Variable well location deployment SAGD optimization method and device

Abstract

The invention relates to the technical field of oil sand and super heavy oil reservoir development, in particular to a variable well position deployment SAGD optimization method and device. According to the method, the variable well position deployment SAGD development mode different from a conventional SAGD development mode is considered, namely, the SAGD encrypted production horizontal well development mode and the dislocation well pair SAGD development mode are considered, and the SAGD development effect of the oil sand and super heavy oil reservoir can be further improved. By utilizing secondary coupling of a support vector regression method and a genetic algorithm in the field of machine learning, a variable well location deployment SAGD optimization design workflow based on machine learning is established, and rapid, reliable and intelligent optimization design of related well location deployment parameters and operation parameters of a variable well location deployment SAGD development mode can be realized. The method is suitable for well position deployment and operation parameter design of an oil sand and super heavy oil reservoir variable well position deployment SAGD development mode, and technical support is provided for efficient decision making and benefit development of a production site.

Description

Technical Field

[0001] This disclosure relates to the field of oil sands and extra-heavy oil reservoir development technology, and in particular to a method and apparatus for optimizing SAGD deployment with variable well location. Background Technology

[0002] The conventional well deployment mode of steam-assisted gravity drainage (SAGD) technology involves placing a pair of horizontal wells parallel to each other, one above the other, near the bottom of the reservoir. The upper horizontal well is the steam injection well, and the lower well is the production well. The longitudinal distance between the two wells is usually 5-7 meters. In application, steam is continuously injected into the formation through the upper injection well. The steam gradually expands in the formation to form a steam chamber. During the expansion of the steam chamber, heat exchange occurs between the steam chamber and the crude oil in the formation. The heated crude oil and steam condensate are discharged into the production well under the action of gravity and extracted.

[0003] Conventional dual-horizontal-well SAGD technology exhibits low thermal efficiency and poor development performance in thin reservoirs, reservoirs with poor physical properties, reservoirs with high water-cut layers, and reservoirs with shale interlayers. Therefore, variable-well-location SAGD development methods can be adopted, namely, SAGD infill production horizontal well development mode and staggered well SAGD development mode, to improve development results. However, variable-well-location SAGD development methods involve many variable settings, including well location parameters and well-related operational parameters. Relying solely on experience or conventional optimization design methods cannot achieve satisfactory design results, thus affecting the final development outcome.

[0004] Conventional optimization design methods mainly include single-factor analysis and full-factor experimentation. These methods consider the numerical changes of individual variables one by one for research and analysis, but cannot take into account the coupling effects between various factors. Furthermore, if there are too many factors, full-factor experimentation requires a huge amount of computation and manual work, making it impossible to achieve rapid optimization design in the SAGD development mode with variable well location deployment. Summary of the Invention

[0005] To address the aforementioned issues, this disclosure provides a method and apparatus for optimizing SAGD deployment with variable well locations, enabling rapid and reliable optimization design of well location deployment parameters and operating parameters related to the SAGD development mode with variable well locations, thus providing strong technical support for production sites.

[0006] Firstly, a method for optimizing SAGD deployment in variable well locations, the method comprising:

[0007] Based on the variable well location deployment SAGD development method, determine the input parameters of the proxy research model;

[0008] A three-dimensional numerical simulation model was established by combining reservoir numerical simulation technology with the Latin hypercube sampling method, and the output parameters of the proxy study model were determined.

[0009] Based on the input and output parameters of the proxy research model, a machine learning algorithm is used to construct the proxy research model.

[0010] The hyperparameters of the surrogate research model are optimized to determine the optimal combination of hyperparameters, thus obtaining the final surrogate research model.

[0011] For the final agent research model, the combination of input parameters that maximizes the recovery rate or system efficiency is solved, and the corresponding combination of input parameters is the final optimization design result.

[0012] Furthermore, based on the variable well location deployment SAGD development method, the input parameters of the proxy research model are determined, including:

[0013] When the SAGD infill production horizontal well development mode is selected, the input parameters of the proxy research model are the well location deployment parameters and related operating parameters of the SAGD infill production horizontal well, specifically including the infill location of the SAGD infill production horizontal well, the operating pressure of the steam injection well, the maximum allowable steam volume of the production well, and the bottom hole flowing pressure of the infill production horizontal well.

[0014] Furthermore, based on the variable well location deployment SAGD development method, the input parameters of the proxy research model are determined, including:

[0015] When the staggered well pair SAGD development mode is selected, the input parameters of the proxy research model are the well location deployment parameters and related operation parameters of the steam injection well and production well in the staggered SAGD well pair. Specifically, these include the longitudinal and horizontal spacing between the steam injection well and the production well in the staggered SAGD well pair, the operating pressure of the steam injection well, and the bottom flow pressure of the production well.

