A method, device and medium for predicting a geological carbon dioxide storage capacity

Abstract

The application discloses a carbon dioxide geological storage amount prediction method, device and medium, relates to the field of carbon dioxide capture, utilization and storage technology, and comprises the following steps: determining whether a to-be-predicted oil and gas reservoir and a control oil and gas reservoir are in the same block; if yes, inputting geological data, production dynamic data and injection parameter combination schemes of the to-be-predicted oil and gas reservoir into an optimal control oil and gas reservoir carbon dioxide geological storage amount prediction model to determine a full-area cumulative carbon dioxide geological storage amount of the to-be-predicted oil and gas reservoir; if not, obtaining geological data, production data and injection parameter value ranges of the to-be-predicted oil and gas reservoir, performing migration training on the optimal control oil and gas reservoir carbon dioxide geological storage amount prediction model to obtain a to-be-predicted oil and gas reservoir carbon dioxide geological storage amount prediction model, and then determining the full-area cumulative carbon dioxide geological storage amount of the to-be-predicted oil and gas reservoir. The application improves the accuracy and efficiency of carbon dioxide geological storage amount prediction.

Description

Technical Field

[0001] This application relates to the field of carbon dioxide capture, utilization and storage technology, and in particular to a method, equipment and medium for predicting the geological reserves of carbon dioxide. Background Technology

[0002] After a period of development, oil and gas reservoirs may become unproductive due to limitations in technology and economic conditions at the time, rendering some crude oil unrecoverable. These reservoirs are often referred to as abandoned oil and gas reservoirs. Utilizing abandoned oil and gas reservoirs for geological carbon dioxide sequestration offers several advantages. First, it allows for the full utilization of existing exploration and development data, well sites, and injection equipment, saving investment and engineering time. Second, sequestering industrially emitted carbon dioxide within these reservoirs effectively reduces overall carbon dioxide emissions. Accurately predicting the amount of carbon dioxide that can be geologically sequestered is crucial for the selection and site evaluation of abandoned oil and gas reservoirs.

[0003] Currently, the commonly used method for predicting carbon dioxide geological reserves is numerical simulation. This involves setting geological and injection parameters based on field experience in the target block and conducting reservoir numerical simulations to obtain the carbon dioxide geological reserves. This method requires a large amount of data, much of which is difficult to obtain in the oilfield. Moreover, numerical simulations need to be repeated for each block and each set of injection parameters, making the calculations very time-consuming and impractical. Summary of the Invention

[0004] The purpose of this application is to provide a method, equipment, and medium for predicting carbon dioxide geological reserves, which can improve the accuracy and efficiency of carbon dioxide geological reserves prediction.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] Firstly, this application provides a method for predicting the geological reserves of carbon dioxide, including:

[0007] Determine whether the oil and gas reservoir to be predicted and the control oil and gas reservoir belong to the same block;

[0008] If so, the geological data, production dynamics data, and injection parameter combination scheme of the oil and gas reservoir to be predicted are input into the optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model to obtain the total cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir to be predicted. The optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model is obtained by using the geological data, production dynamics data, and injection parameter combination scheme of the control oil and gas reservoir as input and the total cumulative carbon dioxide geological sequestration of the control oil and gas reservoir as output. The hyperparameters in the multi-feature embedded carbon dioxide geological sequestration prediction model are optimized by using an optimization algorithm, and the multi-feature embedded carbon dioxide geological sequestration prediction model is obtained after multiple trainings.

[0009] If not, obtain the geological data, production dynamics data, and injection parameter range of the oil and gas reservoir to be predicted;

[0010] Based on the geological data and production dynamic data of the oil and gas reservoir to be predicted, a numerical simulation model of carbon dioxide sequestration in the oil and gas reservoir to be predicted is constructed.

[0011] A sampling is conducted within the range of carbon dioxide injection parameters for the oil and gas reservoir to be predicted to obtain multiple sets of carbon dioxide injection parameter combinations for the oil and gas reservoir to be predicted; the carbon dioxide injection parameters include, but are not limited to: carbon dioxide injection volume, carbon dioxide injection rate, carbon dioxide injection time, carbon dioxide injection pressure, size of the injected carbon dioxide slug, water / gas injection ratio, and well shut-in time.

[0012] Multiple sets of injection parameter combinations for the oil and gas reservoirs to be predicted are input into the numerical simulation model of carbon dioxide sequestration of the oil and gas reservoirs to be predicted, and the total cumulative geological carbon dioxide sequestration of the whole area corresponding to different injection parameter combinations for the oil and gas reservoirs to be predicted is obtained.

