A method and system for enhanced oil recovery based on neural network models

Abstract

The application relates to a method and system for predicting enhanced oil recovery based on a neural network model, and belongs to the technical field of deep learning and enhanced oil recovery, and comprises the following steps: constructing and training a neural network model; inputting actual reservoir parameters of an oilfield development unit into the trained neural network model, and obtaining an economic limit enhanced oil recovery value that can be reached by adopting heterogeneous complex flooding development of the oilfield development unit through the trained neural network model. The neural network model comprises an input layer, a hidden layer and an output layer; the construction of the neural network model comprises the following steps: screening reservoir parameters of the input layer of the neural network model; determining the recovery rate under the optimal injection-production condition of the heterogeneous complex flooding by adopting a particle swarm optimization algorithm, and constructing evaluation indexes of the output layer of the neural network model. The application does not need to artificially assume the correlation between the prediction indexes and the reservoir parameters in advance, and avoids the problem that the prediction model has errors due to differences in cognitive ability and experience level.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of deep learning and enhanced oil recovery, specifically relating to a method and system for predicting enhanced oil recovery based on a neural network model. Background Technology

[0002] Heterogeneous composite flooding, which injects viscoelastic particles, viscosifiers, and auxiliary agent solutions into reservoirs, can further enhance oil recovery in high water-cut oilfields and has been successfully applied in oilfields such as Shengli in my country. However, high water-cut oilfields have diverse geological conditions, and heterogeneous composite flooding involves large investments and high costs. Accurately evaluating the maximum enhanced oil recovery achievable by using optimal injection and production parameters while meeting economic requirements is crucial for the rational selection of reservoir blocks for further implementation of heterogeneous composite flooding and has significant practical guiding significance.

[0003] Currently, methods for predicting enhanced oil recovery (EOR) mainly include empirical analogy, regression formulas, and deep learning. Empirical analogy methods determine EOR values ​​by comparing key indicators such as formation crude oil viscosity, average permeability, and reservoir heterogeneity with developed reservoirs. However, due to the complex and varied geological conditions of actual reservoirs, only a general distribution range can be obtained through analogy, making accurate prediction of EOR values ​​impossible. Regression formula methods establish regression formulas based on the geological conditions and EOR data of developed reservoirs, then substitute these formulas with actual reservoir parameters to solve for the prediction. However, this type of method requires pre-assumed formula forms, is susceptible to human interference, and suffers from significant prediction errors when sample data is insufficient. Deep learning, through artificial neural networks, can obtain predictive models of the relationship between evaluation indicators and influencing factors without requiring pre-assumed correlations, and exhibits high prediction accuracy.

[0004] Artificial neural networks are trained on a large number of samples with known input and output parameters, and their weights are continuously adjusted through an internal adaptive algorithm to achieve the best fitting prediction effect. However, heterogeneous composite flooding, as an emerging method for enhancing oil recovery, suffers from limited field sample data, and no applied and validated neural network structures have been reported. Furthermore, currently known enhanced oil recovery values ​​for heterogeneous composite flooding are all obtained under single development conditions, while the maximum potential value for enhanced oil recovery under optimal development conditions remains unclear. Therefore, there is an urgent need to develop a method and system for predicting enhanced oil recovery based on a neural network model. Summary of the Invention

[0005] To address the shortcomings of existing technologies and the characteristics of neural network models, this invention proposes a method for predicting enhanced oil recovery based on a neural network model. The method involves selecting and determining reservoir parameters for the neural network input layer based on sensitivity analysis, using particle swarm optimization to determine the economic limit of enhanced oil recovery under optimal injection-production conditions in heterogeneous composite flooding, and using this limit as a prediction index for the neural network output layer. Numerical simulation is used to construct a large number of training samples to optimize and adjust the hidden layer network structure, thereby establishing a neural network prediction model for enhanced oil recovery.

[0006] Terminology Explanation:

[0007] Orthogonal experimental design: Orthogonal experimental design is a design method for studying the influence of multiple factors. It selects a portion of representative schemes from a comprehensive set of experimental schemes based on orthogonality and conducts experiments. It is a highly efficient, rapid, and economical experimental design method.

[0008] Numerical simulation of heterogeneous composite flooding: Using computers to solve the mathematical model of heterogeneous composite flooding reservoirs, simulating underground oil and water flow, providing the oil and water distribution at a certain moment, assisting in determining reasonable development schemes and adjustment measures, and predicting the development dynamics of heterogeneous composite flooding reservoirs.

[0009] Particle swarm optimization algorithm: By utilizing the information sharing among individuals in a swarm composed of multiple solutions, the entire swarm evolves from disorder to order in the iterative solution space, thereby obtaining the optimal solution to the problem.

[0010] The technical solution adopted in this invention is as follows:

[0011] A method for improving oil recovery based on neural network model prediction includes:

[0012] Build and train a neural network model;

[0013] The actual reservoir parameters of the oilfield development unit are input into the trained neural network model, and the economic limit of enhanced oil recovery that can be achieved by the heterogeneous composite flooding development of the oilfield development unit is predicted by the trained neural network model.

[0014] The neural network model includes an input layer, a hidden layer, and an output layer. The construction of the neural network model includes: screening the reservoir parameters of the input layer of the neural network model; using the particle swarm optimization algorithm to determine the enhanced oil recovery rate under the optimal injection and production conditions of heterogeneous composite flooding; and constructing the evaluation index of the output layer of the neural network model.

[0015] According to a preferred embodiment of the present invention, the reservoir parameters for screening the input layer of a neural network model include:

[0016] Based on understanding of mineral field development, the main influencing factors of enhanced oil recovery (EOR) in heterogeneous composite flooding are used as candidate parameters for the input layer of a neural network model. Numerical simulation studies of the single-factor influence law of EOR in heterogeneous composite flooding are conducted for each candidate parameter. The sensitivity coefficient of each candidate parameter is calculated, and candidate parameters with sensitivity coefficients greater than 0.005 are selected as reservoir parameters for the input layer of the neural network model. The single-factor influence law numerical simulation study is an existing technology, specifically referring to calculating the impact of a single factor on the research target when several different values ​​are taken, while keeping other factors unchanged. The single-factor influence law numerical simulation study of EOR in heterogeneous composite flooding involves changing only the magnitude of one factor at a time while keeping other factors unchanged, and then using heterogeneous composite flooding numerical simulation technology to perform simulation calculations and statistically analyze the changes in EOR corresponding to several different values ​​of that factor.