[0016] Furthermore, a three-dimensional numerical simulation model was established using reservoir numerical simulation technology combined with the Latin hypercube sampling method to determine the output parameters of the surrogate study model, including:

[0017] Relevant geological static parameters and conventional SAGD well operation parameters collected from the reservoir production site were used as the basic parameters of the reservoir numerical simulation model. Multiple combinations of surrogate research model input parameters were designed using the Latin hypercube sampling method. Each combination was used to construct the reservoir numerical simulation model under different conditions with the basic parameter combination of the reservoir numerical simulation model, thus obtaining the reservoir numerical simulation model.

[0018] Among them, the relevant geological static parameters collected from the reservoir production site include reservoir effective thickness, reservoir horizontal permeability, the ratio of reservoir vertical permeability to horizontal permeability, reservoir porosity, initial reservoir pressure, initial oil saturation, initial reservoir temperature, and thermal conductivity of the reservoir and fluid; engineering parameters include steam injection temperature and bottom hole steam dryness.

[0019] The reservoir numerical simulation model was used to perform simulation calculations and obtain multiple sets of output parameters, which were used as output parameters for the proxy research model. The output parameters include the recovery rate value and the system efficiency value R. The recovery rate value is the ratio of the reservoir's cumulative oil production to its geological reserves, and the system efficiency value R is the ratio of the cumulative oil production value to the cumulative gasoline ratio value.

[0020] Furthermore, based on the input and output parameters of the proxy research model, a machine learning algorithm is used to construct the proxy research model, including:

[0021] A proxy research model was constructed using a support vector regression (SVR) machine.

[0022] Using the radial basis function (RBF) as the kernel function in the SVR regression function, the final SVR model is:

[0023]

[0024] In the formula: α i α i * To assist the non-negative Lagrange multipliers; N is the number of samples; γ is the width parameter of the radial basis kernel function, γ>0; b is the bias; x i Let be the value of the parameter of the i-th sample.

[0025] Furthermore, before constructing the proxy research model, the input parameters of the proxy research model are preprocessed. The preprocessing expression is as follows:

[0026]

[0027] Where x is the value of a parameter in the input dataset, x min x is the minimum value of this parameter. max This is the maximum value of this parameter. This is the normalized value, which is between [-1, 1].

[0028] Furthermore, hyperparameter optimization is performed on the surrogate research model to determine the optimal combination of hyperparameters, resulting in the final surrogate research model, including:

[0029] Genetic algorithms were used to optimize the hyperparameters of the support vector regression machine model to find the optimal model architecture. The three hyperparameters to be optimized were the penalty factor C, the width parameter γ, and the loss factor ε in the SVR model. The model with the combination of hyperparameters was used as the final proxy model for research.

[0030] Secondly, a variable well location deployment SAGD optimization device includes: an input parameter determination unit, an output parameter determination unit, a model building unit, a model parameter optimization unit, and a solution unit;

[0031] The input parameter determination unit is used to determine the input parameters of the proxy research model based on the SAGD development method of variable well location deployment.

[0032] The output parameter determination unit is used to establish a three-dimensional numerical simulation model by combining reservoir numerical simulation technology with the Latin hypercube sampling method, and to determine the output parameters of the proxy study model.

[0033] The model building unit is used to build the proxy research model using machine learning algorithms based on the input and output parameters of the proxy research model.

[0034] The model parameter optimization unit is used to perform hyperparameter optimization on the surrogate research model, determine the optimal combination of model hyperparameters, and obtain the final surrogate research model.

[0035] The solution unit is used to solve the final agent research model for the combination of input parameters that maximizes the recovery rate or system efficiency. The corresponding combination of input parameters is the final optimization design result.

[0036] Thirdly, an electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0037] Memory, which stores computer programs;

[0038] When a processor executes a computer program stored in memory, it implements the aforementioned variable well location deployment SAGD optimization method.

[0039] Fourthly, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned variable well location deployment SAGD optimization method.

[0040] This disclosure has at least the following beneficial effects:

[0041] This disclosure considers a variable-well-location SAGD development model, distinct from conventional SAGD development methods, specifically the SAGD infill production horizontal well development model and the staggered well pair SAGD development model, which can further improve the SAGD development effect in oil sands and extra-heavy oil reservoirs. Utilizing a secondary coupling of the Support Vector Regression (SVR) method and a genetic algorithm from the field of machine learning, a machine learning-based workflow for the optimized design of variable-well-location SAGD is established. This workflow enables rapid, reliable, and intelligent optimization design of well location deployment parameters and operational parameters related to the variable-well-location SAGD development model. This method is applicable to the well location deployment and operational parameter design of variable-well-location SAGD development models in oil sands and extra-heavy oil reservoirs, providing technical support for efficient decision-making and profitable development at the production site.

[0042] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart of the optimization method according to an embodiment of the present disclosure;

[0045] Figure 2 This is a schematic diagram of the optimized device structure according to an embodiment of the present disclosure;

[0046] Figure 3 This is a schematic diagram of the electronic device structure;

[0047] Figure 4 This is a schematic diagram of a reservoir profile developed using conventional SAGD technology.

[0048] Figure 5 Diagram of SAGD infill production horizontal well development model;

[0049] Figure 6 This is a diagram illustrating the biological evolution model of a genetic algorithm.

