Oil reservoir injection-production optimization method, system and equipment integrating multiple agent models

Abstract

The invention relates to an oil reservoir injection-production optimization method, system and equipment fusing multiple proxy models. The method comprises the steps that a geologic model and a carbon dioxide flooding three-phase component model are established; determining constraint conditions, and obtaining an initial population; determining a target function; compiling a reference vector guide optimization algorithm code fusing multiple proxy models, and establishing a target database based on the carbon dioxide flooding three-phase component model, the initial population and the target function; adjusting parameters of the plurality of proxy models based on the plurality of established target databases; generating a filial generation population based on a genetic algorithm, and executing a reference vector guided optimization algorithm to update the target database; based on a given agent updating condition and the updated target database, preferably selecting an agent model of each stage; and obtaining a non-dominated scheme set of carbon dioxide flooding and a corresponding digital-analog result based on the optimized agent model of each stage. The method can be widely applied to the field of oil reservoir production optimization.

Description

Technical Field

[0001] This invention belongs to the field of reservoir production optimization and CCUS research, specifically involving a reservoir injection and production optimization method, system and equipment that integrates multiple proxy models. Background Technology

[0002] To address the dual challenges of global climate change and energy development efficiency, there is an urgent need to research carbon dioxide enhanced oil recovery (CO2-EOR) technology. The CO2 injection-production scheme is a key factor affecting reservoir oil displacement efficiency and carbon storage effectiveness. Traditional CO2 injection-production scheme design often relies on the experience of reservoir engineers to manually design multiple schemes for comparison and selection. This method depends on experience, has high computational costs, and makes it difficult to obtain a truly optimal scheme.

[0003] With the continuous advancement of machine learning technology, data-driven surrogate-assisted optimization methods have become a research hotspot in reservoir development. These methods rely on historical reservoir data and numerical simulation results to construct surrogate models that efficiently capture the relationship between injection and production parameters and the dynamic response of the reservoir. Combined with intelligent optimization algorithms, they achieve accurate prediction and optimization of injection and production parameters. This approach is particularly suitable for complex reservoir environments, enabling the rapid formulation of optimal injection and production strategies, thereby improving development efficiency and optimizing economic benefits.

[0004] However, traditional injection-production optimization methods usually rely on a single surrogate model. When faced with complex and variable reservoir conditions, it may be difficult to take into account global adaptability, resulting in large prediction errors in certain parameter spaces, affecting the robustness of the model, and reducing computational efficiency. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a reservoir injection-production optimization method, system, and equipment that integrates multiple surrogate models. This method can adaptively select the optimal surrogate model, improve optimization accuracy and computational stability, and has significant engineering application value.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a reservoir injection-production optimization method that integrates multiple surrogate models, including:

[0008] Obtain oilfield geological information, establish a geological model of the oilfield using reservoir geological modeling software, and construct a three-phase component model for carbon dioxide flooding based on a reservoir numerical simulator;

[0009] Based on the actual production conditions of the oilfield, the constraints that the injection-production system and optimization cycle must meet are determined, and the initial population under the constraints is obtained using Latin hypercube sampling.

[0010] Using oilfield revenue and carbon dioxide storage as objective functions, the objectives that need to be proxied and optimized are determined.

[0011] The reference vector-guided optimization algorithm code, which integrates multiple proxy models, was written using MATLAB. It is used to establish a target database based on the three-phase component model of carbon dioxide flooding, the initial population, and the objective function.

[0012] Based on the established multiple target databases, the parameters of various proxy models are adjusted;

[0013] The offspring population is generated based on a genetic algorithm, and a reference vector-guided optimization algorithm is executed to update the target database.

[0014] Based on the given agent update conditions and the updated target database, the agent model for each stage is optimized.

[0015] Based on the offspring population generated by the surrogate model selected at each stage, the optimal potential solutions are numerically simulated, recorded, and added to the target database until the optimization ends, resulting in a set of non-dominated schemes for carbon dioxide flooding and the corresponding numerical simulation results.

[0016] Furthermore, the constraints that the injection and extraction system and optimization cycle must meet are as follows:

[0017] st Xg low ≤Xg≤Xg up

[0018] P low ≤P≤P up or Pw low ≤Pw≤Pw up

[0019] In the formula, Xg is the gas injection rate of the injection well, Xg low Xg is the lower bound of the injection rate. up P is the upper bound of the gas injection rate; P is the bottom hole pressure of the production well. low P is the lower bound of the bottom hole pressure of the production well. up Pw represents the upper limit of the bottom hole pressure of the production well; Pw represents the fluid production rate of the production well. low Pw is the lower bound of the bottom hole pressure of the production well. up This is the upper limit of the bottom pressure of the production well.

[0020] Furthermore, the oilfield revenue is expressed as follows:

[0021]

[0022] Where J represents oilfield revenue (USD); T represents the total development cycle of the oilfield; Δt represents the time step (d); and Q represents the total oilfield revenue. o,l For the crude oil production at time step l, m 3 / d;r o For the profit per barrel of crude oil, USD / m 3 Q gi,l m is the amount of carbon dioxide injected at time step l. 3 / d;r gi Cost per unit volume of carbon dioxide injected, USD / m 3 Q gp,l m represents the carbon dioxide production at time step l. 3 / d;r gp Cost per unit volume of carbon dioxide treated, USD / m 3 Q wi,l For the water injection volume of the oilfield at time step l, m 3 / d;r wi The cost per cubic meter of water injection is USD / m 3 Q wp,l For the water production of the oilfield at time step l, m 3 / d;r wp The cost of treating each barrel of produced water, USD / m 3 ;t l b represents the cumulative production time in step l, in years; b is the asset discount rate; N o N represents the number of producing wells. i For the number of injection wells, C; o Cost of drilling production wells, USD / well; C w Drilling injection cost, USD / wellbore.

[0023] Furthermore, the reference vector-guided optimization algorithm code, which integrates multiple surrogate models and is written in MATLAB, is used to establish a target database based on the three-phase component model of carbon dioxide flooding, the initial population, and the objective function. This database includes:

[0024] The initial population POP obtained using Latin hypercube sampling will be used as the data sample.