[0013] Using the geological data and production dynamic data of the oil and gas reservoir to be predicted, as well as the injection parameter combination scheme of the oil and gas reservoir to be predicted, as input, and the total cumulative carbon dioxide geological sequestration of the whole area corresponding to the injection parameter combination scheme of the oil and gas reservoir to be predicted as output, the optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model is transferred and trained to obtain the prediction model of carbon dioxide geological sequestration of the oil and gas reservoir to be predicted.

[0014] The geological data, production dynamic data, and injection parameter combination scheme of the oil and gas reservoir to be predicted are input into the prediction model of carbon dioxide geological sequestration of the oil and gas reservoir to be predicted, so as to obtain the cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir.

[0015] Optionally, the multi-feature embedded carbon dioxide geological sequestration prediction model includes a physical feature layer and a multi-feature embedded prediction layer connected in sequence.

[0016] Optionally, the physical feature layer includes: a geological data feature sequence extraction branch, a production dynamics data feature sequence extraction branch, a parameter feature sequence extraction branch, a trend feature sequence extraction branch, and a splicing layer; the output ends of the production dynamics data feature sequence extraction branch, the geological data feature sequence extraction branch, the parameter feature sequence extraction branch, and the trend feature sequence extraction branch are all connected to the input end of the splicing layer;

[0017] The geological data feature sequence extraction branch includes a geological encoder; the input of the geological data feature sequence extraction branch is the geological data of the oil and gas reservoir to be predicted.

[0018] The feature sequence extraction branch of the production dynamic data includes a production encoder; the input of the feature sequence extraction branch of the production dynamic data is the production dynamic data of the oil and gas reservoir to be predicted.

[0019] The parameter feature sequence extraction branch includes a parallel value encoder and a position encoder; the input of the parameter feature sequence extraction branch is the geological data and production dynamic data of the oil and gas reservoir to be predicted.

[0020] The trend feature sequence extraction branch includes a trend encoder; the input to the trend feature sequence extraction branch is the geological data and production dynamic data of the oil and gas reservoir to be predicted.

[0021] The production dynamic data feature sequence extraction branch and the geological data feature sequence extraction branch extract the overall features of all parameters in the production dynamic data and geological data, respectively. The parameter feature sequence extraction branch extracts the individual features of any parameter in the production dynamic data and geological data, and the trend feature sequence extraction branch extracts the trend features of any parameter in the production dynamic data and geological data. The splicing layer is formed by splicing the production dynamic data feature sequence extraction branch, the geological data feature sequence extraction branch, the parameter feature sequence extraction branch, and the trend feature sequence extraction branch in a horizontal splicing manner. The output end of the splicing layer is connected to the multi-feature embedding prediction layer. The horizontal splicing method aligns the heads of the production dynamic data feature sequence extraction branch, the geological data feature sequence extraction branch, the parameter feature sequence extraction branch, and the trend feature sequence extraction branch, and splices the aligned extraction branches in a left-to-right order to obtain the splicing layer.

[0022] Optionally, the multi-feature embedding prediction layer includes:

[0023] The branches at time t, time t+1, the first fully connected layer, and the activation function are connected in sequence.

[0024] The input terminals of the branch at time t and the branch at time t+1 are both connected to the output terminal of the physical feature layer.

[0025] The input terminals of the branch at time t and the branch at time t+1 are also used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted.

[0026] Optionally, the branch at time t includes: a first multi-head attention layer, a first residual layer, a first regularization layer, a second fully connected layer, a second residual layer, and a second regularization layer connected in sequence;

[0027] The branch at time t+1 includes: a second multi-head attention layer, a third residual layer, a third regularization layer, a third fully connected layer, a fourth residual layer, and a fourth regularization layer connected in sequence;

[0028] The input terminals of the first multi-head attention layer and the first regularization layer are used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted and the data at time t in the physical feature sequence; the physical feature sequence is obtained by splicing the production dynamic data feature sequence, geological data feature sequence, parameter feature sequence and trend feature sequence;

[0029] The output of the first regularization layer is also connected to the input of the second regularization layer;

[0030] The output of the second regularization layer is also connected to the input of the third regularization layer;

[0031] The input end of the second multi-head attention layer is used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted and the data at time t+1 in the physical feature sequence;

[0032] The output of the third regularization layer is also connected to the input of the fourth regularization layer;

[0033] The output of the fourth regularization layer is also connected to the input of the first fully connected layer. Optionally, before determining whether the oil and gas reservoir to be predicted and the control oil and gas reservoir belong to the same block, the method further includes:

[0034] Obtain geological data, production dynamics data, and injection parameter ranges for the reference oil and gas reservoir;

[0035] Based on the geological and production dynamic data of the reference oil and gas reservoir, a numerical simulation model of carbon dioxide sequestration of the reference oil and gas reservoir was constructed.