[0017] According to a preferred embodiment of the present invention, a particle swarm optimization algorithm is used to determine the enhanced oil recovery rate under optimal injection-production conditions in heterogeneous composite flooding, and an evaluation index for the output layer of a neural network model is constructed; including:

[0018] The curves of enhanced oil recovery (EOR) and incremental cumulative net present value (NPV) in heterogeneous composite flooding as a function of chemical injection volume were statistically analyzed. The EOR value corresponding to the gradual decrease of the incremental NPV from a positive value to 0 was taken as the economic limit EOR value of heterogeneous composite flooding and used as an evaluation index for the output layer of a neural network. The chemical agents injected into the reservoir in heterogeneous composite flooding include viscoelastic particles, thickeners, and additives. The statistical process for the curves of EOR and incremental NPV as a function of chemical injection volume in heterogeneous composite flooding was as follows: A series of chemical injection volume schemes (slug size, with three chemicals injected simultaneously and with consistent slug size) were designed from small to large. For each chemical injection volume scheme, the injection and production parameters of each well were optimized using a particle swarm optimization algorithm to obtain the maximum EOR and incremental NPV under that chemical injection volume. Curves were plotted based on the EOR and incremental NPV obtained under different chemical injection volumes.

[0019] Further preferred candidate parameters for the input layer of the neural network model include: average permeability, underground crude oil viscosity, interlayer permeability gradient, permeability variation coefficient, effective formation thickness, and reservoir pressure.

[0020] A further preferred method for calculating the sensitivity coefficient is shown in equation (1):

[0021] (1);

[0022] In equation (1), Represents the sensitivity coefficient; This indicates the number of values ​​used in a numerical simulation study of the single-factor influence of a candidate parameter. Indicates the first candidate parameter j Each value corresponds to a heterogeneous composite flooding method that improves oil recovery. Indicates a candidate parameter The values ​​represent the average values ​​of the enhanced oil recovery rates corresponding to heterogeneous composite flooding.

[0023] A further preferred formula for calculating the incremental cumulative net present value is shown in equation (2):

[0024] (2);

[0025] In equation (2), Represents the incremental cumulative net present value; Indicates the crude oil commodity rate; Indicates the first The annual increase in oil production from heterogeneous composite flooding; Indicates the selling price of crude oil; This indicates the operating cost per ton of oil. Indicates the first Annual viscoelastic particle injection volume; This indicates the purchase price of viscoelastic particles; Indicates the first Annual thickener injection volume; This indicates the purchase price of the tackifier; Indicates the first Annual dosage of auxiliary agents; Indicates the purchase price of the excipients; Indicates the resource tax rate; Indicates the overall tax rate; Indicates the rate of return; This indicates the number of years since the development of heterogeneous composite drives.

[0026] According to a preferred embodiment of the present invention, the training process of the neural network model includes:

[0027] Establish a training sample set for the neural network model; including: based on the reservoir parameters of the input layer of the determined neural network model, compile orthogonal test schemes; for each orthogonal test scheme, use particle swarm optimization algorithm to optimize the injection and production parameters of heterogeneous composite flooding to obtain the maximum economic limit enhanced oil recovery value that each orthogonal test scheme can achieve; use the reservoir parameters and the corresponding maximum economic limit enhanced oil recovery value in each orthogonal test scheme as input parameters and output parameters, respectively, to establish a training sample set for the neural network model;

[0028] Optimize the network structure of the hidden layers of a neural network model; including: based on a determined training sample set of the neural network model, by comparing the training fitting effect of the neural network model, optimize and determine the number of hidden layers and the number of neurons in each hidden layer of the neural network model, wherein the number of hidden layers ranges from 1 to 3, and the number of neurons in each hidden layer ranges from 5 to 10.

[0029] Further preferred, the injection and recovery parameters of heterogeneous composite flooding are optimized using a particle swarm optimization algorithm, including:

[0030] First, initialize the injection rate, viscoelastic particle injection concentration and volume, viscoelastic agent injection concentration and volume, auxiliary agent injection concentration and volume, and fluid production rate of each injection well; then, use a simulator to conduct a numerical simulation of heterogeneous composite flooding, and statistically analyze the annual increase in oil production, viscoelastic particle injection volume, viscoelastic agent injection volume, and auxiliary agent injection volume of heterogeneous composite flooding during the development process. Based on the formula for calculating the incremental cumulative net present value, calculate the first economic limit of enhanced oil recovery under the current combination of development parameters.

[0031] Then, the particle swarm optimization algorithm is used to update the dynamic parameters of the development, including the injection rate of each injection well, the concentration and volume of viscoelastic particles, the concentration and volume of viscosifier, the concentration and volume of auxiliary agent, and the fluid production rate of each production well. Then, the simulator is called to carry out the numerical simulation of heterogeneous composite flooding. According to the formula for calculating the incremental cumulative net present value, the second economic limit of improved recovery rate under the combination of development parameters updated by the particle swarm optimization algorithm is calculated.

[0032] Finally, the calculated first and second economic limit enhanced oil recovery rates are compared. If the difference is less than 0.5%, the larger of the two is taken as the maximum economic limit enhanced oil recovery rate achievable by the orthogonal experimental scheme after optimizing the injection and recovery parameters of the heterogeneous composite flooding using the particle swarm optimization algorithm. Otherwise, the second economic limit enhanced oil recovery rate is assigned to the first economic limit enhanced oil recovery rate, the particle swarm optimization algorithm is used to update the dynamic parameters, the simulator is called to carry out numerical simulation of heterogeneous composite flooding, and a new second economic limit enhanced oil recovery rate is calculated according to the formula for calculating incremental cumulative net present value.

[0033] Repeat the above steps until the difference between the first and second economic limit enhancement rates is less than 0.5%, and take the larger of the two as the maximum economic limit enhancement rate that the orthogonal experimental scheme can achieve.

[0034] Based on the formula for calculating incremental cumulative net present value, calculate the first economic limit for enhanced oil recovery under the current combination of development parameters; including:

[0035] The number of heterogeneous composite flooding development years corresponding to the point where the incremental cumulative net present value is 0 is obtained through a trial-and-error method. The enhanced oil recovery rate at this point is the first economic limit enhanced oil recovery rate. The specific implementation process includes:

[0036] The number of years of heterogeneous composite flooding in equation (2) Increasing sequentially, if After one year, the incremental cumulative net present value calculated by equation (2) changes from a positive value to a negative value. Therefore, further... Trial calculations were performed between the two points to obtain the accurate time corresponding to the incremental cumulative net present value of 0. The injection and extraction data for each full year before that time were statistically analyzed. For those less than a full year, they were recorded as one year. The first economic limit to improve the recovery rate was calculated, and the calculation formula is shown in Equation (3).

[0037] (3);

[0038] In equation (3), This indicates the first economic limit for increasing the recovery rate. Indicates the first The annual increase in oil production from heterogeneous composite flooding; Indicates the number of years since the development of heterogeneous composite drives; This indicates the original geological reserves of the oil reservoir.