[0050] Figure 7 The fitness change curve;

[0051] Figure 8 Demonstrating the prediction performance of the SVR model;

[0052] Figure 9A flowchart for the overall optimization process;

[0053] Figure 10 This is a diagram illustrating the SAGD development model for misaligned wells. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0055] To address the challenge of rapidly and reliably optimizing well location deployment and operational parameters in the variable-well-location SAGD development model, this technology can be applied to the development of oil sands and extra-heavy oil reservoirs with thin reservoirs, poor reservoir properties, high water-cut layers, and shale interlayers, providing a favorable theoretical basis for reservoir design of variable-well-location SAGD development technology schemes.

[0056] like Figure 1 As shown, a method for optimizing SAGD deployment in variable well locations includes:

[0057] S101, Based on the variable well location deployment SAGD development method, determine the input parameters of the proxy research model;

[0058] S102, a three-dimensional numerical simulation model was established by combining reservoir numerical simulation technology with the Latin hypercube sampling method, and the output parameters of the proxy study model were determined.

[0059] S103, Based on the input and output parameters of the proxy research model, a machine learning algorithm is used to construct the proxy research model;

[0060] S104, Perform hyperparameter optimization on the surrogate research model to determine the optimal combination of model hyperparameters and obtain the final surrogate research model;

[0061] S105, for the final agent research model, solve for the combination of input parameters that maximizes the recovery rate or system efficiency value. The corresponding combination of input parameters is the final optimization design result.

[0062] The specific implementation details are as follows:

[0063] The variable well location deployment SAGD development method is determined to determine the input parameters of the proxy research model. The variable well location deployment SAGD development method mainly includes the SAGD infill production horizontal well development mode and the staggered well SAGD development mode.

[0064] The SAGD infill production horizontal well development mode involves adding a production horizontal well between adjacent well pairs after the SAGD development platform has been in operation for a certain number of years and the steam chambers of each well pair have been fully formed. This utilizes the remaining oil area between adjacent well pairs. Because the target development area of ​​the infill production horizontal well has been heated, it can be directly or after preheating before entering the production stage. The staggered well pair SAGD development technology considers that the upper and lower injection / production horizontal wells in a conventional SAGD well pair are no longer directly aligned, meaning there is a certain horizontal displacement between them. In this special well deployment combination mode, high-temperature steam is injected into the reservoir through the injection well, and crude oil is produced through the production well for development. If the target reservoir has already undergone conventional SAGD development and the steam chambers of each well pair have been largely formed, the SAGD infill production horizontal well development mode can be used to improve the SAGD development effect. If the target reservoir has not undergone conventional SAGD development, the staggered well pair SAGD development mode can be used to enhance the development effect.

[0065] If the SAGD infill production horizontal well development mode is selected, the input parameters of the proxy research model are the well location deployment parameters and related operating parameters of the SAGD infill production horizontal well, specifically including the infill location of the SAGD infill production horizontal well, the operating pressure of the steam injection well, the maximum allowable steam volume of the production well, and the bottom hole flowing pressure of the infill production horizontal well. If the staggered well pair SAGD development mode is selected, the input parameters of the proxy research model are the well location deployment parameters and related operating parameters of the steam injection well and production well in the staggered SAGD well pair, specifically including the longitudinal and horizontal spacing between the steam injection well and the production well in the staggered SAGD well pair, the operating pressure of the steam injection well, and the bottom hole flowing pressure of the production well.

[0066] To obtain the necessary basic data for constructing a three-dimensional reservoir numerical simulation model, it is necessary to collect relevant geological static parameters and conventional SAGD well operation parameters from the reservoir production site. Then, using reservoir numerical simulation technology combined with the Latin hypercube sampling method, three-dimensional numerical simulation models under different conditions are established to determine the output parameters of the proxy research model.

[0067] In the specific implementation process, relevant geological static parameters collected from the reservoir production site include reservoir effective thickness, reservoir horizontal permeability, the ratio of reservoir vertical permeability to horizontal permeability, reservoir porosity, initial reservoir pressure, initial oil saturation, initial reservoir temperature, and the thermal conductivity of the reservoir and fluids; engineering parameters include steam injection temperature and bottom hole steam dryness. The above parameters are the basic parameters for constructing the reservoir numerical simulation model.

[0068] The Latin hypercube sampling method possesses the characteristic of uniform stratification and can obtain tail sample values ​​with a smaller sample size, making it more efficient than ordinary sampling methods. Multiple combinations of surrogate research model input parameters are designed using the Latin hypercube sampling method. Each combination, when combined with the basic parameters of the aforementioned reservoir numerical simulation model, completes the construction of numerical simulation models under different conditions.

[0069] Based on this, numerical simulation calculations of the reservoir were performed, yielding multiple sets of output parameters, which serve as the foundation for constructing the proxy research model. These output parameters include the recovery rate and the system efficiency (R). The recovery rate is the ratio of the reservoir's cumulative oil production to its geological reserves, while the system efficiency is defined as follows:

[0070] R = COP / CSOR / 1000 (1)

[0071] In the formula: COP is the cumulative oil production value; CSOR is the cumulative gasoline ratio value. Generally, the larger the R value, the higher the system efficiency of the entire scheme.