[0036] Secondary sampling was conducted within the range of carbon dioxide injection parameters to obtain multiple sets of carbon dioxide injection parameter combinations for reference oil and gas reservoirs.

[0037] Multiple sets of injection parameter combinations for the control oil and gas reservoirs were input into the numerical simulation model of carbon dioxide sequestration for the control oil and gas reservoirs to obtain the total cumulative geological carbon dioxide sequestration in the whole area corresponding to different injection parameter combinations for the control oil and gas reservoirs.

[0038] A small sample dataset was constructed using the control oil and gas reservoir injection parameter combination scheme as input and the total cumulative carbon dioxide geological sequestration in the whole area corresponding to different control oil and gas reservoir injection parameter combination schemes as output.

[0039] The small sample dataset is augmented using small perturbation data to obtain an augmented sample dataset. The augmentation process includes, but is not limited to, small perturbation data conforming to Gaussian, uniform, or Laplace distributions. The small sample dataset and the augmented sample dataset are then concatenated vertically to obtain a training dataset. The vertical concatenation method aligns the left sides of the augmented sample dataset and the small sample dataset, and the aligned datasets are stacked sequentially from top to bottom to obtain the training dataset.

[0040] Using geological and production dynamic data of the control oil and gas reservoir, as well as the injection parameter combination scheme of the control oil and gas reservoir in the training dataset as input, and the cumulative carbon dioxide geological storage of the entire region in the training dataset as output, the multi-feature embedded carbon dioxide geological storage prediction model is trained multiple times. The hyperparameters in the multi-feature embedded carbon dioxide geological storage prediction model are optimized using optimization algorithms. The evaluation index of the control oil and gas reservoir carbon dioxide geological storage prediction model obtained in each training is determined. Based on the evaluation index, the optimal control oil and gas reservoir carbon dioxide geological storage prediction model is obtained. The evaluation index includes, but is not limited to, goodness of fit to perturbation, error of the storage process under small perturbation, and mean absolute error of geological storage under small perturbation.

[0041] The goodness of fit of the perturbation is: ;

[0042] The error in the sealing process under minor disturbances is:

[0043] ;

[0044] The mean absolute error of geological reserves under minor disturbances is:

[0045] ;

[0046] In the formula, CCSR is the goodness of fit to the perturbation, CCSME is the error in the sealing process under small perturbations, CCSMA is the mean absolute error of geological sealing under small perturbations, N is the number of samples in the training dataset, i is the sample set number, j is the sample number in a certain sample set, and N j Let be the total number of samples in the i-th sample set. This represents the predicted geological carbon dioxide sequestration amount for the j-th sample in the i-th sample set at time t. The numerical simulation results of the geological carbon dioxide sequestration at time t for the j-th sample in the i-th sample set are as follows. This represents the arithmetic mean of the carbon dioxide geological sequestration results for the j-th sample in the i-th sample set over the carbon dioxide injection time period, where num is the total number of samples, Total is the carbon dioxide injection time, and t is the single time step of the carbon dioxide injection time. Let be the distribution function corresponding to the small perturbation data in the i-th sample set. The numerical simulation results of the geological carbon dioxide sequestration at time Total for the j-th sample in the i-th sample set are shown. This represents the predicted geological carbon dioxide sequestration at time Total for the j-th sample in the i-th sample set.

[0047] Optionally, based on the geological data and production dynamics data of the oil and gas reservoir to be predicted, a numerical simulation model of carbon dioxide sequestration in the oil and gas reservoir to be predicted is constructed, including:

[0048] Based on the geological data of the oil and gas reservoir to be predicted, a geological model of the oil and gas reservoir to be predicted is constructed using geological modeling software;

[0049] The geological model of the oil and gas reservoir to be predicted is imported into the reservoir numerical simulation software. The production dynamics of the oil and gas reservoir to be predicted are historically fitted using production dynamic data to obtain the numerical simulation model for the development of the oil and gas reservoir to be predicted.