[0039] According to a preferred embodiment of the present invention, the network structure of the hidden layer of an optimized neural network model includes:

[0040] First, based on the range of the number of hidden layers in the neural network model and the number of neurons in each hidden layer, the network structure of the hidden layers of the neural network model with different combinations of the number of layers and the number of neurons is established.

[0041] Secondly, the input parameters from the determined neural network training sample set are substituted into the network structure of the hidden layer of each neural network model to calculate the economic limit of enhanced oil recovery. The values ​​are then compared with the output parameters from the determined neural network training sample set. The network structure of the hidden layer of the neural network model with the smallest difference is selected as the network structure of the hidden layer of the neural network model to be used.

[0042] According to a preferred embodiment of the present invention, verifying a neural network model includes:

[0043] The input parameters from the determined neural network training sample set are substituted into the network structure of the hidden layer of the optimized neural network model to calculate the economic limit enhanced oil recovery value. This value is then compared with the output parameters from the determined neural network training sample set. If the coefficient of determination between the two is greater than 95%, it indicates that the accuracy of the established neural network model can be used to predict the economic limit enhanced oil recovery of heterogeneous composite flooding.

[0044] A further preferred formula for calculating the coefficient of determination is shown in equation (4):

[0045] (4);

[0046] In equation (4), Indicates the coefficient of determination; This represents the number of samples in the training sample set of the neural network; Indicates the first The actual values ​​of the output parameters for each training sample; express The average of the actual values ​​of the output parameters of each training sample; Indicates according to the first k The input parameters of each training sample are predicted values ​​calculated by the neural network model.

[0047] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a method for predicting improved oil recovery based on a neural network model.

[0048] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for predicting improved recovery rate based on a neural network model.

[0049] A system for improving oil recovery based on neural network model prediction includes:

[0050] The neural network model construction and training module is configured to: construct and train a neural network model; wherein the neural network model includes an input layer, a hidden layer, and an output layer; constructing the neural network model includes: screening the reservoir parameters of the input layer of the neural network model; using the particle swarm optimization algorithm to determine the enhanced oil recovery rate under the optimal injection and production conditions of heterogeneous composite flooding; and constructing the evaluation index of the output layer of the neural network model.

[0051] The prediction module is configured to input the actual reservoir parameters of the oilfield development unit into the trained neural network model, and predict the economic limit of enhanced oil recovery that the oilfield development unit can achieve by using heterogeneous composite flooding.

[0052] The beneficial effects of this invention are as follows:

[0053] 1. By establishing a prediction model for the economic limit of enhanced oil recovery in heterogeneous composite flooding through artificial neural networks, there is no need to pre-assume the correlation between prediction indicators and reservoir parameters, thus avoiding the problem of errors in prediction models caused by differences in cognitive ability and experience level.

[0054] 2. The injection and production parameters of heterogeneous composite flooding are optimized by particle swarm optimization to obtain the maximum economic limit for improving the recovery rate. Based on this, the hidden layer network structure of the neural network is trained to ensure that the prediction result of the neural network model is the maximum potential value for improving the recovery rate of heterogeneous composite flooding, which helps mine staff to make correct screening decisions.

[0055] 3. Through orthogonal experimental design and numerical simulation, the results of the economic limit of enhanced oil recovery under different reservoir parameter combinations were obtained, providing a large number of input and output parameters for neural network training, thus avoiding the problem of insufficient actual field samples making it difficult to apply neural network modeling. Attached Figure Description

[0056] Figure 1 A schematic diagram illustrating the process of constructing a neural network model;

[0057] Figure 2 This is a schematic diagram of a numerical simulation model for heterogeneous composite drive.

[0058] Figure 3 A schematic diagram illustrating the changes in oil recovery rate and incremental cumulative net present value with the amount of chemical agent injected;

[0059] Figure 4 This is a scatter plot comparing the neural network's predicted values ​​with the actual values. Detailed Implementation

[0060] To make the above and other objects, features and advantages of the present invention more apparent and understandable, a further detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments.

[0061] Example 1

[0062] A method for improving oil recovery based on neural network model prediction includes:

[0063] Build and train a neural network model;

[0064] The actual reservoir parameters of the oilfield development unit are input into the trained neural network model, and the economic limit of enhanced oil recovery that can be achieved by the heterogeneous composite flooding development of the oilfield development unit is predicted by the trained neural network model.

[0065] The neural network model includes an input layer, a hidden layer, and an output layer. The construction of the neural network model includes: screening the reservoir parameters of the input layer of the neural network model; using the particle swarm optimization algorithm to determine the enhanced oil recovery rate under the optimal injection and production conditions of heterogeneous composite flooding; and constructing the evaluation index of the output layer of the neural network model.

[0066] Example 2

[0067] The method for improving oil recovery rate based on neural network model prediction described in Example 1 differs in that:

[0068] Filtering reservoir parameters in the input layer of a neural network model; including:

[0069] Based on understanding of mineral field development, the main influencing factors of enhanced oil recovery (EOR) in heterogeneous composite flooding are used as candidate parameters for the input layer of a neural network model. Numerical simulation studies of the single-factor influence on EOR in heterogeneous composite flooding are conducted for each candidate parameter. The sensitivity coefficient of each candidate parameter is calculated, and candidate parameters with sensitivity coefficients greater than 0.005 are selected as reservoir parameters for the input layer of the neural network model. The single-factor influence numerical simulation study is an existing technology, specifically referring to calculating the impact of a single factor on the research target when several different values ​​are taken, while keeping other factors unchanged. The single-factor influence numerical simulation study of EOR in heterogeneous composite flooding involves changing only one factor at a time while keeping other factors unchanged, and then using heterogeneous composite flooding numerical simulation technology to perform simulation calculations and statistically analyze the changes in EOR corresponding to several different values ​​of that factor.

[0070] The particle swarm optimization algorithm is used to determine the enhanced oil recovery rate under optimal injection-production conditions in heterogeneous composite flooding, and evaluation indicators for the output layer of a neural network model are constructed, including:

[0071] The curves of enhanced oil recovery (EOR) and incremental cumulative net present value (NPV) in heterogeneous composite flooding as a function of chemical injection volume were statistically analyzed. The EOR value corresponding to the gradual decrease of the incremental NPV from a positive value to 0 was taken as the economic limit EOR value of heterogeneous composite flooding and used as an evaluation index for the output layer of a neural network. The chemical agents injected into the reservoir in heterogeneous composite flooding include viscoelastic particles, thickeners, and additives. The statistical process for the curves of EOR and incremental NPV as a function of chemical injection volume in heterogeneous composite flooding was as follows: A series of chemical injection volume schemes (slug size, with three chemicals injected simultaneously and with consistent slug size) were designed from small to large. For each chemical injection volume scheme, the injection and production parameters of each well were optimized using a particle swarm optimization algorithm to obtain the maximum EOR and incremental NPV under that chemical injection volume. Curves were plotted based on the EOR and incremental NPV obtained under different chemical injection volumes.

