer 308, which can then issue the prediction 311 for the particular commanded drilling parameter (s).
    Before using the DNN 300 at the time of execution, the DNN 300 can be trained with labeled or transcribed data, including one or more drilling parameters. For example, during training, a predicted drilling parameter value 311 may be labeled or previously transcribed. As such, prediction 311 can be applied to DNN 300, as described above, and node outputs 310 from each layer, including prediction 311, can be compared to expected or "true" output values.
    As shown in the illustration, the DNN 300 is considered to be "fully connected" because the node output 310 of each node 302 of the input layer 304 and of the hidden layers 306 is connected to the input of each node 302 in the next hidden layer 306 or in the output layer 308. Therefore, each node 302 receives its input values from a previous layer 304, 306, with the exception of nodes 302 in the layer of input 304 which receive the feature vectors 303 from the feature extraction module 202, as described above.
    In another exemplary embodiment, the DNN 300 can be implemented in the form of a block of long and short term memory memory (LSTM). Each LSTM memory block can include one or more LSTM memory cells and each LSTM memory cell can generate an aggregated cell output to generate the LSTM output for the time step. [Eig. 3B] illustrates a schematic representation of the connections between stacked LSTM cells 312a, 312b constituting a deep recurrent neural network according to an embodiment of the present invention.
    In [Eig. 3B], p t represents a drilling parameter variable (such as ROP) at different time steps. More specifically, p * t _ 2 313a and p 2 t2 313b represent the values of the drilling parameters in time step t-2, p * ti 313c and p 2 , i 313d represent the values of the drilling parameters in time step t-1 and p't313e and p 2 t 313f represent drilling parameter values at time step t. The x 315 input is then passed to the deep LSTM recurrent neural network to predict the drilling parameters. It has been observed that the present embodiments as described provide a predictive system which achieves greater accuracy than that of conventional predictive systems. In the embodiment shown in [Fig. 3B], the input x 315 includes the instantaneous penetration rate (r R0P ), the weight on the drill bit (r WOB ) and the flow rate (r Q ) and is shared by all the stacked layers 312a and 312b. Each horizontal row 314a, 314b of the LSTM cells 312a, 312b represents a deep RNN layer and each vertical section 316a, 316b represents an individual time step.
    According to an embodiment of the present invention, the state of cell C 322 and the predicted output generated (variable p 313) of an individual layer in the deep RNN are transmitted to the next step in the same layer and form the basis of the input formulation at the next time step. In other words, the cell states c * t _i 322c and c 2 t_i 322d and the output of the predicted variable generated p * t_i 313c and p 2 (, 313d are transmitted from cells 312a and 312b to the respective cells 312c and 312d in the same layers 314a and 314b. The final value of the drilling parameter p (for example, the instantaneous penetration rate) is obtained by the combination of the predicted variable outputs p * t 313e and p 2 t 313f of all the stacked layers 314a to 314b at the last time step 316b. In various embodiments, the respective outputs can be combined using methods of loss of error of mean square root and / or BPTT (backpropagation in time) known in the field. , among others. Thus, a prediction model based on deep learning, such as stacked LSTM or other variants of deep RNN (depending on the implementation), makes it possible to capture highly nonlinear variations in data from the This series property of the prediction model based on deep learning makes it perfectly suited to the real-time prediction of one or more drilling parameters on the basis of information collected during multi-stage drilling operations.