[0050] Numerical simulation of carbon dioxide sequestration in the predicted oil and gas reservoir was conducted based on the numerical simulation model of the predicted oil and gas reservoir development, resulting in the numerical simulation model of carbon dioxide sequestration in the predicted oil and gas reservoir.

[0051] Optionally, the number of samples in a single sampling is 5 to 10% of the number of samples in a double sampling.

[0052] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for predicting carbon dioxide geological reserves.

[0053] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for predicting the geological reserves of carbon dioxide.

[0054] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0055] This application provides a method, equipment, and medium for predicting carbon dioxide geological reserves. A numerical simulation model for carbon dioxide reserves is established using geological data and production dynamics data. Different injection parameter combinations are formed through sampling methods. Numerical simulations are performed based on these different injection parameter combinations to obtain the cumulative carbon dioxide geological reserves for the entire region. Then, a multi-feature embedded carbon dioxide geological reserves prediction model is trained using the geological data, production dynamics data, and the cumulative carbon dioxide geological reserves for the entire region. Finally, transfer learning is used to predict the cumulative carbon dioxide geological reserves for the entire region of the oil and gas reservoir to be predicted.

[0056] The carbon dioxide geological storage prediction model trained using the multi-feature embedding method described in this application can efficiently, quickly, and accurately predict the cumulative carbon dioxide geological storage of other oil and gas reservoirs across the entire region through transfer learning. For a given oil and gas reservoir, predicting carbon dioxide geological storage using existing methods requires rebuilding a numerical simulation model and performing numerical simulations, which is not only data-intensive but also time-consuming and labor-intensive. Furthermore, when the target oil and gas reservoir or injection parameters change, numerical simulations must be repeated, resulting in poor model portability. However, using the method provided in this application, only 5%–10% of new samples need to be added to the model trained on existing oil and gas reservoirs for retraining, achieving better prediction results. Moreover, the trained model can directly replace numerical simulations for carbon dioxide storage prediction, eliminating the need for further numerical simulations. Therefore, this application demonstrates superior generalization ability and significantly improved computational efficiency.

[0057] Based on the aforementioned method and system for predicting carbon dioxide geological reserves, this application extracts geological data and production dynamic data features of abandoned oil and gas reservoirs through multiple encoders such as geological encoders, production encoders, and trend encoders, and embeds them as feature embedding layers into the carbon dioxide geological reserves prediction model. This preserves the original information while giving it physical meaning. The hyperparameters in the multi-feature embedded carbon dioxide geological reserves prediction model are optimized through hyperparameter optimization methods, ensuring the model's prediction accuracy even when performing cumulative carbon dioxide geological reserves prediction tasks for different target oil reservoirs. Therefore, this application demonstrates strong robustness. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of a method for predicting the geological reserves of carbon dioxide in one embodiment of this application;

[0060] Figure 2 This is a schematic diagram of a method for predicting the geological reserves of carbon dioxide in one embodiment of this application.

[0061] Figure 3 This is a schematic diagram of the physical feature layer in one embodiment of this application;

[0062] Figure 4 This is a schematic diagram of a multi-feature embedding prediction layer in one embodiment of this application;

[0063] Figure 5This is a 45° intersection diagram of the predicted and actual values ​​of carbon dioxide geological sequestration in one embodiment of this application;

[0064] Figure 6 This is a schematic diagram of a carbon dioxide geological sequestration prediction device according to one embodiment of this application. Detailed Implementation

[0065] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] In one exemplary embodiment, such as Figure 1 As shown, a method for predicting carbon dioxide geological reserves is provided, including:

[0068] Step 101: Determine whether the oil and gas reservoir to be predicted and the control oil and gas reservoir are in the same block.

[0069] Step 102: If so, input the geological data, production dynamics data, and injection parameter combination scheme of the oil and gas reservoir to be predicted into the optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model to obtain the total cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir. The optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model takes the geological data, production dynamics data, and injection parameter combination scheme of the control oil and gas reservoir as input, and the total cumulative carbon dioxide geological sequestration of the control oil and gas reservoir as output. It uses an optimization algorithm to optimize the hyperparameters in the multi-feature embedded carbon dioxide geological sequestration prediction model and obtains it after multiple trainings.

[0070] Step 103: If not, obtain the geological data, production dynamic data and injection parameter range of the oil and gas reservoir to be predicted.

[0071] Step 104: Based on the geological data and production dynamic data of the oil and gas reservoir to be predicted, construct a numerical simulation model of carbon dioxide sequestration in the oil and gas reservoir to be predicted.