[0072] Candidate parameters for the input layer of the neural network model include: average permeability, underground crude oil viscosity, interlayer permeability gradient, permeability variation coefficient, effective formation thickness, and reservoir pressure.

[0073] The formula for calculating the sensitivity coefficient is shown in equation (1):

[0074] (1);

[0075] In equation (1), Represents the sensitivity coefficient; This indicates the number of values ​​used in a numerical simulation study of the single-factor influence of a candidate parameter. Indicates the first candidate parameter j Each value corresponds to a heterogeneous composite flooding method that improves oil recovery. Indicates a candidate parameter The values ​​represent the average values ​​of the enhanced oil recovery rates corresponding to heterogeneous composite flooding.

[0076] The formula for calculating the incremental cumulative net present value is shown in equation (2):

[0077] (2);

[0078] In equation (2), Represents the incremental cumulative net present value; Indicates the crude oil commodity rate; Indicates the first The increase in oil production from heterogeneous composite flooding in 2018; Indicates the selling price of crude oil; This indicates the operating cost per ton of oil. Indicates the first Annual viscoelastic particle injection volume; This indicates the purchase price of viscoelastic particles; Indicates the first Annual thickener injection volume; This indicates the purchase price of the tackifier; Indicates the first Annual dosage of auxiliary agents; Indicates the purchase price of the excipients; Indicates the resource tax rate; Indicates the overall tax rate; Indicates the rate of return; This indicates the number of years since the development of heterogeneous composite drives.

[0079] The training process of a neural network model includes:

[0080] Establish a training sample set for the neural network model; including: based on the reservoir parameters of the input layer of the determined neural network model, compile orthogonal test schemes; for each orthogonal test scheme, use particle swarm optimization algorithm to optimize the injection and production parameters of heterogeneous composite flooding to obtain the maximum economic limit enhanced oil recovery value that each orthogonal test scheme can achieve; use the reservoir parameters and the corresponding maximum economic limit enhanced oil recovery value in each orthogonal test scheme as input parameters and output parameters, respectively, to establish a training sample set for the neural network model;

[0081] Optimize the network structure of the hidden layers of a neural network model; including: based on a determined training sample set of the neural network model, by comparing the training fitting effect of the neural network model, optimize and determine the number of hidden layers and the number of neurons in each hidden layer of the neural network model, wherein the number of hidden layers ranges from 1 to 3, and the number of neurons in each hidden layer ranges from 5 to 10.

[0082] The particle swarm optimization algorithm is used to optimize the injection and recovery parameters of heterogeneous composite flooding, including:

[0083] First, initialize the injection rate, viscoelastic particle injection concentration and volume, viscoelastic agent injection concentration and volume, auxiliary agent injection concentration and volume, and fluid production rate of each injection well; then, use a simulator to conduct a numerical simulation of heterogeneous composite flooding, and statistically analyze the annual increase in oil production, viscoelastic particle injection volume, viscoelastic agent injection volume, and auxiliary agent injection volume of heterogeneous composite flooding during the development process. Based on the formula for calculating the incremental cumulative net present value, calculate the first economic limit of enhanced oil recovery under the current combination of development parameters.

[0084] Then, the particle swarm optimization algorithm is used to update the dynamic parameters of the development, including the injection rate of each injection well, the concentration and volume of viscoelastic particles, the concentration and volume of viscosifier, the concentration and volume of auxiliary agent, and the fluid production rate of each production well. Then, the simulator is called to carry out the numerical simulation of heterogeneous composite flooding. According to the formula for calculating the incremental cumulative net present value, the second economic limit of improved recovery rate under the combination of development parameters updated by the particle swarm optimization algorithm is calculated.

[0085] Finally, the calculated first and second economic limit enhanced oil recovery rates are compared. If the difference is less than 0.5%, the larger of the two is taken as the maximum economic limit enhanced oil recovery rate achievable by the orthogonal experimental scheme after optimizing the injection and recovery parameters of the heterogeneous composite flooding using the particle swarm optimization algorithm. Otherwise, the second economic limit enhanced oil recovery rate is assigned to the first economic limit enhanced oil recovery rate, the particle swarm optimization algorithm is used to update the dynamic parameters, the simulator is called to carry out numerical simulation of heterogeneous composite flooding, and a new second economic limit enhanced oil recovery rate is calculated according to the formula for calculating incremental cumulative net present value.

[0086] Repeat the above steps until the difference between the first and second economic limit enhancement rates is less than 0.5%, and take the larger of the two as the maximum economic limit enhancement rate that the orthogonal experimental scheme can achieve.

[0087] Based on the formula for calculating incremental cumulative net present value, calculate the first economic limit for enhanced oil recovery under the current combination of development parameters; including:

[0088] The number of heterogeneous composite flooding development years corresponding to the point where the incremental cumulative net present value is 0 is obtained through a trial-and-error method. The enhanced oil recovery rate at this point is the first economic limit enhanced oil recovery rate. The specific implementation process includes:

[0089] The number of years of heterogeneous composite flooding in equation (2) Increasing sequentially, if After one year, the incremental cumulative net present value calculated by equation (2) changes from a positive value to a negative value. Therefore, further... Trial calculations were performed between the two points to obtain the accurate time corresponding to the incremental cumulative net present value of 0. The injection and extraction data for each full year before that time were statistically analyzed. For those less than a full year, they were recorded as one year. The first economic limit to improve the recovery rate was calculated, and the calculation formula is shown in Equation (3).

[0090] (3);

[0091] In equation (3), This indicates the first economic limit for increasing the recovery rate. Indicates the first The annual increase in oil production from heterogeneous composite flooding; Indicates the number of years since the development of heterogeneous composite drives; This indicates the original geological reserves of the oil reservoir.

[0092] Optimize the network structure of the hidden layers in a neural network model; including:

[0093] First, based on the range of the number of hidden layers in the neural network model and the number of neurons in each hidden layer, the network structure of the hidden layers of the neural network model with different combinations of the number of layers and the number of neurons is established.

[0094] Secondly, the input parameters from the determined neural network training sample set are substituted into the network structure of the hidden layer of each neural network model to calculate the economic limit of enhanced oil recovery. The values ​​are then compared with the output parameters from the determined neural network training sample set. The network structure of the hidden layer of the neural network model with the smallest difference is selected as the network structure of the hidden layer of the neural network model to be used.

[0095] Validating the neural network model includes:

[0096] The input parameters from the determined neural network training sample set are substituted into the network structure of the hidden layer of the optimized neural network model to calculate the economic limit enhanced oil recovery value. This value is then compared with the output parameters from the determined neural network training sample set. If the coefficient of determination between the two is greater than 95%, it indicates that the accuracy of the established neural network model can be used to predict the economic limit enhanced oil recovery of heterogeneous composite flooding.