[0025] Clean the data samples;

[0026] Based on the cleaned data samples, numerical simulation software is called to perform numerical simulations, and the input and output data corresponding to each objective are added to the target database of the task.

[0027] Furthermore, based on the cleaned data samples, numerical simulation software is invoked to perform numerical simulations, and the input and output data corresponding to each objective are added to the target database of the task, including:

[0028] ① Write the data collection and processing system corresponding to a certain data sample into the system document according to the format requirements;

[0029] ② Call the reservoir numerical simulator to load the three-phase component model of carbon dioxide flooding, complete the numerical simulation of the current data sample, and calculate the oilfield revenue and carbon dioxide storage of the current data sample.

[0030] ③ Repeat steps ① to ② to read the input and output data corresponding to each data sample and add them to the target database of the task.

[0031] Furthermore, the adjustment of parameters for the multi-agent model based on the established multiple target databases includes:

[0032] The surrogate models were selected, including seven types of surrogate models: radial basis functions based on Gaussian kernel, multi-binomial kernel, inverse multi-binomial kernel, and thin-plate spline kernel, and standard Gaussian processes based on exponential kernel, square exponential kernel, and Matern kernel.

[0033] The multi-agent model is trained based on the established target databases, and the prediction accuracy of the multi-agent model is evaluated using preset indicators. Based on the evaluation results, the agent model for the next target is determined.

[0034] Furthermore, the step of generating a progeny population based on a genetic algorithm and executing a reference vector-guided optimization algorithm to update the target database includes:

[0035] Based on the initial population, offspring populations are generated using gene exchange and gene mutation operations.

[0036] Mutation operations are performed on individuals in a population generated by gene exchange to obtain the mutated individuals;

[0037] Based on the objective function value y of the individual population i From the mutated individuals, select u potential solutions and add them to the target database. Repeat the evolution process until the number of iterations reaches a predetermined value.

[0038] Furthermore, the process of selecting the optimal proxy model for each stage based on given proxy update conditions and the updated target database includes:

[0039] By integrating various target databases and setting the agent update to meet the condition Tp, each agent update will increase the number of data samples by Tp compared to the previous update. Based on the given maximum number of iterations FEs, the multi-agent model is updated.

[0040] Based on preset criteria, the updated multi-agent model is optimized to determine the agent model for the next stage of each objective.

[0041] Secondly, the present invention provides a reservoir injection-production optimization system that integrates multiple proxy models, including:

[0042] The component model construction module is used to acquire oilfield geological information, establish a geological model of the oilfield using reservoir geological modeling software, and construct a three-phase component model for carbon dioxide flooding based on the reservoir numerical simulator.

[0043] The constraint determination module is used to determine the constraints that the injection-production system and optimization cycle must meet based on the actual production conditions of the oilfield, and to obtain the initial population under the constraints using Latin hypercube sampling.

[0044] The objective function is used to determine the objectives that need to be proxied and optimized, with oilfield revenue and carbon dioxide storage as the objective functions.

[0045] The algorithm writing module is used to write reference vector-guided optimization algorithm code that integrates multiple proxy models, and to establish a target database based on the three-phase component model of carbon dioxide flooding, the initial population and the objective function;

[0046] The proxy model training module is used to adjust the parameters of various proxy models based on multiple established target databases.

[0047] The optimization and update module is used to generate offspring populations based on genetic algorithms and execute reference vector-guided optimization algorithms to update the target database.

[0048] The proxy model selection module is used to select the best proxy model for each stage based on the given proxy update conditions and the updated target database.

[0049] The results output module is used to perform numerical simulations on the offspring population generated by the surrogate model selected at each stage, select the potential solutions, record and add them to the target database until the optimization ends, and obtain the set of non-dominated schemes for carbon dioxide flooding and the corresponding numerical simulation results.

[0050] Thirdly, the present invention provides a computing device comprising: one or more processors and a memory, wherein the memory stores one or more programs and is configured to be executed by the one or more processors, the one or more programs including instructions for performing any method.

[0051] The present invention has the following advantages due to the adoption of the above technical solutions:

[0052] 1. The injection and sampling optimization method of the present invention, which integrates multiple surrogate models, can select the most suitable surrogate model according to the characteristics of the current target database, avoid overfitting due to excessive model complexity at a certain stage, and improve the accuracy and stability of the optimization solution while ensuring computational efficiency. It is particularly suitable for high-dimensional complex engineering optimization problems.

[0053] 2. This invention fully combines the advantages of standard Gaussian processes and radial basis functions. Standard Gaussian processes possess uncertainty quantification capabilities, providing reliable point-addition strategies for the optimization process, and are particularly suitable for accurate modeling of complex nonlinear regions. In contrast, radial basis functions, due to their strong high-dimensional adaptability and lower computational complexity, can provide more stable prediction results. Compared to traditional methods, this invention integrates the complementary characteristics of Gaussian processes and radial basis functions, significantly improving the optimization efficiency of the algorithm while enhancing the flexibility and robustness of the surrogate model, ultimately achieving efficient optimization of reservoir injection and production schemes.

[0054] Therefore, this invention can be widely applied in the fields of reservoir production optimization and CCUS research. Attached Figure Description

[0055] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings:

[0056] Figure 1 This is a flowchart of a reservoir injection and production optimization method that integrates multiple proxy models in an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of a preferred proxy model in an embodiment of the present invention;

[0058] Figure 3 To maximize the reservoir net present value and carbon dioxide burial volume 1000 times in this embodiment of the invention, a Pareto front is generated.

[0059] Figure 4 This is a curve showing the cumulative oil production over production time at the maximum reservoir net present value in this embodiment of the invention.