    [Fig. 4A] and [Fig. 4B] illustrate a comparison between the actual optimal drilling parameter value and the predicted optimal drilling parameter value or the best point, in accordance with embodiments of the present invention. [Fig. 4A] shows the comparison between the best real point and predicted without range constraint. As shown in [Fig. 4A], the prediction 311 calculated by the drilling control system 140 is very close to the actual optimum value of the drilling operating parameter. The predicted value 404 of the drilling operating parameter (for example, ROP) is 1.1, while the actual optimal value 402 of the drilling operating parameter is 1.0. According to an embodiment of the present invention, the drilling control system 140 calculates the predicted value of ROP using the following equation (2):
    [Math. 4]
    ROP = (WOB RPM) 1.12 [0048] [Fig. 4B] shows the comparison between the best real and predicted point with range constraints. The range constraints (applied in step 210) impose large gradients that the DNN 300 can capture with more hidden layers 306 and nodes 302. In other words, it is possible to impose small modifications to the parameters using gradient clipping, which controls the explosion of the gradient and uses regularization. Likewise, according to an embodiment of the present invention, the drilling control system 140 calculates the predicted value of ROP using the equation (1) indicated above. In the example illustrated, the drilling control system 140 applied the constraints specific to a domain with the zero value of ROP between the values 0.99 and 1.0. In the illustrated case, the actual optimal value 406 is approximately equal to 0.9899 and the predicted value 408 of the drilling operating parameter (for example, the ROP) is 0.9. According to an embodiment of the present invention, the drilling control system 140 can improve the results of the prediction performed using a hyper optimization technique of the DNN 300.
    Consequently, as indicated above, the embodiments described herein can be implemented in several ways. In general, in one aspect, the embodiments described relate to a method of optimizing drilling. The method includes, among other steps, the steps: (i) creating a multi-layered deep neural network (DNN) from real-time input drilling data; (ii) extracting a plurality of characteristics of drilling parameters from the DNN; (iii) creating a linear regression model based on the extracted plurality of characteristics of drilling parameters; and (iv) applying the linear regression model to predict one or more drilling parameters.
    In one or more embodiments, the drilling optimization method can also individually include any one of the following characteristics or a combination of two or more of these characteristics: (a) the application step the linear regression model, further comprising applying a constrained data range to predict the one or more drilling parameters; (b) DNN comprising a convolutional neural network (CNN); (c) the linear regression model comprising a linear support vector machine (SVM) model; (d) the SVM model further comprising an SVM model with a radial base function (RBF) core; and (e) the step of maximizing an expected improvement value based on the linear regression model, the maximum expected improvement corresponds to a predicted value of the one or more drilling parameters.
    In general, in yet another aspect, the embodiments described relate to a drilling control system. The system includes a processor and a memory device coupled to the processor. The memory device contains a set of instructions which, when executed by the processor, cause the processor to: (i) create a multilayer deep neural network (DNN) from real-time input drilling data ; (ii) extracting a plurality of characteristics of drilling parameters from the DNN; (üi) creating a linear regression model based on the extracted plurality of characteristics of drilling parameters; and (iv) apply the linear regression model to predict one or more drilling parameters.
    In one or more embodiments, the drilling control system may further individually include any of the following characteristics or a combination of two or more of these characteristics: (a) the set of instructions which causes the processor to apply the linear regression model, further causing the processor to apply a constrained range of data to predict the one or more drilling parameters; (b) DNN comprising a convolutional neural network (CNN); (c) the linear regression model comprising a linear support vector machine (SVM) model; (d) the SVM model further comprising an SVM model with a radial base function (RBF) core; and (e) the set of instructions which further causes the processor to maximize an expected improvement value based on the linear regression model, the maximum expected improvement corresponds to a predicted value of the one or more drilling parameters. .
    Although particular aspects, implementations and applications of the present invention have been illustrated and described, it should be understood that the present invention is not limited to the construction and to the precise compositions described herein , and that various modifications, various changes and various variations may appear from the preceding descriptions without departing from the spirit and scope of the embodiments described as defined in the claims appended hereto.