[0097] The formula for calculating the coefficient of determination is shown in equation (4):

[0098] (4);

[0099] In equation (4), Indicates the coefficient of determination; This represents the number of samples in the training sample set of the neural network; Indicates the first The actual values ​​of the output parameters for each training sample; express The average of the actual values ​​of the output parameters of each training sample; Indicates according to the first k The input parameters of each training sample are predicted values ​​calculated by the neural network model.

[0100] Example 3

[0101] The method for improving oil recovery rate based on neural network model prediction described in Example 1 differs in that:

[0102] like Figure 1 As shown, the steps are as follows:

[0103] Screening reservoir parameters for the input layer of the neural network model; including: establishing a heterogeneous composite flooding numerical simulation model based on the X reservoir physical property parameters provided by the oilfield and combined with the reservoir development characteristics, such as... Figure 2 As shown. Figure 2 In the model, INJ-1, INJ-2, and INJ-3 are the names of the injection wells; PRO-1, PRO-2, PRO-3, PRO-4, PRO-5, and PRO-6 are the names of the production wells. The heterogeneous composite flooding numerical simulation model adopts a rectangular grid system, which is divided into 53×53×5=14045 grids. The well grid adopts a row-column well grid, that is, the water injection well rows and the production well rows are arranged alternately.

[0104] Based on the understanding of mine development, the main influencing factors of the enhanced oil recovery rate of heterogeneous composite flooding are used as candidate parameters for the input layer of the neural network model, including: average permeability, underground crude oil viscosity, inter-layer permeability difference, permeability variation coefficient, effective formation thickness, and reservoir pressure.

[0105] For each candidate parameter, five equally spaced values ​​are selected and input into the heterogeneous composite flooding numerical simulation model for numerical calculation. The enhanced oil recovery value is statistically analyzed, and the sensitivity coefficient is calculated. The formula for calculating the sensitivity coefficient is shown in Equation (1):

[0106] (1);

[0107] In equation (1), Represents the sensitivity coefficient; This indicates the number of values ​​used in a numerical simulation study of the single-factor influence of a candidate parameter. Indicates the first candidate parameter j Each value corresponds to a heterogeneous composite flooding method that improves oil recovery. Indicates a candidate parameter The values ​​represent the average values ​​of the enhanced oil recovery rates corresponding to heterogeneous composite flooding.

[0108] Table 1. Candidate parameters and the obtained sensitivity coefficients;

[0109] Candidate parameters Average penetration underground crude oil viscosity Interlayer permeability gradient coefficient of variation of permeability Effective thickness of formation reservoir pressure Increase harvest rate by 1%. 15.29 18.80 15.67 15.06 15.40 15.29 Increase the recovery rate by 2%. 15.27 16.29 15.52 15.14 15.34 15.27 Increase harvest rate by 3%. 16.28 15.28 15.28 15.28 15.28 15.28 Increased harvest rate by 4%. 17.27 14.65 14.97 15.32 15.25 15.27 Increase harvest rate by 5%. 17.23 13.97 14.69 15.43 15.56 15.23 Sensitivity coefficient 0.054 0.106 0.023 0.008 0.007 0.001

[0110] As shown in Table 1, candidate parameters with a sensitivity coefficient greater than 0.005 were selected as reservoir parameters for the input layer of the neural network model, including average permeability, underground crude oil viscosity, interlayer permeability gradient, permeability variation coefficient, and effective formation thickness.

[0111] The evaluation index of the output layer of the neural network model is constructed; based on the numerical simulation results of heterogeneous composite flooding, the amount of oil added, the amount of viscoelastic particles injected, the amount of thickener injected, and the amount of auxiliary agent injected in each year during the development process are statistically analyzed, and the incremental cumulative net present value under different chemical agent injection amounts is calculated using formula (2):

[0112] (2);

[0113] In equation (2), Represents the incremental cumulative net present value; Indicates the crude oil commodity rate; Indicates the first The annual increase in oil production from heterogeneous composite flooding; Indicates the selling price of crude oil; This indicates the operating cost per ton of oil. Indicates the first Annual viscoelastic particle injection volume; This indicates the purchase price of viscoelastic particles; Indicates the first Annual thickener injection volume; This indicates the purchase price of the tackifier; Indicates the first Annual dosage of auxiliary agents; Indicates the purchase price of the excipients; Indicates the resource tax rate; Indicates the overall tax rate; Indicates the rate of return; This indicates the number of years since the development of heterogeneous composite drives.

[0114] Based on this, curves were plotted showing the changes in enhanced oil recovery rate and incremental cumulative net present value of heterogeneous composite flooding with the amount of chemical agent injected. Figure 3 The figure shows the results of a certain scheme. The enhanced oil recovery value corresponding to the incremental cumulative net present value gradually decreasing from a positive value to 0 is used as the economic limit enhanced oil recovery value of heterogeneous composite flooding, and is used as the evaluation index of the output layer of the neural network model. The economic limit enhanced oil recovery value corresponding to this embodiment is 17.85%.

[0115] Establish a training sample set for the neural network model; including: based on the reservoir parameters of the input layer of the determined neural network model, develop an orthogonal experimental scheme with 5 parameters and 7 levels; for each experimental scheme, use the particle swarm optimization algorithm to optimize the injection and production parameters of heterogeneous composite flooding, so as to obtain the maximum economic limit of enhanced oil recovery achievable by each orthogonal experimental scheme. The specific implementation process includes:

[0116] First, initialize the injection rate, viscoelastic particle injection concentration and volume, viscoelastic agent injection concentration and volume, auxiliary agent injection concentration and volume, and fluid production rate of each injection well; then, use a simulator to conduct a numerical simulation of heterogeneous composite flooding, and statistically analyze the annual increase in oil production, viscoelastic particle injection volume, viscoelastic agent injection volume, and auxiliary agent injection volume of heterogeneous composite flooding during the development process. Based on the formula for calculating the incremental cumulative net present value, calculate the first economic limit of enhanced oil recovery under the current combination of development parameters.

[0117] Then, the particle swarm optimization algorithm is used to update the dynamic parameters of the development, including the injection rate of each injection well, the concentration and volume of viscoelastic particles, the concentration and volume of viscosifier, the concentration and volume of auxiliary agent, and the fluid production rate of each production well. Then, the simulator is called to carry out the numerical simulation of heterogeneous composite flooding. According to the formula for calculating the incremental cumulative net present value, the second economic limit of improved recovery rate under the combination of development parameters updated by the particle swarm optimization algorithm is calculated.