[0072] Step 105: Conduct a sampling within the range of carbon dioxide injection parameters for the oil and gas reservoir to be predicted to obtain multiple sets of carbon dioxide injection parameter combinations for the oil and gas reservoir to be predicted; the carbon dioxide injection parameters include, but are not limited to: carbon dioxide injection volume, carbon dioxide injection rate, carbon dioxide injection time, carbon dioxide injection pressure, size of the injected carbon dioxide slug, water / gas injection ratio, and well shut-in time.

[0073] Step 106: Input multiple sets of injection parameter combination schemes for the oil and gas reservoirs to be predicted into the numerical simulation model of carbon dioxide sequestration of the oil and gas reservoirs to be predicted, and obtain the total cumulative geological carbon dioxide sequestration of the whole area corresponding to different injection parameter combination schemes for the oil and gas reservoirs to be predicted.

[0074] Step 107: Using the geological data and production dynamic data of the oil and gas reservoir to be predicted, as well as the injection parameter combination scheme of the oil and gas reservoir to be predicted, as input, and the total cumulative carbon dioxide geological sequestration of the whole area corresponding to the injection parameter combination scheme of the oil and gas reservoir to be predicted as output, the optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model is transferred and trained to obtain the prediction model of carbon dioxide geological sequestration of the oil and gas reservoir to be predicted.

[0075] Step 108: Input the geological data, production dynamic data and injection parameter combination scheme of the oil and gas reservoir to be predicted into the prediction model of carbon dioxide geological sequestration of the oil and gas reservoir to be predicted, and obtain the cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir to be predicted.

[0076] The multi-feature embedding model for predicting carbon dioxide geological reserves consists of a sequentially connected physical feature layer and a multi-feature embedding prediction layer. For example... Figure 3 The physical feature layer includes: a geological data feature sequence extraction branch, a production dynamics data feature sequence extraction branch, a parameter feature sequence extraction branch, a trend feature sequence extraction branch, and a splicing layer; the output ends of the production dynamics data feature sequence extraction branch, the geological data feature sequence extraction branch, the parameter feature sequence extraction branch, and the trend feature sequence extraction branch are all connected to the input end of the splicing layer.

[0077] The geological data feature sequence extraction branch includes a geological encoder; the input of the geological data feature sequence extraction branch is the geological data of the oil and gas reservoir to be predicted.

[0078] The feature sequence extraction branch for production dynamics data includes a production encoder. The input to the feature sequence extraction branch for production dynamics data is the production dynamics data of the oil and gas reservoir to be predicted.

[0079] The parametric feature sequence extraction branch includes a parallel value encoder and a position encoder. The input to the parametric feature sequence extraction branch is the geological data and production dynamics data of the oil and gas reservoir to be predicted.

[0080] The trend feature sequence extraction branch includes a trend encoder. The input to this branch is the geological data and production dynamics data of the oil and gas reservoir to be predicted. The production dynamics data feature sequence extraction branch and the geological data feature sequence extraction branch extract the overall features of all parameters in the production dynamics data and geological data, respectively. The parameter feature sequence extraction branch extracts the individual features of any parameter in the production dynamics data and geological data, and the trend feature sequence extraction branch extracts the trend features of any parameter in the production dynamics data and geological data. The stitching layer is formed by stitching the production dynamics data feature sequence extraction branch, geological data feature sequence extraction branch, parameter feature sequence extraction branch, and trend feature sequence extraction branch together in a horizontal stitching manner. The output of the stitching layer is connected to the multi-feature embedding prediction layer. The horizontal stitching method aligns the heads of the production dynamics data feature sequence extraction branch, geological data feature sequence extraction branch, parameter feature sequence extraction branch, and trend feature sequence extraction branch, and then stitches the aligned extraction branches sequentially from left to right to obtain the stitching layer.

[0081] like Figure 4 The multi-feature embedding prediction layer includes: a branch at time t, a branch at time t+1, a first fully connected layer, and an activation function connected in sequence.

[0082] The input terminals of the branch at time t and the branch at time t+1 are both connected to the output terminal of the physical feature layer.

[0083] The input terminals of the branch at time t and time t+1 are also used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted.

[0084] The branch at time t includes: a first multi-head attention layer, a first residual layer, a first regularization layer, a second fully connected layer, a second residual layer, and a second regularization layer connected in sequence.