[0060] Figure 5 This is a curve showing the cumulative burial volume of the maximum carbon dioxide burial volume in this embodiment of the invention as a function of production time. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0062] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0063] To address the problem of poor optimization performance caused by insufficient robustness of traditional surrogate-assisted optimization models, some embodiments of the present invention provide a reservoir injection-production optimization method that integrates multiple surrogate models. This method fully combines the advantages of standard Gaussian processes and radial basis functions, and dynamically adjusts the model structure and parameters based on the surrogate model evaluation results, thereby improving optimization performance.

[0064] Correspondingly, in other embodiments of the present invention, a reservoir injection and production optimization system and device that integrates multiple proxy models is provided.

[0065] Example 1

[0066] like Figure 1 As shown, this invention provides a reservoir injection-production optimization method that integrates multiple surrogate models, comprising the following steps:

[0067] (1) Obtain oilfield geological information, establish a geological model of the oilfield using reservoir geological modeling software, and construct a three-phase component model for carbon dioxide flooding based on the reservoir numerical simulator;

[0068] (2) Determine the constraints that the injection-production system and optimization cycle must meet based on the actual production conditions of the oilfield, and use Latin hypercube sampling to obtain the initial population under the constraints;

[0069] (3) Using oilfield revenue and carbon dioxide storage as objective functions, determine the objectives that need to be proxied and optimized;

[0070] (4) Use MATLAB to write reference vector guided optimization algorithm code that integrates multiple proxy models, and use it to establish a target database based on the carbon dioxide flooding three-phase component model, initial population and objective function;

[0071] (5) Adjust the parameters of the multi-agent model based on the established multiple target databases;

[0072] (6) Generate offspring population based on genetic algorithm and execute reference vector guided optimization algorithm to update target database;

[0073] (7) Based on the given agent update conditions and target database, select the best agent model for each stage;

[0074] (8) Based on the offspring population generated by the preferred proxy model at each stage, the preferred potential solutions are numerically simulated, recorded and added to the target database until the optimization ends, and the set of non-dominated schemes for carbon dioxide flooding and the corresponding numerical simulation results are saved.

[0075] Furthermore, step (2) above includes the following steps:

[0076] (2.1) Based on the actual production conditions of the oilfield, determine the constraints that the injection-production system and optimization cycle must meet, namely the upper and lower limits of gas injection in injection wells and the upper and lower limits of liquid production or bottom-hole flowing pressure in production wells as constraints.

[0077] In this embodiment, the decision variables for optimizing carbon dioxide injection and production include the time-series injection and production regime (i.e., the carbon dioxide flooding injection and production scheme); the optimization variable for each injection well is the gas injection rate, and the optimization variable for each production well is the production rate or bottom hole pressure; the dimensions of the decision variables... Therefore, the constraints on the optimal variables for carbon dioxide flooding and injection can be expressed as:

[0078] st Xg low ≤Xg≤Xg up

[0079] P low ≤P≤P up or Pw low ≤Pw≤Pw up

[0080] In the formula, Xg is the gas injection rate of the injection well, Xg low Xg is the lower bound of the injection rate. up P is the upper bound of the gas injection rate; P is the bottom hole pressure of the production well. low P is the lower bound of the bottom hole pressure of the production well. up Pw represents the upper limit of the bottom hole pressure of the production well; Pw represents the fluid production rate of the production well. low Pw is the lower bound of the bottom hole pressure of the production well. up This is the upper limit of the bottom pressure of the production well.

[0081] (2.2) Determine the initial population POP size m, and use Latin hypercube sampling to generate an injection and sampling regime that meets the constraints, as the initial population.

[0082] Furthermore, in step (3) above, this embodiment uses oilfield revenue and carbon dioxide storage as objective functions to determine the two objectives that need to be proxies and optimized. Oilfield revenue and carbon dioxide storage can be obtained through reservoir numerical simulation results. Oilfield revenue mainly includes the cumulative oil production, cumulative carbon dioxide injection, cumulative carbon dioxide output, and cumulative water production calculated through numerical simulation. Considering development costs, the expression for oilfield revenue J(x,t) is as follows:

[0083]

[0084] Where J represents oilfield revenue (USD); T represents the total development cycle of the oilfield; Δt represents the time step (d); and Q represents the total oilfield revenue. o,l For the crude oil production at time step l, m 3 / d;r o For the profit per barrel of crude oil, USD / m 3 Q gi,l m is the amount of carbon dioxide injected at time step l. 3 / d;r gi Cost per unit volume of carbon dioxide injected, USD / m 3 Q gp,l m represents the carbon dioxide production at time step l. 3 / d;r gp Cost per unit volume of carbon dioxide treated, USD / m 3 Q wi,l For the water injection volume of the oilfield at time step l, m 3 / d;r wi The cost per cubic meter of water injection is USD / m 3 Q wp,l For the water production of the oilfield at time step l, m 3 / d;r wp The cost of treating each barrel of produced water, USD / m 3 ;t l b represents the cumulative production time in step l, in years; b is the asset discount rate; N o N represents the number of producing wells. i For the number of injection wells, C; o Cost of drilling production wells, USD / well; C w Drilling injection cost, USD / wellbore.

[0085] Furthermore, in step (4) above, the reference vector-guided optimization algorithm code that integrates multiple surrogate models is written using MATLAB, which includes three parts: establishing databases of various optimization targets, cleaning data samples, and calling numerical simulation software.

[0086] Specifically, it includes the following steps:

[0087] (4.1) The initial population POP obtained using Latin hypercube sampling will be used as the data sample.

[0088] (4.2) Clean the data samples (i.e., the historical data of the injection and collection system), mainly including deleting duplicate, missing and abnormal data samples to improve data quality.

[0089] (4.3) Based on the cleaned data samples, numerical simulation software is called to perform numerical simulation, and the input and output data corresponding to each target are added to the target database of the task.

[0090] Furthermore, the specific implementation of step (4.3) above includes the following steps:

[0091] (4.3.1) Write the data collection and processing system corresponding to a certain data sample into the system document according to the format requirements;

[0092] (4.3.2) Call the reservoir numerical simulator to load the three-phase component model of carbon dioxide flooding, complete the numerical simulation of the current data sample, and calculate the oilfield revenue and carbon dioxide storage of the current data sample.