[0118] Finally, the calculated first and second economic limit enhanced oil recovery rates are compared. If the difference is less than 0.5%, the larger of the two is taken as the maximum economic limit enhanced oil recovery rate achievable by the orthogonal experimental scheme after optimizing the injection and recovery parameters of the heterogeneous composite flooding using the particle swarm optimization algorithm. Otherwise, the second economic limit enhanced oil recovery rate is assigned to the first economic limit enhanced oil recovery rate, the particle swarm optimization algorithm is used to update the dynamic parameters, the simulator is called to carry out numerical simulation of heterogeneous composite flooding, and a new second economic limit enhanced oil recovery rate is calculated according to the formula for calculating incremental cumulative net present value.

[0119] Repeat the above steps until the difference between the first and second economic limit enhancement rates is less than 0.5%, and take the larger of the two as the maximum economic limit enhancement rate that the orthogonal experimental scheme can achieve.

[0120] In this embodiment, the calculation results of the orthogonal experimental scheme and the economic limit enhanced oil recovery value obtained by particle swarm optimization are shown in Table 2.

[0121] Table 2. Calculation results of orthogonal experimental schemes and economic limit enhanced oil recovery values ​​obtained by particle swarm optimization.

[0122] Serial Number Effective thickness of formation Average penetration underground crude oil viscosity Interlayer permeability gradient Permeability variation coefficient Economic limits to increase recovery rate 1 6 60 10 1 0.2 11.52 2 6 500 60 5 1.2 14.23 3 6 1000 20 11 0.8 20.15 4 6 1500 80 3 0.4 17.44 5 6 2000 30 9 1.4 21.73 6 6 1500 100 2 1 20.84 7 6 2000 50 7 0.6 24.92 8 8 60 100 9 1.2 7.07 9 8 500 50 2 0.8 13.03 10 8 1000 10 7 0.4 19.49 11 8 500 60 1 1.4 17.70 12 8 1000 20 5 1 21.81 13 8 1500 80 11 0.6 21.43 14 8 2000 30 3 0.2 27.21 15 10 60 80 5 0.8 6.34 16 10 2000 30 11 0.4 14.11 17 10 60 100 3 1.4 13.89 18 10 500 50 9 1 17.40 19 10 1000 10 2 0.6 25.03 20 10 1500 60 7 0.2 22.48 21 10 1000 20 1 1.2 27.17 22 12 1500 60 2 0.4 7.17 23 12 2000 20 7 1.4 13.77 24 12 60 80 1 1 15.30 25 12 500 30 5 0.6 18.88 26 12 2000 100 11 0.2 17.10 27 12 60 50 3 1.2 23.53 28 12 500 10 9 0.8 28.17 29 14 1000 50 11 1.4 7.63 30 14 1500 10 3 1 16.44 31 14 1000 60 9 0.6 15.07 32 14 1500 20 2 0.2 19.62 33 14 2000 80 7 1.2 17.45 34 14 60 30 1 0.8 22.79 35 14 500 100 5 0.4 22.33 36 16 60 30 7 1 9.16 37 16 500 100 1 0.6 9.71 38 16 1000 50 5 0.2 14.49 39 16 1500 10 11 1.2 19.91 40 16 2000 60 3 0.8 16.72 41 16 1500 20 9 0.4 24.04 42 16 2000 80 2 1.4 22.80 43 18 60 20 3 0.6 8.45 44 18 500 80 9 0.2 10.36 45 18 1000 30 2 1.2 14.47 46 18 500 100 7 0.8 16.19 47 18 1000 50 1 0.4 19.15 48 18 1500 10 5 1.4 25.62 49 18 2000 60 11 1 22.92

[0123] As shown in Table 2, the average permeability, underground crude oil viscosity, inter-layer permeability difference, permeability variation coefficient and effective formation thickness in each orthogonal experimental scheme were used as input parameters for the training samples of the neural network model, and the economic limit enhanced oil recovery rate was used as the output parameter for the training samples of the neural network model, thus establishing a training sample set of the neural network model containing a total of 49 samples.

[0124] Optimize the hidden layer network structure of the neural network; in this embodiment, the number of hidden layers in the neural network model ranges from 1 to 3, and the number of neurons in each hidden layer ranges from 5 to 10. Establish the hidden layer network structure of the neural network model with different combinations of the number of layers and the number of neurons; substitute the input parameters in the determined neural network training sample set into the hidden layer network structure of each neural network model, calculate the economic limit enhanced recovery rate value, and compare it with the output parameters in the determined neural network training sample set. The deviation of the calculated values ​​of each network structure is shown in Table 3.

[0125] Table 3. Deviation of calculated values ​​for each network structure;

[0126] Serial Number number of floors Number of neurons Calculated deviation Serial Number number of floors Number of neurons Calculated deviation 1 1 5 0.0012 10 2 8 0.0007 2 1 6 0.0011 11 2 9 0.0006 3 1 7 0.0009 12 2 10 0.0006 4 1 8 0.0009 13 3 5 0.0009 5 1 9 0.0008 14 3 6 0.0007 6 1 10 0.0008 15 3 7 0.0007 7 2 5 0.0010 16 3 8 0.0006 8 2 6 0.0009 17 3 9 0.0006 9 2 7 0.0008 18 3 10 0.0006

[0127] As shown in Table 3, the minimum deviation of the calculated values ​​for each network structure is 0.0006. Under the same deviation, the network structure with the smaller number of layers and neurons is selected. Therefore, the hidden layer network structure of the neural network model selected in this embodiment has 2 layers and 9 neurons.

[0128] Validate the neural network prediction model;

[0129] The input parameters from the determined neural network training sample set are substituted into the network structure of the hidden layer of the optimized neural network model to calculate the economic limit of enhanced oil recovery. This value is then compared with the output parameters from the determined neural network training sample set. The scatter plot distribution is as follows: Figure 4 As shown. The determination coefficient between the two calculated using formula (4) is 98.92%, which is greater than 95%. Therefore, the established neural network model can be used to predict the economic limit of heterogeneous composite flooding to improve the recovery rate.

[0130] The formula for calculating the coefficient of determination is shown in equation (4):

[0131] (4);

[0132] In equation (4), Indicates the coefficient of determination; This represents the number of samples in the training sample set of the neural network; Indicates the first The actual values ​​of the output parameters for each training sample; express The average of the actual values ​​of the output parameters of each training sample; Indicates according to the first k The input parameters of each training sample are predicted values ​​calculated by the neural network model.

[0133] The actual reservoir parameters of an oilfield development unit are input into a trained neural network model. The trained neural network model then predicts the economic limit of enhanced oil recovery achievable by heterogeneous composite flooding in that oilfield development unit. The average permeability of the reservoir in an oilfield where heterogeneous composite flooding is planned is 1204 × 10⁻⁶. -3 µm 2 The underground crude oil viscosity is 67 mPa·s, the interlayer permeability difference is 5.4, the permeability variation coefficient is 0.76, and the effective formation thickness is 13.7 m. Substituting the above reservoir parameters into the validated neural network model, the economic limit of enhanced oil recovery that can be achieved by using heterogeneous composite flooding in this oilfield development unit is calculated to be 18.7%.