[0072] Based on the determined input and output parameters of the proxy research model, a suitable machine learning algorithm is selected to construct the proxy research model.

[0073] The input parameters for the proxy research model are the well location deployment parameters and related operational parameters of the SAGD infill production horizontal wells, or the well location deployment parameters and related operational parameters of the misaligned SAGD wells aligning the steam injection wells and production wells. The machine learning algorithm is a supervised machine learning algorithm, namely the support vector regression machine learning algorithm.

[0074] The structural risk function of a Support Vector Regression (SVR) machine can be expressed in the following form:

[0075]

[0076] In the formula: ω is the regression parameter vector; N is the number of samples; f(x) is the calculated value of the model; y is the true value. The loss function L can be expressed in the following form:

[0077]

[0078] Based on the principle of risk minimization, the objective function expression for the SVR problem is obtained as follows:

[0079]

[0080] In the formula: C is the penalty factor, which controls the risk of the SVR method; L is the loss function, which is 0 when the sample point falls within the interval band, otherwise it is included in the loss; ε is the loss factor, and b is the bias.

[0081] Introduce relaxation factors ξ and ξ on both sides of the spacer band respectively. * To reduce the error, the objective function is transformed into the following form:

[0082]

[0083] The constraints are as follows:

[0084]

[0085] In Equation (5), the Lagrange function is introduced to transform the optimization problem into an equivalent dual problem, that is, auxiliary non-negative Lagrange multipliers α and α' are introduced. * To ensure the equivalence of the primal and dual problems, the KKT conditions must be satisfied, resulting in the dual problem:

[0086]

[0087] The constraints are:

[0088]

[0089] In equation (8), the sample points corresponding to the Lagrange multipliers are the support vectors in the SVR method:

[0090]

[0091] Based on functional correlation theory, a kernel function satisfying the Mercer condition is introduced. This allows us to complete linear regression tasks in high-dimensional feature spaces, and the resulting SVR regression function is:

[0092]

[0093] Using the radial basis function (RBF) as the kernel function in the SVR regression function, the final SVR model is:

[0094]

[0095] In the formula: γ is the width parameter of the radial basis kernel function, γ>0.

[0096] Based on the above model, intelligent optimization algorithms from the field of machine learning are coupled to optimize the model hyperparameters, determine the optimal combination of model hyperparameters, and use the model under this combination as the final proxy research model.

[0097] Before constructing the proxy research model, the input parameters need to be preprocessed to improve the computational accuracy of the proxy research model. The preprocessing expression is as follows:

[0098]

[0099] Where x is the value of a parameter in the input dataset, x min x is the minimum value of this parameter. max This is the maximum value of this parameter. This is the normalized value, which is between [-1, 1].

[0100] The intelligent optimization algorithm used in the machine learning field for coupling is the genetic algorithm. The genetic algorithm is used to perform hyperparameter optimization on the support vector regression model to find the optimal architecture of the model. The three hyperparameters to be optimized are the penalty factor C, the width parameter γ, and the loss factor ε in the SVR model. The optimization is carried out until the termination criterion is met, and the optimization result is output. The final surrogate research model is constructed using the result.

[0101] The final proxy research model is coupled twice with intelligent optimization algorithms in the field of machine learning to find the proxy research model input parameters that maximize or minimize the objective function. The corresponding input parameters are the final optimization design results.

[0102] The intelligent optimization algorithm used in the machine learning field for the secondary coupling is the genetic algorithm. The genetic algorithm is coupled with the final surrogate research model in a secondary manner. The output parameters of the surrogate research model are used as the fitness values ​​in the genetic algorithm operation process. The operation stops when the termination criterion is met. Based on this, the combination of input parameters that maximizes the recovery rate or system efficiency value is solved. The corresponding input parameters are the desired results. Thus, the optimal well location deployment parameters and related operation parameters under the variable well location deployment SAGD development mode are determined, and the corresponding optimization design work is completed.

[0103] like Figure 2 As shown, a variable well location deployment SAGD optimization device includes: an input parameter determination unit 201, an output parameter determination unit 202, a model construction unit 203, a model parameter optimization unit 204, and a solution unit 205;

[0104] The input parameter determination unit 201 is used to determine the input parameters of the proxy research model according to the SAGD development method of variable well location deployment;

[0105] The output parameter determination unit 202 is used to establish a three-dimensional numerical simulation model by combining reservoir numerical simulation technology with the Latin hypercube sampling method, and to determine the output parameters of the proxy research model.

[0106] The model building unit 203 is used to build the proxy research model using machine learning algorithms based on the input and output parameters of the proxy research model.

[0107] The model parameter optimization unit 204 is used to perform hyperparameter optimization on the surrogate research model, determine the optimal combination of model hyperparameters, and obtain the final surrogate research model.

[0108] Solver 205 is used to solve the final agent research model for the combination of input parameters that maximizes the recovery rate or system efficiency. The corresponding combination of input parameters is the final optimization design result.