[0093] (4.3.4) Repeat steps (4.3.1) to (4.3.2) to read the input and output data corresponding to each data sample and add the input and output data to the target database of this task. Among them, the input data refers to the injection and production system, and the output data refers to the oilfield revenue and carbon dioxide storage.

[0094] Furthermore, in step (5) above, based on the established multiple target databases, the parameters of the multi-agent model are adjusted, including:

[0095] (5.1) Screening multiple surrogate models. In this embodiment, seven surrogate models are selected: radial basis functions based on Gaussian kernel, multi-binomial kernel, inverse multi-binomial kernel and thin plate spline kernel, and standard Gaussian process based on exponential kernel, square exponential kernel and Matern kernel.

[0096] Radial basis functions based on Gaussian kernels, multi-binomial kernels, inverse multi-binomial kernels, and thin-plate spline kernels are functions that radiate outwards from the distance to the center point and vary with distance. They are commonly used in interpolation and function approximation problems. This surrogate model includes an input layer, a hidden layer, and an output layer. The interpolation form is as follows:

[0097]

[0098] In the formula, is the Euclidean norm. This refers to the kernel function. Commonly used kernel functions include the Gaussian kernel, multi-binomial kernel, inverse multi-binomial kernel, and thin-plate spline kernel. The formula for calculating the Gaussian kernel function is:

[0099]

[0100] In the formula, r = ||xx i || represents the distance between data point x and center point x. i The Euclidean norm between them; σ is the scale parameter.

[0101] The formula for calculating the multi-binomial kernel function is:

[0102]

[0103] In the formula, c is the shape parameter.

[0104] The formula for calculating the inverse multinomial kernel function is:

[0105]

[0106] The formula for calculating the kernel function of a thin plate spline is:

[0107]

[0108] In standard Gaussian processes based on exponential kernels, squared exponential kernels, Matern kernels, and exponential kernels, the standard Gaussian process is a commonly used regression-type surrogate model, widely applied in machine learning and reservoir production optimization. Unlike parametric modeling, Gaussian processes model the objective function, generating a nonparametric model and extending multi-source Gaussian processes to infinite dimensions. Theoretically, Gaussian processes can approximate any continuous function and possess unique capabilities for calculating uncertainty. A Gaussian process is specified by mean and covariance kernel functions, expressed as follows:

[0109] f(x)~GP(m(x),k(x,x′))

[0110] In the formula, m(x) is the mean of the distribution, which is usually set to 0 in the absence of prior knowledge; k(x,x′) is the kernel function, used to connect two points x and x′ in space. Commonly used kernel functions include the exponential kernel, the squared exponential kernel, and the Matern kernel. The formula for calculating the exponential kernel is:

[0111]

[0112] In the formula, σ 2 is the variance of the kernel function; l is the length scale.

[0113] The formula for calculating the squared exponent kernel is:

[0114]

[0115] In the formula, l is a parameter reflecting the smoothness of the objective function; x i and xj Let be any two points in space.

[0116] The formula for calculating the Matern kernel is:

[0117]

[0118] In the formula, v is the smoothing parameter; K v (·) represents the v-th order modified Bessel function; Γ(·) represents the gamma function.

[0119] If we have N observed sample data points x and y, and introduce a noise term with the mean of the distribution set to 0, then the predicted distribution of GP can be given directly in analytical form. The predicted distribution at the new test point x* is:

[0120]

[0121] in

[0122]

[0123] In the formula, This is the model's predicted value at x*; Let be the prediction variance of the model at x*, reflecting the uncertainty of the model at x*; K(X,X) is an N×N Gram matrix, reflecting the relationship between known points; k * =K(x) * (X), reflecting the relationship between the point to be measured and the known points. Transpose it; k ** =k(x * ,x * This reflects the relationship between the measured point and itself; This represents the noise term. It's worth noting that the prediction formula above is given only for a single test point x*, but the Gaussian process supports simultaneous input of multiple test points and simultaneous prediction of them, while the basic principles and formulas remain unchanged.

[0124] (5.2) The multi-agent model is trained based on the established target databases, and the prediction accuracy of the multi-agent model is evaluated using preset indicators. Based on the evaluation results, the agent model for the target in the next stage is determined.

[0125] In this embodiment, the deterministic coefficient (R) is selected. 2 The coefficient of determination is used as an indicator to evaluate the prediction accuracy of the surrogate model. The value of the coefficient of determination ranges from 0 to 1; the closer the value is to 1, the higher the accuracy of the model. As the surrogate model for the next stage of the objective, the expression for the coefficient of determination for each objective is:

[0126]

[0127] In the formula, y(x k ) represents the test sample x k The true value; Representative test sample x k The average value; Representative test sample x k The predicted value.

[0128] Furthermore, in step (6) above, the genetic algorithm generates a population with its own characteristics, and reference vector-guided optimization is performed, including the following steps:

[0129] (6.1) Based on the initial population, offspring populations are generated using gene exchange and gene mutation operations;

[0130] In this embodiment, it is assumed that in the g-th iteration, two different individuals X are randomly selected from the offspring population. p1 (g), X p2 (g) Gene exchange individual c is generated. i :

[0131]

[0132] In the formula, Represents the j-th gene of parent individual 1. Represents the j-th gene of parent individual 1; c i (g) represents the j-th individual generated after crossover; β is used to control the degree of offset in the offspring.

[0133] (6.2) Perform mutation operations on the offspring population generated by gene exchange in step (6.1) to obtain the mutated individuals m. i,j (g):

[0134] m i,j (g)=c i (g)+δ(x max -x min )

[0135]

[0136] δ1=(c i (g)-x min ) / (x max -x min )

[0137] δ2=(x max -c i (g)) / (x max -x min )

[0138] In the formula, m iFor individuals with genetic variations; δ represents the perturbation factor, following a multinomial distribution; δ1 and δ2 represent the normalized distances of the current variable value relative to the upper and lower bounds, respectively; η m The distribution index represents the control for variation.