[0134] Example 4

[0135] A computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method for improving oil recovery based on neural network model prediction as described in any of Examples 1-3.

[0136] Example 5

[0137] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for improving recovery rate based on neural network model prediction as described in any of Embodiments 1-3.

[0138] Example 6

[0139] A system for improving oil recovery based on neural network model prediction includes:

[0140] The neural network model construction and training module is configured to: construct and train a neural network model; wherein the neural network model includes an input layer, a hidden layer, and an output layer; constructing the neural network model includes: screening the reservoir parameters of the input layer of the neural network model; using the particle swarm optimization algorithm to determine the enhanced oil recovery rate under the optimal injection and production conditions of heterogeneous composite flooding; and constructing the evaluation index of the output layer of the neural network model.

[0141] The prediction module is configured to input the actual reservoir parameters of the oilfield development unit into the trained neural network model, and predict the economic limit of enhanced oil recovery that the oilfield development unit can achieve by using heterogeneous composite flooding.

Claims

1. A method for improving oil recovery rate based on neural network model prediction, characterized in that, include: Build and train a neural network model; The actual reservoir parameters of the oilfield development unit are input into the trained neural network model, and the economic limit of enhanced oil recovery that can be achieved by the heterogeneous composite flooding development of the oilfield development unit is predicted by the trained neural network model. The neural network model includes an input layer, a hidden layer, and an output layer; constructing the neural network model includes: selecting reservoir parameters for the input layer of the neural network model; The particle swarm optimization algorithm is used to determine the enhanced oil recovery rate under the optimal injection and production conditions of heterogeneous composite flooding, and the evaluation index of the output layer of the neural network model is constructed. This includes: statistically analyzing the changes in the enhanced oil recovery rate and incremental cumulative net present value of heterogeneous composite flooding with the amount of chemical agent injected, and using the enhanced oil recovery rate corresponding to the incremental cumulative net present value gradually decreasing from a positive value to 0 as the evaluation index of the output layer of the neural network. The training process of a neural network model includes: Establish a training sample set for the neural network model; including: based on the reservoir parameters of the input layer of the determined neural network model, compile orthogonal test schemes; for each orthogonal test scheme, use particle swarm optimization algorithm to optimize the injection and production parameters of heterogeneous composite flooding to obtain the maximum economic limit enhanced oil recovery value that each orthogonal test scheme can achieve; use the reservoir parameters and the corresponding maximum economic limit enhanced oil recovery value in each orthogonal test scheme as input parameters and output parameters, respectively, to establish a training sample set for the neural network model; Optimize the network structure of the hidden layers of a neural network model; including: based on a determined training sample set of the neural network model, by comparing the training fitting effect of the neural network model, optimize and determine the number of hidden layers and the number of neurons in each hidden layer of the neural network model, wherein the number of hidden layers ranges from 1 to 3, and the number of neurons in each hidden layer ranges from 5 to 10. The particle swarm optimization algorithm is used to optimize the injection and recovery parameters of heterogeneous composite flooding, including: First, initialize the injection rate, viscoelastic particle injection concentration and volume, viscoelastic agent injection concentration and volume, auxiliary agent injection concentration and volume, and fluid production rate of each injection well; then, use a simulator to conduct a numerical simulation of heterogeneous composite flooding, and statistically analyze the annual increase in oil production, viscoelastic particle injection volume, viscoelastic agent injection volume, and auxiliary agent injection volume of heterogeneous composite flooding during the development process. Based on the formula for calculating the incremental cumulative net present value, calculate the first economic limit of enhanced oil recovery under the current combination of development parameters. Then, the particle swarm optimization algorithm is used to update the dynamic parameters of the development, including the injection rate of each injection well, the concentration and volume of viscoelastic particles, the concentration and volume of viscosifier, the concentration and volume of auxiliary agent, and the fluid production rate of each production well. Then, the simulator is called to carry out the numerical simulation of heterogeneous composite flooding. According to the formula for calculating the incremental cumulative net present value, the second economic limit of improved recovery rate under the combination of development parameters updated by the particle swarm optimization algorithm is calculated. Finally, the calculated first and second economic limit enhanced oil recovery rates are compared. If the difference is less than 0.5%, the larger of the two is taken as the maximum economic limit enhanced oil recovery rate achievable by the orthogonal experimental scheme after optimizing the injection and recovery parameters of the heterogeneous composite flooding using the particle swarm optimization algorithm. Otherwise, the second economic limit enhanced oil recovery rate is assigned to the first economic limit enhanced oil recovery rate, the particle swarm optimization algorithm is used to update the dynamic parameters, the simulator is called to carry out numerical simulation of heterogeneous composite flooding, and a new second economic limit enhanced oil recovery rate is calculated according to the formula for calculating incremental cumulative net present value. Repeat the above steps until the difference between the first and second economic limit enhancement rates is less than 0.5%, and take the larger of the two as the maximum economic limit enhancement rate that the orthogonal experimental scheme can achieve. The formula for calculating the first economic limit to increase the recovery rate is shown in equation (3): （3） In equation (3), This indicates the first economic limit for increasing the recovery rate. Indicates the first The annual increase in oil production from heterogeneous composite flooding; Indicates the number of years since the development of heterogeneous composite drives; This indicates the original geological reserves of the oil reservoir.

2. The method for improving oil recovery rate based on neural network model prediction according to claim 1, characterized in that, Filtering reservoir parameters in the input layer of a neural network model; including: Based on the understanding of mine development, the main influencing factors of the enhanced oil recovery rate of heterogeneous composite flooding are used as candidate parameters of the input layer of the neural network model. Numerical simulation studies on the single-factor influence law of enhanced oil recovery of heterogeneous composite flooding are carried out for each candidate parameter. The sensitivity coefficient of each candidate parameter is calculated, and candidate parameters with sensitivity coefficients greater than 0.005 are selected as reservoir parameters of the input layer of the neural network model. Candidate parameters for the input layer of the neural network model include: average permeability, underground crude oil viscosity, interlayer permeability gradient, permeability variation coefficient, effective formation thickness, and reservoir pressure. The formula for calculating the sensitivity coefficient is shown in equation (1): （1） In equation (1), Represents the sensitivity coefficient; This indicates the number of values ​​used in a numerical simulation study of the single-factor influence of a candidate parameter. This indicates that the j-th value of a candidate parameter corresponds to the improved oil recovery rate in heterogeneous composite flooding. Indicates a candidate parameter The values ​​represent the average values ​​of the enhanced oil recovery rates corresponding to heterogeneous composite flooding.