[0109] like Figure 3 As shown, this disclosure provides an electronic device, including a processor 301, a communication interface 302, a memory 303, and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304;

[0110] Memory 303 stores computer programs;

[0111] The processor 301 implements the above method when executing a computer program stored in the memory 303.

[0112] This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0113] The computer-readable storage medium may be included in the device / apparatus described in the above embodiments; or it may exist independently and not assembled into the device / apparatus. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0114] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0115] To enable those skilled in the art to better understand this disclosure, the principles of this disclosure are explained below in conjunction with the accompanying drawings:

[0116] First, determine the SAGD development method for variable well location deployment.

[0117] First, let's take an oil sands reservoir that has been developed using conventional SAGD technology as an example. Figure 4 The figure shows a schematic diagram of a reservoir profile developed using conventional SAGD technology. As the SAGD development platform has been in operation for a certain number of years, the steam chambers of adjacent well pairs have been fully formed. At this point, a horizontal production well can be drilled between adjacent well pairs to utilize the remaining oil area between them. Because the target development area of ​​the drilled horizontal production well has been heated, it can be directly or after preheating to enter the production stage. Compared to the conventional method of drilling SAGD well pairs, this single-well drilling development technology saves one steam injection well, significantly reducing drilling and steam facility investment. Simultaneously, the drilling scheme can improve recovery and accelerate development. Therefore, for this situation, the SAGD drilled horizontal production well development model is selected. Figure 5 At this point, the input parameters of the proxy research model are the well location deployment parameters and related operating parameters of the SAGD infill production horizontal well, specifically including the infill location of the SAGD infill production horizontal well, the operating pressure of the steam injection well, the maximum allowable steam volume of the production well, and the bottom hole flowing pressure of the infill production horizontal well.

[0118] To obtain the necessary basic data for constructing a three-dimensional reservoir numerical simulation model, it is necessary to collect relevant geological static parameters and conventional SAGD well operation parameters from the reservoir production site. Then, using reservoir numerical simulation technology combined with the Latin hypercube sampling method, three-dimensional numerical simulation models under different conditions are established to determine the output parameters of the proxy research model.

[0119] In the specific implementation process, relevant geological static parameters collected from the reservoir production site include reservoir effective thickness, reservoir horizontal permeability, the ratio of reservoir vertical permeability to horizontal permeability, reservoir porosity, initial reservoir pressure, initial oil saturation, initial reservoir temperature, and the thermal conductivity of the reservoir and fluids; engineering parameters include steam injection temperature and bottom hole steam dryness. These parameters form the basis for constructing the reservoir numerical simulation model, and the specific parameter values ​​for this reservoir are shown in Table 1. A three-dimensional reservoir numerical simulation baseline model is constructed based on the parameters in Table 1.

[0120] Table 1 Basic Parameters of Numerical Simulation Model

[0121]

[0122] In this case, the input parameters of the surrogate study model are the well location deployment parameters and related operational parameters of the SAGD infill production horizontal wells, specifically including the infill location of the SAGD infill production horizontal wells, the operating pressure of the steam injection wells, the maximum allowable steam output of the production wells, and the bottomhole flowing pressure of the infill production horizontal wells. The Latin hypercube sampling method has the characteristic of uniform stratification, and it can obtain tail sample values ​​with fewer samples; these two characteristics make it more efficient than ordinary sampling methods. Based on the value range of the surrogate study model (as shown in Table 2), 200 sets of surrogate study model input parameter combinations were designed using the Latin hypercube sampling method. Numerical simulation models under different conditions were constructed based on the three-dimensional reservoir numerical simulation baseline model and the input parameters of the surrogate study model.

[0123] Table 2. Range of Input Parameters for the Agent Research Model

[0124]

[0125] The reservoir numerical simulation calculations for the above 200 models were completed, and the numerical simulation output results were obtained. The recovery rate, cumulative oil production, and cumulative gasoline ratio for each scheme were extracted. Based on the cumulative oil production and cumulative gasoline ratio, the system efficiency value (R) for each scheme can be calculated. The system efficiency value is defined as follows:

[0126] R = COP / CSOR / 1000 (1)

[0127] In the formula: COP is the cumulative oil production value; CSOR is the cumulative gasoline ratio value. Generally, the larger the R value, the higher the system efficiency of the entire scheme.

[0128] The recovery rate and R value of each scheme are taken as the output parameters of the surrogate research model, and together with the input parameters of the surrogate research model, the overall dataset of the surrogate research model is constructed.

[0129] Based on the overall dataset, the Support Vector Regression (SVR) machine learning algorithm was selected to construct the proxy research model.

[0130] The structural risk function of a Support Vector Regression (SVR) machine can be expressed in the following form:

[0131]

[0132] In the formula: ω is the regression parameter vector; N is the number of samples; f(x) is the calculated value of the model; y is the true value. The loss function L can be expressed in the following form:

[0133]

[0134] Based on the principle of risk minimization, the objective function expression for the SVR problem is obtained as follows:

[0135]

[0136] In the formula: C is the penalty factor, which controls the risk of the SVR method; L is the loss function, which is 0 when the sample point falls within the interval band, otherwise it is included in the loss; is the loss factor, and b is the bias.