[0139] (6.3) Reference vector-guided selection of potential solutions: Based on the objective function value y of the individual population i From the mutated individuals, select u potential solutions and add them to the target database. Repeat the evolution process until the number of iterations reaches a predetermined value.

[0140] Specifically, it includes four parts: generating reference vectors, assigning individuals to reference vectors, selecting potential solutions based on angle-penalized distance, and adaptive adjustment of reference vectors. The formula for generating reference vectors is as follows:

[0141]

[0142] In the formula, i = 1, 2, ..., N, where N is the number of uniformly distributed points; m is the number of objective functions; and H is a positive integer representing the simplex lattice design. The formula for calculating the cosine value between the objective vector and the reference vector is:

[0143]

[0144] In the formula, f i,t Let i represent the target vector of individual i in generation t; θ represents the minimum objective vector of the current population P; i,j Represents the target vector With reference vector v j The angle between them. If and only if... and v j When the inter-cosine value is maximized, individual i is assigned to a subpopulation. The formula for calculating the angle penalty distance value is:

[0145]

[0146] In the formula, θ is the distance from the origin to the target vector corresponding to the i-th individual; i Let v be the angle between the i-th individual and its assigned reference vector; let v be the reference vector. i Its nearest neighbor reference vector v j The minimum angle between them. After calculating the angle penalty distance values ​​of all individuals in each subpopulation, the individual with the smallest angle penalty distance value is selected from each subpopulation as the next generation individual. The adaptive adjustment formula for the reference vector is:

[0147]

[0148] In the formula, v t+1,i v is the i-th adjustment reference vector applicable to the next generation t+1; 0,i The initial vector; ||°|| represents the Hadamard product of two matrices of the same size; and Let be the maximum and minimum values ​​of each objective function in the t-th generation, respectively.

[0149] Furthermore, in step (7) above, the step of selecting the optimal phased proxy model based on the proxy update condition is as follows:

[0150] (7.1) Integrate the target databases and set the agent update to satisfy condition Tp. Each agent update will increase the number of data samples by Tp compared to the previous update. Based on the given maximum number of iterations FEs, the total number of agent model updates G in the entire optimization process satisfies:

[0151] G = math.floor(FEs / Tp)

[0152] (7.2) The criteria for selecting the staged proxy model are consistent with those in step (5.2), based on the coefficient of determination R. 2 Optimize the proxy model for the next stage of each objective.

[0153] Example 2

[0154] This embodiment further describes the reservoir injection and production optimization method that integrates multiple proxy models proposed in this invention using a specific research block.

[0155] In this embodiment, the study block is composed of a 23*58*3 grid. An Eclipse 300 component model is used to simulate a three-phase oil-gas-water drive. This model includes 9 production wells and 4 injection wells, with a planned production period of 1800 days. The gas injection rate and the bottomhole pressure of the production wells are used as optimization variables for carbon dioxide flooding. The upper and lower bounds of the single-well gas injection rate are 5000 m³ / s. 3 / d and 15000m 3 / d; The upper and lower limits of the bottom pressure of the production well are 290 bar and 190 bar, respectively.

[0156] In this embodiment, the algorithm iteration count FEs = 1000, the number of potential solutions selected in each iteration μ = 5, and the condition Tp for phased selection of surrogates = 100 are set. The specific steps include:

[0157] (1) Obtain oilfield geological information, establish a geological model of the oilfield using the reservoir geological modeling software Petrel, and construct an oil-gas-water three-phase component model based on the reservoir numerical simulator Eclipse 300;

[0158] (2) Based on the actual production situation of the oilfield, determine the necessary conditions that the injection-production system and optimization cycle must meet, and use Latin hypercube sampling to obtain the initial population under constraints. The specific steps include:

[0159] (2.1) Based on the oilfield production status, the decision variable for optimizing carbon dioxide flooding injection and production is determined to be the time-series injection and production regime. The upper and lower bounds of the gas injection rate Xg for each injection well are Xg and Xg, respectively. up =15000m 3 and Xg low =5000m 3 The upper and lower bounds of the bottom hole pressure P of the production well are P0 and P1, respectively. up =290 bar and P low =190 bar. Dimensions of decision variables. The constraints on the optimization variables of the CO2-WAG strategy can be expressed as follows:

[0160] st Xg low ≤Xg≤Xg up

[0161] P low ≤P≤P up or Pw low ≤Pw≤Pw up

[0162] (2.2) Determine the initial population POP=100, and generate a carbon dioxide flooding and extraction scheme that meets the constraints based on Latin hypercube sampling.

[0163] (3) Calculate the target values ​​of each carbon dioxide flooding and production scheme with the reservoir economic benefits and carbon dioxide storage as the targets to be represented.

[0164] (3.1) Oilfield revenue includes the cumulative oil production, cumulative carbon dioxide injection, cumulative carbon dioxide output, and cumulative water production calculated through numerical simulation. After considering development costs, the oilfield revenue J(x,t) is expressed as follows:

[0165]

[0166] Where J represents oilfield revenue (USD); T represents the total development cycle of the oilfield; Δt represents the time step (d); and Q represents the total oilfield revenue. o,l For the crude oil production at time step l, m 3 / d;r o For the profit per barrel of crude oil, USD / m 3 Q gi,l m is the amount of carbon dioxide injected at time step l. 3 / d;r gi Cost per unit volume of carbon dioxide injected, USD / m 3 Qgp,l m represents the carbon dioxide production at time step l. 3 / d;r gp Cost per unit volume of carbon dioxide treated, USD / m 3 Q wi,l For the water injection volume of the oilfield at time step l, m 3 / d;r wi The cost per cubic meter of water injection is USD / m 3 Q wp,l For the water production of the oilfield at time step l, m 3 / d;r wp The cost of treating each barrel of produced water, USD / m 3 ;t l b represents the cumulative production time in step l, in years; b is the asset discount rate; N o N represents the number of producing wells. i For the number of injection wells, C; o Cost of drilling production wells, USD / well; C w Drilling injection cost, USD / wellbore.