    权利要求:
    Claims (1)
    [1" id="c-fr-0001]
    claims [Claim 1] Method for optimizing the drilling of a well, the method comprising the steps:creating a multi-layered deep neural network (DNN) from real-time input drilling data from the well;extracting a plurality of drilling parameter characteristics from real-time input drilling data using DNN; creating a linear regression model based on the extracted plurality of characteristics of drilling parameters;applying the linear regression model to the real-time input data to predict one or more drilling parameters for the well; and drilling the well using one or more drilling parameters. [Claim 2] The method of claim 1, wherein the step of applying the linear regression model further comprises applying a constrained data range to the real-time input drilling data to predict the one or more parameters of drilling. [Claim 3] The method of claim 1, wherein the DNN comprises a convolutional neural network (CNN). [Claim 4] The method of claim 1, wherein the linear regression model comprises a linear machine vector model (SVM). [Claim 5] The method of claim 4, wherein the SVM model comprises an SVM model with a radial base function (RBF) core. [Claim 6] The method of claim 1, further comprising determining an expected improvement value based on the linear regression model, wherein the expected improvement value corresponds to a predicted value of the one or more drilling parameters. [Claim 7] The method of claim 1, wherein the one or more drilling parameters comprise one or more: of a weight on bit (WOB), of a number of rotations per minute of bit (RPM), of a rate of flow (Q) and a penetration rate (ROP). [Claim 8] The method of claim 6, further comprising continuously updating the one or more drilling parameters based on the expected improvement value in real time during a drilling operation. [Claim 9] A drilling control system for a well, the system comprising
    a processor and a memory device coupled to the processor, the memory device containing a set of instructions which, when executed by the processor, cause the processor: to control a downhole tool disposed inside the well to obtain real-time input drilling data from the well; creating a multi-layered deep neural network (DNN) from real-time input drilling data from the well;extracting a plurality of drilling parameter characteristics from the real-time input drilling data using the DNN;creating a linear regression model based on the extracted plurality of characteristics of drilling parameters;applying the linear regression model to the real-time input data to predict one or more drilling parameters; and drilling the well using one or more drilling parameters. [Claim 10] The system of claim 9, wherein the instruction set which causes the processor to apply the linear regression model further causes the processor to apply a constrained range of data to the real-time input drilling data in order to predict one or more drilling parameters. [Claim 11] The system of claim 9, wherein the DNN comprises a convolutional neural network (CNN). [Claim 12] The system of claim 9, wherein the linear regression model comprises a linear vector machine (SVM) model. [Claim 13] The system of claim 12, wherein the SVM model comprises an SVM model with a radial base function (RBF) core. [Claim 14] The system of claim 9, wherein the instruction set further causes the processor to determine an expected improvement value based on the linear regression model, wherein the expected improvement value corresponds to a predicted value of one or more drilling parameters. [Claim 15] The system of claim 9, wherein the one or more drilling parameters comprise one or more: of a weight on bit (WOB), of a number of rotations per minute of bit (RPM), of a flow flow (Q) and a penetration rate (ROP).
    1/6
    类似技术:
    公开号 | 公开日 | 专利标题