3. The method for improving oil recovery rate based on neural network model prediction according to claim 1, characterized in that, The formula for calculating the incremental cumulative net present value is shown in equation (2): （2） In equation (2), Represents the incremental cumulative net present value; Indicates the crude oil commodity rate; Indicates the first The annual increase in oil production from heterogeneous composite flooding; Indicates the selling price of crude oil; This indicates the operating cost per ton of oil. Indicates the first Annual viscoelastic particle injection volume; This indicates the purchase price of viscoelastic particles; Indicates the first Annual thickener injection volume; This indicates the purchase price of the tackifier; Indicates the first Annual dosage of auxiliary agents; Indicates the purchase price of the excipients; Indicates the resource tax rate; Indicates the overall tax rate; Indicates the rate of return; This indicates the number of years since the development of heterogeneous composite drives.

4. The method for improving oil recovery rate based on neural network model prediction according to claim 1, characterized in that, Optimize the network structure of the hidden layers in a neural network model; including: First, based on the range of the number of hidden layers in the neural network model and the number of neurons in each hidden layer, the network structure of the hidden layers of the neural network model with different combinations of the number of layers and the number of neurons is established. Secondly, the input parameters from the determined neural network training sample set are substituted into the network structure of the hidden layer of each neural network model to calculate the economic limit of enhanced oil recovery. The values ​​are then compared with the output parameters from the determined neural network training sample set. The network structure of the hidden layer of the neural network model with the smallest difference is selected as the network structure of the hidden layer of the neural network model to be used.

5. The method for improving oil recovery rate based on neural network model prediction according to claim 1, characterized in that, Validating the neural network model includes: The input parameters from the determined neural network training sample set are substituted into the network structure of the hidden layer of the optimized neural network model to calculate the economic limit enhanced oil recovery value. This value is then compared with the output parameters from the determined neural network training sample set. If the coefficient of determination between the two is greater than 95%, it indicates that the accuracy of the established neural network model can be used to predict the economic limit enhanced oil recovery of heterogeneous composite flooding.

6. The method for improving oil recovery rate based on neural network model prediction according to claim 5, characterized in that, The formula for calculating the coefficient of determination is shown in equation (4): （4） In equation (4), Indicates the coefficient of determination; This represents the number of samples in the training sample set of the neural network; This represents the actual value of the output parameter of the k-th training sample; express The average of the actual values ​​of the output parameters of each training sample; This represents the predicted value calculated by the neural network model based on the input parameters of the k-th training sample.

7. A system for improving oil recovery rate based on neural network model prediction, characterized in that, include: The neural network model construction and training module is configured to: construct and train a neural network model; wherein the neural network model includes an input layer, a hidden layer, and an output layer; constructing the neural network model includes: screening the reservoir parameters of the input layer of the neural network model; using the particle swarm optimization algorithm to determine the enhanced oil recovery rate under the optimal injection and production conditions of heterogeneous composite flooding; and constructing the evaluation index of the output layer of the neural network model. The prediction module is configured to input the actual reservoir parameters of the oilfield development unit into the trained neural network model, and predict the economic limit of enhanced oil recovery that the oilfield development unit can achieve by using heterogeneous composite flooding. The particle swarm optimization algorithm is used to determine the enhanced oil recovery rate under the optimal injection and production conditions of heterogeneous composite flooding, and the evaluation index of the output layer of the neural network model is constructed. This includes: statistically analyzing the changes in the enhanced oil recovery rate and incremental cumulative net present value of heterogeneous composite flooding with the amount of chemical agent injected, and using the enhanced oil recovery rate corresponding to the incremental cumulative net present value gradually decreasing from a positive value to 0 as the evaluation index of the output layer of the neural network. The training process of a neural network model includes: Establish a training sample set for the neural network model; including: based on the reservoir parameters of the input layer of the determined neural network model, compile orthogonal test schemes; for each orthogonal test scheme, use particle swarm optimization algorithm to optimize the injection and production parameters of heterogeneous composite flooding to obtain the maximum economic limit enhanced oil recovery value that each orthogonal test scheme can achieve; use the reservoir parameters and the corresponding maximum economic limit enhanced oil recovery value in each orthogonal test scheme as input parameters and output parameters, respectively, to establish a training sample set for the neural network model; Optimize the network structure of the hidden layers of a neural network model; including: based on a determined training sample set of the neural network model, by comparing the training fitting effect of the neural network model, optimize and determine the number of hidden layers and the number of neurons in each hidden layer of the neural network model, wherein the number of hidden layers ranges from 1 to 3, and the number of neurons in each hidden layer ranges from 5 to 10. The particle swarm optimization algorithm is used to optimize the injection and recovery parameters of heterogeneous composite flooding, including: First, initialize the injection rate, viscoelastic particle injection concentration and volume, viscoelastic agent injection concentration and volume, auxiliary agent injection concentration and volume, and fluid production rate of each injection well; then, use a simulator to conduct a numerical simulation of heterogeneous composite flooding, and statistically analyze the annual increase in oil production, viscoelastic particle injection volume, viscoelastic agent injection volume, and auxiliary agent injection volume of heterogeneous composite flooding during the development process. Based on the formula for calculating the incremental cumulative net present value, calculate the first economic limit of enhanced oil recovery under the current combination of development parameters. Then, the particle swarm optimization algorithm is used to update the dynamic parameters of the development, including the injection rate of each injection well, the concentration and volume of viscoelastic particles, the concentration and volume of viscosifier, the concentration and volume of auxiliary agent, and the fluid production rate of each production well. Then, the simulator is called to carry out the numerical simulation of heterogeneous composite flooding. According to the formula for calculating the incremental cumulative net present value, the second economic limit of improved recovery rate under the combination of development parameters updated by the particle swarm optimization algorithm is calculated. Finally, the calculated first and second economic limit enhanced oil recovery rates are compared. If the difference is less than 0.5%, the larger of the two is taken as the maximum economic limit enhanced oil recovery rate achievable by the orthogonal experimental scheme after optimizing the injection and recovery parameters of the heterogeneous composite flooding using the particle swarm optimization algorithm. Otherwise, the second economic limit enhanced oil recovery rate is assigned to the first economic limit enhanced oil recovery rate, the particle swarm optimization algorithm is used to update the dynamic parameters, the simulator is called to carry out numerical simulation of heterogeneous composite flooding, and a new second economic limit enhanced oil recovery rate is calculated according to the formula for calculating incremental cumulative net present value. Repeat the above steps until the difference between the first and second economic limit enhancement rates is less than 0.5%, and take the larger of the two as the maximum economic limit enhancement rate that the orthogonal experimental scheme can achieve. The formula for calculating the first economic limit to increase the recovery rate is shown in equation (3): （3） In equation (3), This indicates the first economic limit for increasing the recovery rate. Indicates the first The annual increase in oil production from heterogeneous composite flooding; Indicates the number of years since the development of heterogeneous composite drives; This indicates the original geological reserves of the oil reservoir.

Images