[0167] In this embodiment, the number of injection wells N i The number of production wells is 9, N. o The value is 4, the time step T is set to 180, the time step size Δt = 30 days, the production cycle is 1800 days, and the revenue per cubic meter of crude oil is r. o Set to 503, cost per unit volume of carbon dioxide injection r gi The cost of treating carbon dioxide per unit volume is 0.083, r. gp The cost of treating each cubic meter of produced water is 0.027, r. wp It is 31.5.

[0168] (3.2) The carbon dioxide storage capacity can be obtained from the results of reservoir numerical simulation.

[0169] (4) Use MATLAB to write a reference vector-guided optimization algorithm that integrates multiple surrogate models. The specific MATLAB code includes establishing the target databases, cleaning data samples, and calling numerical simulation software:

[0170] (4.1) Establish the target databases and, after completing constrained Latin hypercube sampling, obtain the initial population POP, where POP = {x1, x2, ..., x...}. i ,...,x m}, where m is the initial population size, x i For the optimization strategy of sample i, numerical simulation is carried out based on the established oilfield component model to obtain the result matrix Y = {Y1, Y2, ..., Y} for each objective. j ,...,Yk}, where k = 2, Y j Dimension D j =100. Database DB = {DB1, DB2, ..., DB} is used to divide different tasks. j ,...,DB k}, DB j DB is the database for task j. j ={(x i ,Y ji )}.

[0171] (4.2) Clean the data samples. Write MATLAB code to clean the data samples, delete duplicate, missing and abnormal data samples, and improve the data quality.

[0172] (4.3) Calling numerical simulation software, the specific implementation steps include:

[0173] (4.3.1) Write the sampling and collection system for a certain sample into the system document according to the format requirements;

[0174] (4.3.2) Call the component simulator to load the model, complete the numerical simulation of the current sample, and calculate and obtain the reservoir economic benefits and carbon dioxide burial amount of the sample according to the formula.

[0175] (4.3.3) Read the input and output data corresponding to each optimization objective and add the data to the database of the task.

[0176] (5) Based on the established target databases, select the best proxy model for each target. The specific steps include:

[0177] (5.1) Screening multiple surrogate models: Seven surrogate models were selected, including radial basis functions based on Gaussian kernel, multi-binomial kernel, inverse multi-binomial kernel and thin plate spline kernel, and standard Gaussian process based on exponential kernel, square exponential kernel and Matern kernel.

[0178] (5.2) Evaluate the performance of the surrogate model by dividing the multi-objective database into training and test sets, and selecting the coefficient of determination (R²). 2 The closer the surrogate model is to 1, the better it will be used as the surrogate model for the next stage of the target.

[0179] (6) Based on the given initial population size, optimization population size, number of iterations, and objective function, perform reference vector-guided optimization. The specific steps include:

[0180] (6.1) Gene exchange: In the g-th iteration, two different individuals X are randomly selected from the population. p1 (g), X p2 (g) Gene exchange individual c is generated. i :

[0181]

[0182] (6.2) Gene mutation operation: Perform mutation operation on the population generated by gene exchange in step (6.1) to obtain the mutated m. i,j (g):

[0183] m i,j (g)=c i (g)+δ(x max -x min )

[0184]

[0185] δ1=(c i (g)-x min ) / (x max -x min )

[0186] δ2=(x max -C i (G)) / (X max -X min )

[0187] In the formula, m i For individuals with genetic variations; δ represents the perturbation factor, following a multinomial distribution; δ1 and δ2 represent the normalized distances of the current variable value relative to the upper and lower bounds, respectively; η m η represents the distribution index that controls for variation. In the example, η m =20.

[0188] (6.3) The reference vector-guided selection of potential solutions is based on the objective function value y of the individual population. i The process involves selecting μ potential solutions from offspring individuals and adding them to the database, then iterating repeatedly until a predetermined number of iterations is reached. Specifically, it includes four parts: generating a reference vector, assigning individuals to the reference vector, selecting potential solutions based on angle penalty distance, and adaptively adjusting the reference vector. The formula for generating the reference vector is as follows:

[0189]

[0190] In the formula, i = 1, 2, ..., N, where N is the number of uniformly distributed points; m is the number of objective functions; and H is a positive integer representing the simplex lattice design. In this example, N = 50, and the formula for calculating the cosine value between the objective vector and the reference vector is:

[0191]

[0192] In the formula, f i,t Let i represent the target vector of individual i in generation t; θ represents the minimum objective vector of the current population P; i,j Represents the target vector With reference vector v j The angle between them. If and only if... and v j When the inter-cosine value is maximized, individual i is assigned to a subpopulation. The formula for calculating the angle penalty distance value is:

[0193]

[0194] In the formula, θ is the distance from the origin to the target vector corresponding to the i-th individual; i Let v be the angle between the i-th individual and its assigned reference vector; let v be the reference vector. i Its nearest neighbor reference vector v j The minimum angle between them. After calculating the angle penalty distance values ​​of all individuals in each subpopulation, the individual with the smallest angle penalty distance value is selected from each subpopulation as the next generation individual. The adaptive adjustment formula for the reference vector is:

[0195]

[0196] In the formula, v t+1,i v is the i-th adjustment reference vector applicable to the next generation t+1; 0,i The initial vector; ||°|| represents the Hadamard product of two matrices of the same size; and Let be the maximum and minimum values ​​of each objective function in the t-th generation, respectively.

[0197] (7) The steps for selecting the optimal phased proxy based on the proxy update conditions are as follows:

[0198] (7.1) Integrate the target databases. In this example, the proxy update condition is Tp = 100, the maximum number of iterations is FEs = 1000, and the total number of proxy model updates during the entire optimization process is G = math.floor(FEs / Tp) = 10 times.

[0199] (7.2) The criteria for selecting the staged proxy model are consistent with those in step (5.2), based on the coefficient of determination R. 2 Optimize the proxy model for the next stage of each objective.