[0137] Introduce relaxation factors ξ and ξ on both sides of the spacer band respectively. * To reduce the error, the objective function is transformed into the following form:

[0138]

[0139] The constraints are as follows:

[0140]

[0141] In Equation (5), the Lagrange function is introduced to transform the optimization problem into an equivalent dual problem, that is, auxiliary non-negative Lagrange multipliers α and α' are introduced. * To ensure the equivalence of the primal and dual problems, the KKT conditions must be satisfied, resulting in the dual problem:

[0142]

[0143] The constraints are:

[0144]

[0145] In equation (8), the sample points corresponding to the Lagrange multipliers are the support vectors in the SVR method:

[0146]

[0147] Based on functional correlation theory, a kernel function satisfying the Mercer condition is introduced. This allows us to complete linear regression tasks in high-dimensional feature spaces, and the resulting SVR regression function is:

[0148]

[0149] In this disclosure, the radial basis function (RBF) is used as the kernel function in the SVR regression function, and the final SVR model is as follows:

[0150]

[0151] In the formula: γ is the width parameter of the radial basis kernel function, γ>0.

[0152] The entire dataset, consisting of 200 schemes, was randomly divided into training and test sets in an 80%:20% ratio. These sets were used for training the SVR model and evaluating its accuracy, respectively. Notably, before training the SVR model, the input parameters of the entire dataset were preprocessed to improve prediction accuracy. The preprocessing expression is as follows:

[0153]

[0154] Where x is the value of a parameter in the input dataset, x min x is the minimum value of this parameter. max This is the maximum value of this parameter. This is the normalized value, which is between [-1, 1].

[0155] After preprocessing, an SVR surrogate research model is constructed using all input and output samples from the training set. The SVR model is then trained using a genetic algorithm (e.g., ...). Figure 6 As shown, the hyperparameters of the SVR model are optimized to find the optimal architecture of the SVR model, namely the penalty factor C, the width parameter γ, and the loss factor ε, until the termination criterion is met, and the optimization result is output. At this time, the three optimal model hyperparameters are the penalty factor C of 486.2194, the width parameter γ of 0.1052, and the loss factor ε of 0.0312.

[0156] like Figure 7 As shown, after a certain number of generations of genetic evolution, the fitness value has decreased to an extremely low level. The optimal SVR model hyperparameters after the above evolutionary process were selected to construct an SVR surrogate research model, which was then tested using test set data. The results are shown in the figure below. Figure 8 As shown, the data points are evenly distributed near the 45° line. The surrogate research model has good calculation performance and high accuracy. The mean square error of the training set is 0.0011 and the R² value is 0.9997. The mean square error of the test set is 0.0067 and the R² value is 0.9985.

[0157] Based on the SVR agent research model, a genetic algorithm is coupled again to optimize the operational parameters. The overall method is as follows: Figure 9 As shown, the output parameters of the surrogate research model are used as the fitness values ​​in the genetic algorithm operation process. The operation stops when the decision termination criterion is met. Based on this, the combination of input parameters that maximizes the recovery rate or system efficiency value is found. The corresponding combination of input parameters is the desired result. Thus, the optimal well location deployment parameters and related operating parameters under the variable well location deployment SAGD development mode are determined, and the corresponding optimization design work is completed.

[0158] The optimization results are shown in Tables 3 and 4. Based on these results, the SVR surrogate research model and numerical simulation model were compared and verified. The obtained recovery rate values ​​or R values ​​were very close, with very small errors, further confirming the reliability of the SVR surrogate research model. If the optimization design aims to maximize the recovery rate, the optimized infill production horizontal well location is 2m from the bottom of the reservoir, with higher operating pressure in the steam injection well and relatively lower bottomhole flowing pressure in the infill well. If the optimization design aims to maximize system efficiency, the optimized infill production horizontal well location is 2m from the bottom of the reservoir, with a lower maximum allowable steam volume in the optimized production well and a bottomhole flowing pressure in the infill well within a moderate range. Overall, to achieve a higher recovery rate, it is necessary to increase the operating pressure of the SAGD well pair as much as possible within the allowable range, while maintaining a low bottomhole flowing pressure in the infill production horizontal well to maximize the reservoir's oil production capacity. However, this approach does not result in high system efficiency. To achieve higher system efficiency, it is necessary to comprehensively consider the influence of various design parameters. The results of the optimized design show that the recovery rate or system efficiency of the variable well location SAGD development scheme is significantly improved compared with the basic scheme.

[0159] Table 3. Optimization Design Results with Maximum Recovery as the Objective

[0160]

[0161] Table 4 shows the optimization design results with the goal of maximizing system efficiency.