    FR3081026A1|2019-11-15|LEARNING-BASED BAYESIAN Optimization to Optimize CONTROLLED Drilling Parameters
    FR3070180A1|2019-02-22|NEURON NETWORK MODELS FOR REAL-TIME OPTIMIZATION OF DRILLING PARAMETERS DURING DRILLING OPERATIONS
    FR2869067A1|2005-10-21|SYSTEM AND METHOD FOR FIELD SYNTHESIS FOR OPTIMIZING A DRILLING DEVICE
    FR3027049A1|2016-04-15|PREDICTION OF DOWNHOLE TOOL DEFECT INDUCED BY TEMPERATURE CYCLE
    FR2866922A1|2005-09-02|SYSTEM AND METHOD FOR DRILLING WELLS FROM AN REMOTE CONTROL CENTER
    FR2910922A1|2008-07-04|PUMP CONTROL FOR LAYER TESTING
    US20200277848A1|2020-09-03|Well planning system
    US10719893B2|2020-07-21|Symbolic rigstate system
    FR3037352A1|2016-12-16|
    FR3085053A1|2020-02-21|PREDICTION OF OIL TANK BEHAVIOR USING A PROXY FLOW MODEL TECHNICAL AREA
    FR3027050A1|2016-04-15|AUTOMATED FRACTURING PLANNING METHODS FOR MULTI-WELL FIELDS
    FR3031131A1|2016-07-01|REAL-TIME PERFORMANCE ANALYZER FOR DRILLING OPERATIONS
    FR3075434A1|2019-06-21|RECURRENT NEURONAL NETWORK MODEL FOR PUMPING IN SEVERAL STEPS
    FR3086779A1|2020-04-03|MATCHING AN AUTOMATED PRODUCTION HISTORY USING BAYESIAN OPTIMIZATION
    FR3085404A1|2020-03-06|AUTOMATED OPTIMIZATION OF THE PENETRATION RATE FOR DRILLING
    US20200149354A1|2020-05-14|Drill bit repair type prediction using machine learning
    FR3084102A1|2020-01-24|ADJUSTING THE FUNCTIONING OF A WELL TOOL FOR HANDLING THE PENETRATION RATE | OF A DRILL BIT ON THE BASIS OF MULTIPLE ROP FORECASTS
    FR3080210A1|2019-10-18|RECURRENT NEURONAL NETWORK MODEL FOR DOWNHOLE PRESSURE AND TEMPERATURE IN LOWERING ANALYSIS
    FR3059704A1|2018-06-08|INTELLIGENT RESPONSE IN REAL TIME TO CHANGES IN EQUILIBRIUM OF A PETROLEUM FIELD
    FR3070179A1|2019-02-22|OPTIMIZATION OF PENETRATING SPEED FOR WELLBARS USING MACHINE LEARNING
    FR3075858A1|2019-06-28|RELIABLE AND PRECISE DETECTION OF A-COUP USING REAL-TIME DRILLING DATA
    FR3059108A1|2018-05-25|AUTOMATIC MULTI-ZONE HORIZON TRACKING
    US20210302612A1|2021-09-30|Dynamic field operations system
    FR3040509A1|2017-03-03|
    FR3034546A1|2016-10-07|
    同族专利:
    公开号 | 公开日
    NO20200987A1|2020-09-09|
    WO2019216891A1|2019-11-14|
    GB202014145D0|2020-10-21|
    GB2585581A|2021-01-13|
    CA3093668A1|2019-11-14|
    US20210047910A1|2021-02-18|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    GB2371625B|2000-09-29|2003-09-10|Baker Hughes Inc|Method and apparatus for prediction control in drilling dynamics using neural network|
    US9022140B2|2012-10-31|2015-05-05|Resource Energy Solutions Inc.|Methods and systems for improved drilling operations using real-time and historical drilling data|
    US10519759B2|2014-04-24|2019-12-31|Conocophillips Company|Growth functions for modeling oil production|
    CN103967478B|2014-05-21|2017-10-27|北京航空航天大学|A kind of peupendicular hole meteor trail echoes method based on conducting probe|
    AU2014396852B2|2014-06-09|2018-05-10|Landmark Graphics Corporation|Employing a target risk attribute predictor while drilling|US20210372259A1|2020-05-26|2021-12-02|Landmark Graphics Corporation|Real-time wellbore drilling with data quality control|
    RU2735794C1|2020-06-23|2020-11-09|Федеральное государственное автономное образовательное учреждение высшего образования "Южно-Уральский государственный университет " ФГАОУ ВО "ЮУрГУ "|Method for prediction of sticking of drilling pipes|
    RU2753289C1|2020-10-20|2021-08-12|Федеральное государственное автономное образовательное учреждение высшего образования «Южно-Уральский государственный университет »|Method for predicting sticking of drilling pipes in process of drilling borehole in real time|
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    WOPCT/US2018/031757|2018-05-09|
    PCT/US2018/031757|WO2019216891A1|2018-05-09|2018-05-09|Learning based bayesian optimization for optimizing controllable drilling parameters|
    [返回顶部]