[0200] (8) Based on the offspring population generated by the current stage proxy evaluation algorithm, select μ=5 potential solutions for numerical simulation, record and add them to the database until the optimization is completed, and save the carbon dioxide non-dominated solution set scheme and the numerical simulation results of reservoir net present value and carbon dioxide burial amount.

[0201] To verify the effectiveness and advantages of this invention, the results of finding optimal carbon dioxide flooding, injection, and production schemes in the decision region were compared using a reference vector-guided evolutionary algorithm based on a Kriging model and the method of this invention. Both schemes used genetic mutation to generate offspring populations, iterated 1000 times, and used constrained Latin hypercube sampling. After optimization, the Pareto fronts of oilfield benefits and carbon dioxide reserves were as follows: Figure 3 As shown, from Figure 3 It can be seen that the reservoir injection-production optimization method that integrates multiple surrogate models is more efficient than the injection-production optimization method based on the Kriging model, and the Pareto front of the algorithm proposed in this invention has been significantly advanced.

[0202] Figure 4 The cumulative oil production as a function of production time for the maximum reservoir net present value is shown in the example curve. Figure 4 As can be seen from the data, the reservoir injection-production optimization method that integrates multiple agent models has a significantly higher cumulative oil production volume with the maximum reservoir net present value, resulting in higher actual economic benefits for the oilfield. Figure 5 The curve showing the cumulative carbon dioxide storage volume as a function of production time in the example with the maximum storage capacity indicates a significantly higher storage volume. (Summary) Figure 4 and Figure 5 It can be observed that the reservoir injection-production optimization method of the present invention, which integrates multiple proxy models, has high optimization efficiency and significantly improves the quality of the non-dominated solution sets of reservoir net present value and carbon dioxide burial quantity.

[0203] Example 3

[0204] The above-described embodiment 1 provides a reservoir injection-production optimization method integrating multiple surrogate models. Correspondingly, this embodiment provides a reservoir injection-production optimization system integrating multiple surrogate models. The system provided in this embodiment can implement the reservoir injection-production optimization method integrating multiple surrogate models of embodiment 1. This system can be implemented through software, hardware, or a combination of both. For example, the system may include integrated or separate functional modules or functional units to execute the corresponding steps in the methods of embodiment 1. Since the system in this embodiment is basically similar to the method embodiment, the description process in this embodiment is relatively simple. Relevant details can be found in the description of embodiment 1. The system embodiment provided in this embodiment is merely illustrative.

[0205] The reservoir injection and production optimization system that integrates multiple proxy models provided in this embodiment includes:

[0206] The component model construction module is used to acquire oilfield geological information, establish a geological model of the oilfield using reservoir geological modeling software, and construct a three-phase component model for carbon dioxide flooding based on the reservoir numerical simulator.

[0207] The constraint determination module is used to determine the constraints that the injection-production system and optimization cycle must meet based on the actual production conditions of the oilfield, and to obtain the initial population under the constraints using Latin hypercube sampling.

[0208] The objective function is used to determine the objectives that need to be proxied and optimized, with oilfield revenue and carbon dioxide storage as the objective functions.

[0209] The algorithm writing module is used to write reference vector-guided optimization algorithm code that integrates multiple proxy models, and to establish a target database based on the three-phase component model of carbon dioxide flooding, the initial population and the objective function;

[0210] The proxy model training module is used to adjust the parameters of various proxy models based on multiple established target databases.

[0211] The optimization and update module is used to generate offspring populations based on genetic algorithms and execute reference vector-guided optimization algorithms to update the target database.

[0212] The proxy model selection module is used to select the best proxy model for each stage based on the given proxy update conditions and the updated target database.

[0213] The results output module is used to perform numerical simulations on the offspring population generated by the surrogate model selected at each stage, select the potential solutions, record and add them to the target database until the optimization ends, and obtain the set of non-dominated schemes for carbon dioxide flooding and the corresponding numerical simulation results.

[0214] Example 4

[0215] This embodiment provides a processing device corresponding to the reservoir injection and production optimization method integrating multiple proxy models provided in Embodiment 1. The processing device can be a client-side processing device, such as a mobile phone, laptop, tablet computer, desktop computer, etc., to execute the method of Embodiment 1.

[0216] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it executes the reservoir injection and production optimization method integrating multiple proxy models provided in Embodiment 1.

[0217] Preferably, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0218] Preferably, the processor can be any type of general-purpose processor such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation herein.

[0219] Example 5

[0220] The reservoir injection-production optimization method integrating multiple proxy models described in Embodiment 1 can be specifically implemented as a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the reservoir injection-production optimization method integrating multiple proxy models described in Embodiment 1 are loaded.

[0221] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0222] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0223] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0224] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0225] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0226] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A reservoir injection-production optimization method integrating multiple surrogate models, characterized in that, include: Obtain oilfield geological information, establish a geological model of the oilfield using reservoir geological modeling software, and construct a three-phase component model for carbon dioxide flooding based on a reservoir numerical simulator; Based on the actual production conditions of the oilfield, the constraints that the injection-production system and optimization cycle must meet are determined, and the initial population under the constraints is obtained using Latin hypercube sampling. Using oilfield revenue and carbon dioxide storage as objective functions, the objectives that need to be proxied and optimized are determined. Write reference vector-guided optimization algorithm code that integrates multiple proxy models, and use it to establish a target database based on the three-phase component model of carbon dioxide flooding, the initial population and the objective function; Based on the established multiple target databases, the parameters of various proxy models are adjusted; The offspring population is generated based on a genetic algorithm, and a reference vector-guided optimization algorithm is executed to update the target database. Based on the given agent update conditions and the updated target database, the agent model for each stage is optimized. Based on the offspring population generated by the surrogate model selected at each stage, the optimal potential solutions are numerically simulated, recorded, and added to the target database until the optimization ends, resulting in a set of non-dominated schemes for carbon dioxide flooding and the corresponding numerical simulation results.