[0162]

[0163] Similarly, if conventional SAGD well pairs are not deployed in the reservoir, a staggered well pair SAGD deployment mode can be used. Figure 10 The diagram illustrates a staggered well pair SAGD deployment mode. As shown, the concept of staggered well pair SAGD development technology considers that in conventional SAGD well pairs, the upper and lower injection / production horizontal wells are no longer directly aligned; that is, the injection / production horizontal wells are displaced in the horizontal direction. In this special well placement combination mode, high-temperature steam is injected into the oil sands reservoir through the steam injection well, and crude oil is produced through the production well for development. Compared with conventional SAGD well pair practices, this deployment mode results in faster lateral development of the steam cavity and a larger lateral sweep area, which to some extent enhances the oil drainage effect, achieving the development effects of increasing recovery rate and reducing gasoline ratio. For this scenario, the input parameters of the proxy research model are the well placement parameters and related operational parameters of the steam injection well and production well in the staggered SAGD well pair, specifically including the longitudinal and horizontal spacing between the steam injection well and the production well, the operating pressure of the steam injection well, and the bottomhole flowing pressure of the production well.

[0164] For this scenario, a staggered well pair deployment scheme is designed, with a longitudinal spacing of 2-6m and a horizontal spacing of 5-20m between the steam injection well and the production well. The design assumes the operating pressure of the steam injection well varies between 1500kPa and 2500kPa, and the bottomhole flowing pressure of the production well varies between 500kPa and 1500kPa. For this scenario, 200 schemes are designed using the Latin hypercube sampling method for numerical simulation studies. The recovery rate, cumulative oil production, and cumulative gasoline ratio are extracted from the numerical simulation outputs for each scheme. Finally, the recovery rate and system efficiency values ​​under each combination of input parameters are used as the output parameters of the surrogate research model.

[0165] The entire dataset, consisting of 200 schemes, was randomly divided into training and test sets at an 80%:20% ratio. These sets were used for training the SVR model and evaluating its accuracy, respectively. Notably, before training the SVR model, the input parameters of the entire dataset were preprocessed to improve prediction accuracy. After preprocessing, the SVR surrogate research model was constructed using all input and output samples from the training set. The SVR model training process then commenced, utilizing genetic algorithms (such as...) Figure 7 As shown, hyperparameter optimization of the SVR model is performed to find the optimal architecture of the SVR model, namely the penalty factor C, the width parameter γ, and the loss factor ε, until the termination criterion is met, and the optimization results are output. At this time, the three optimal model hyperparameters are, respectively, the penalty factor C is 280.3567, the width parameter γ is 0.1675, and the loss factor ε is 0.0248. After a certain number of generations of genetic evolution, the fitness value has decreased to an extremely low level. The optimal SVR model hyperparameters after the above evolution are selected to construct an SVR surrogate research model, and the model is tested with test set data. The surrogate research model has good computational performance and high accuracy. The mean squared error of the training set is 0.0004, and the R2 value is 0.9999. The mean squared error of the test set is 0.0112, and the R2 value is 0.9982.

[0166] Based on the SVR agent research model, a genetic algorithm is coupled again to optimize the operational parameters. The overall method is as follows: Figure 10 As shown, the output parameters of the surrogate research model are used as the fitness values ​​in the genetic algorithm operation process. The operation stops when the decision termination criterion is met. Based on this, the combination of input parameters that maximizes the recovery rate or system efficiency value is found. The corresponding combination of input parameters is the desired result. Thus, the optimal well location deployment parameters and related operating parameters under the variable well location deployment SAGD development mode are determined, and the corresponding optimization design work is completed.

[0167] Based on the optimized design results, the SVR proxy research model and numerical simulation model were substituted into the calculations, and the obtained recovery rate or R value were very close with small errors, further confirming the reliability of the SVR proxy research model. If the optimization aims to maximize the recovery rate, the optimized longitudinal spacing between the injection well and the production well is 5m, and the horizontal spacing is 10m. The optimized injection well has a higher operating pressure of 2500kPa, while the bottom-hole flowing pressure of the production well is relatively lower at 500kPa. If the optimization aims to maximize the R value, the optimized longitudinal spacing between the injection well and the production well is 5m, and the horizontal spacing is 10m. The operating pressure of the injection well is 1850kPa, and the bottom-hole flowing pressure of the production well is 1250kPa. For SAGD development technology with misaligned wells, to achieve a high system efficiency, the pressure difference between the injection well and the production well should not be too small or too large. The optimized design results show that the recovery rate or system efficiency of the variable well location SAGD development scheme is significantly improved compared to the basic scheme.

[0168] This disclosed method considers the SAGD development mode with variable well location deployment. Simultaneously, it couples machine learning methods to achieve efficient design of development parameters for this mode. This provides rapid verification and technical support for improving oil recovery and reducing the gasoline-to-oil ratio in oil sands and extra-heavy oil reservoirs, ultimately enhancing the economic benefits of oilfield development. Specifically, it can be applied to PetroChina's Mackay River oil sands block in Canada and other blocks developed using SAGD technology, showing broad application prospects.