2. The reservoir injection-production optimization method integrating multiple surrogate models as described in claim 1, characterized in that, The constraints that the injection and extraction system and optimization cycle must meet are as follows: st Xg low ≤Xg≤Xg up P low ≤P≤P up or Pw low ≤Pw≤Pw up In the formula, Xg is the gas injection rate of the injection well, Xg low Xg is the lower bound of the injection rate. up P is the upper bound of the gas injection rate; P is the bottom hole pressure of the production well. low P is the lower bound of the bottom hole pressure of the production well. up Pw represents the upper limit of the bottom hole pressure of the production well; Pw represents the production fluid rate of the production well. low Pw is the lower bound of the bottom hole pressure of the production well. up This is the upper limit of the bottom pressure of the production well.

3. The reservoir injection-production optimization method integrating multiple surrogate models as described in claim 1, characterized in that, The oilfield revenue is expressed as follows: Where J represents oilfield revenue (USD); T represents the total development cycle of the oilfield; Δt represents the time step (d); and Q represents the total oilfield revenue. o,l For the crude oil production at time step l, m 3 / d;r o For the profit per barrel of crude oil, USD / m 3 Q gi,l m is the amount of carbon dioxide injected at time step l. 3 / d;r gi Cost per unit volume of carbon dioxide injected, USD / m 3 Q gp,l m represents the carbon dioxide production at time step l. 3 / d;r gp Cost per unit volume of carbon dioxide treated, USD / m 3 Q wi,l For the water injection volume of the oilfield at time step l, m 3 / d;r wi The cost per cubic meter of water injection, USD / m 3 Q wp,l For the water production of the oilfield at time step l, m 3 / d;r wp The cost of treating each barrel of produced water, USD / m 3 ;t l b represents the cumulative production time in step l, in years; b is the asset discount rate; N o N represents the number of producing wells. i For the number of injection wells, C; o Cost of drilling production wells, USD / well; C w Drilling injection cost, USD / wellbore.

4. The reservoir injection-production optimization method integrating multiple surrogate models as described in claim 1, characterized in that, The reference vector-guided optimization algorithm code, written in MATLAB and incorporating multiple surrogate models, is used to establish a target database based on a three-phase component model of carbon dioxide flooding, an initial population, and an objective function. This database includes: The initial population POP obtained using Latin hypercube sampling will be used as the data sample. Clean the data samples; Based on the cleaned data samples, numerical simulation software is called to perform numerical simulations, and the input and output data corresponding to each objective are added to the target database of the task.

5. The reservoir injection-production optimization method integrating multiple surrogate models as described in claim 4, characterized in that, The process involves using cleaned data samples to perform numerical simulations, invoking numerical simulation software, and adding the input and output data corresponding to each objective to the target database for that task. ① Write the data collection and processing system corresponding to a certain data sample into the system document according to the format requirements; ② Call the reservoir numerical simulator to load the three-phase component model of carbon dioxide flooding, complete the numerical simulation of the current data sample, and calculate the oilfield revenue and carbon dioxide storage of the current data sample. ③ Repeat steps ① to ② to read the input and output data corresponding to each data sample and add them to the target database of the task.

6. The reservoir injection-production optimization method integrating multiple surrogate models as described in claim 1, characterized in that, The adjustment of parameters for the multi-agent model based on the established multiple target databases includes: The surrogate models were selected, including seven types of surrogate models: radial basis functions based on Gaussian kernel, multi-binomial kernel, inverse multi-binomial kernel, and thin-plate spline kernel, and standard Gaussian processes based on exponential kernel, square exponential kernel, and Matern kernel. The multi-agent model is trained based on the established target databases, and the prediction accuracy of the multi-agent model is evaluated using preset indicators. Based on the evaluation results, the agent model for the next target is determined.

7. The reservoir injection-production optimization method integrating multiple surrogate models as described in claim 1, characterized in that, The process of generating a progeny population based on a genetic algorithm and executing a reference vector-guided optimization algorithm to update the target database includes: Based on the initial population, offspring populations are generated using gene exchange and gene mutation operations. Mutation operations are performed on individuals in a population generated by gene exchange to obtain the mutated individuals; Based on the objective function value y of the individual population i From the mutated individuals, select u potential solutions and add them to the target database. Repeat the evolution process until the number of iterations reaches a predetermined value.

8. The reservoir injection-production optimization method integrating multiple surrogate models as described in claim 1, characterized in that, The process of selecting the optimal proxy model for each stage based on given proxy update conditions and the updated target database includes: By integrating various target databases and setting the agent update to meet the condition Tp, each agent update will increase the number of data samples by Tp compared to the previous update. Based on the given maximum number of iterations FEs, the multi-agent model is updated. Based on preset criteria, the updated multi-agent model is optimized to determine the agent model for the next stage of each objective.

9. A reservoir injection-production optimization system integrating multiple surrogate models, characterized in that, include: The component model construction module is used to acquire oilfield geological information, establish a geological model of the oilfield using reservoir geological modeling software, and construct a three-phase component model for carbon dioxide flooding based on the reservoir numerical simulator. The constraint determination module is used to determine the constraints that the injection-production system and optimization cycle must meet based on the actual production conditions of the oilfield, and to obtain the initial population under the constraints using Latin hypercube sampling. The objective function is used to determine the objectives that need to be proxied and optimized, with oilfield revenue and carbon dioxide storage as the objective functions. The algorithm writing module is used to write reference vector-guided optimization algorithm code that integrates multiple proxy models, and to establish a target database based on the three-phase component model of carbon dioxide flooding, the initial population and the objective function; The proxy model training module is used to adjust the parameters of various proxy models based on multiple established target databases. The optimization and update module is used to generate offspring populations based on genetic algorithms and execute reference vector-guided optimization algorithms to update the target database. The proxy model selection module is used to select the best proxy model for each stage based on the given proxy update conditions and the updated target database. The results output module is used to perform numerical simulations on the offspring population generated by the surrogate model selected at each stage, select the potential solutions, record and add them to the target database until the optimization ends, and obtain the set of non-dominated schemes for carbon dioxide flooding and the corresponding numerical simulation results.

10. A computing device, characterized in that, include: One or more processors and a memory, wherein the memory stores one or more programs and is configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 8.

Images