[0169] Although the present disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A method for optimizing SAGD deployment with variable well location, characterized in that, The method includes: Based on the variable well location deployment SAGD development method, determine the input parameters of the proxy research model; A three-dimensional numerical simulation model was established by combining reservoir numerical simulation technology with the Latin hypercube sampling method, and the output parameters of the proxy study model were determined. Based on the input and output parameters of the proxy research model, a machine learning algorithm is used to construct the proxy research model. The hyperparameters of the proxy research model are optimized to determine the optimal combination of hyperparameters, thus obtaining the final proxy research model. For the final agent research model, the combination of input parameters that maximizes the recovery rate or system efficiency is solved, and the corresponding combination of input parameters is the final optimization design result.

2. The SAGD optimization method for variable well location deployment according to claim 1, characterized in that, Based on the variable well location deployment SAGD development method, the input parameters of the proxy research model are determined, including: When the SAGD infill production horizontal well development mode is selected, the input parameters of the proxy research model are the well location deployment parameters and related operating parameters of the SAGD infill production horizontal well, specifically including the infill location of the SAGD infill production horizontal well, the operating pressure of the steam injection well, the maximum allowable steam volume of the production well, and the bottom hole flowing pressure of the infill production horizontal well.

3. The SAGD optimization method for variable well location deployment according to claim 1, characterized in that, Based on the variable well location deployment SAGD development method, the input parameters of the proxy research model are determined, including: When the staggered well pair SAGD development mode is selected, the input parameters of the proxy research model are the well location deployment parameters and related operation parameters of the steam injection well and production well in the staggered SAGD well pair. Specifically, these include the longitudinal and horizontal spacing between the steam injection well and the production well in the staggered SAGD well pair, the operating pressure of the steam injection well, and the bottom flow pressure of the production well.

4. The SAGD optimization method for variable well location deployment according to claim 1, characterized in that, A three-dimensional numerical simulation model was established using reservoir numerical simulation technology combined with the Latin hypercube sampling method. The output parameters of the surrogate study model were determined, including: Relevant geological static parameters and conventional SAGD well operation parameters collected from the reservoir production site were used as the basic parameters of the reservoir numerical simulation model. Multiple combinations of surrogate research model input parameters were designed using the Latin hypercube sampling method. Each combination was used to construct the reservoir numerical simulation model under different conditions with the basic parameter combination of the reservoir numerical simulation model, thus obtaining the reservoir numerical simulation model. Among them, the relevant geological static parameters collected from the reservoir production site include reservoir effective thickness, reservoir horizontal permeability, the ratio of reservoir vertical permeability to horizontal permeability, reservoir porosity, initial reservoir pressure, initial oil saturation, initial reservoir temperature, and thermal conductivity of the reservoir and fluid; engineering parameters include steam injection temperature and bottom hole steam dryness. The reservoir numerical simulation model was used to perform simulation calculations and obtain multiple sets of output parameters, which were used as output parameters for the proxy research model. The output parameters include the recovery rate value and the system efficiency value R. The recovery rate value is the ratio of the reservoir's cumulative oil production to its geological reserves, and the system efficiency value R is the ratio of the cumulative oil production value to the cumulative gasoline ratio value.

5. The SAGD optimization method for variable well location deployment according to claim 1, characterized in that, Based on the input and output parameters of the proxy research model, a machine learning algorithm is used to construct the proxy research model, including: A proxy research model was constructed using a support vector regression (SVR) machine. The radial basis function (RBF) is used as the kernel function in the SVR regression function.

6. The SAGD optimization method for variable well location deployment according to claim 1, characterized in that, Before constructing the proxy research model, the input parameters of the proxy research model are normalized, and the normalized values ​​are in the range of [-1, 1].

7. The SAGD optimization method for variable well location deployment according to claim 1, characterized in that, The hyperparameters of the surrogate research model are optimized to determine the optimal combination of hyperparameters, resulting in the final surrogate research model, including: A genetic algorithm was used to optimize the hyperparameters of the support vector regression machine model to find the optimal model architecture. The hyperparameters to be optimized were the penalty factor C, the width parameter γ, and the loss factor ε in the SVR model. The model with the combination of hyperparameters was used as the final proxy model for research.

8. A variable well location deployment SAGD optimization device, characterized in that, include: The unit consists of an input parameter determination unit, an output parameter determination unit, a model building unit, a model parameter optimization unit, and a solution unit. The input parameter determination unit is used to determine the input parameters of the proxy research model based on the SAGD development method of variable well location deployment. The output parameter determination unit is used to establish a three-dimensional numerical simulation model by combining reservoir numerical simulation technology with the Latin hypercube sampling method, and to determine the output parameters of the proxy study model. The model building unit is used to build the proxy research model using machine learning algorithms based on the input and output parameters of the proxy research model. The model parameter optimization unit is used to perform hyperparameter optimization on the surrogate research model, determine the optimal combination of model hyperparameters, and obtain the final surrogate research model. The solution unit is used to solve the final agent research model for the combination of input parameters that maximizes the recovery rate or system efficiency. The corresponding combination of input parameters is the final optimization design result.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, which stores computer programs; A processor, when executing a computer program stored in memory, implements the variable well location deployment SAGD optimization method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements a variable well location deployment SAGD optimization method according to any one of claims 1-7.

Images