[0085] The branch at time t+1 includes: the second multi-head attention layer, the third residual layer, the third regularization layer, the third fully connected layer, the fourth residual layer, and the fourth regularization layer connected in sequence.

[0086] The input terminals of the first multi-head attention layer and the first regularization layer are used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted and the data at time t in the physical feature sequence. The physical feature sequence is obtained by splicing the production dynamic data feature sequence, geological data feature sequence, parameter feature sequence, and trend feature sequence.

[0087] The output of the first regularization layer is also connected to the input of the second regularization layer.

[0088] The output of the second regularization layer is also connected to the input of the third regularization layer.

[0089] The input end of the second multi-head attention layer is used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted and the data at time t+1 in the physical feature sequence.

[0090] The output of the third regularization layer is also connected to the input of the fourth regularization layer.

[0091] The output of the fourth regularization layer is also connected to the input of the first fully connected layer.

[0092] Before step 101, the following is also included:

[0093] Obtain geological data, production dynamics data, and injection parameter ranges for the reference oil and gas reservoir.

[0094] Based on the geological and production dynamic data of the reference oil and gas reservoir, a numerical simulation model of carbon dioxide sequestration for the reference oil and gas reservoir was constructed.

[0095] Secondary sampling was conducted within the range of carbon dioxide injection parameter values ​​to obtain multiple sets of carbon dioxide injection parameter combinations for reference oil and gas reservoirs. The number of samples in the first sampling was 5-10% of the number of samples in the secondary sampling.

[0096] Multiple sets of control oil and gas reservoir injection parameter combinations were input into the control oil and gas reservoir carbon dioxide sequestration numerical simulation model to obtain the total cumulative carbon dioxide geological sequestration in the region corresponding to different control oil and gas reservoir injection parameter combinations.

[0097] Using the control oil and gas reservoir injection parameter combination scheme as input, and the total cumulative carbon dioxide geological storage in the whole area corresponding to different control oil and gas reservoir injection parameter combination schemes as output, a small sample dataset is constructed.

[0098] The small sample dataset is augmented using small perturbation data to obtain an augmented sample dataset. Augmentation includes, but is not limited to, small perturbation data conforming to Gaussian, uniform, or Laplace distributions. The small sample dataset and the augmented sample dataset are then vertically concatenated to obtain the training dataset. The vertical concatenation method aligns the left sides of the augmented sample dataset and the small sample dataset, and the aligned datasets are stacked sequentially from top to bottom to form the training dataset. Using geological and production dynamic data of the control oil and gas reservoir, as well as the injection parameter combination scheme of the control oil and gas reservoir in the training dataset, as input, and the total cumulative carbon dioxide geological reserves in the training dataset as output, the multi-feature embedding carbon dioxide geological reserves prediction model is trained multiple times. The hyperparameters in the multi-feature embedding carbon dioxide geological reserves prediction model are optimized using an optimization algorithm. The evaluation index of the control oil and gas reservoir carbon dioxide geological reserves prediction model obtained in each training iteration is determined, and the optimal control oil and gas reservoir carbon dioxide geological reserves prediction model is obtained based on the evaluation index. Evaluation metrics include, but are not limited to, goodness of fit to disturbances, error in the sealing process under minor disturbances, and average absolute error of geological sealing volume under minor disturbances.

[0099] The goodness of fit of the perturbation is: .

[0100] The error in the sealing process under minor disturbances is:

[0101] .

[0102] The mean absolute error of geological reserves under minor disturbances is:

[0103] .

[0104] In the formula, CCSR is the goodness of fit to the perturbation, CCSME is the error in the sealing process under small perturbations, CCSMA is the mean absolute error of geological sealing under small perturbations, N is the number of samples in the training dataset, i is the sample set number, j is the sample number in a certain sample set, and N j Let be the total number of samples in the i-th sample set. This represents the predicted geological carbon dioxide sequestration amount for the j-th sample in the i-th sample set at time t. The numerical simulation results of the geological carbon dioxide sequestration at time t for the j-th sample in the i-th sample set are as follows. This represents the arithmetic mean of the carbon dioxide geological sequestration results for the j-th sample in the i-th sample set over the carbon dioxide injection time period, where num is the total number of samples, Total is the carbon dioxide injection time, and t is the single time step of the carbon dioxide injection time. Let be the distribution function corresponding to the small perturbation data in the i-th sample set. The numerical simulation results of the geological carbon dioxide sequestration at time Total for the j-th sample in the i-th sample set are shown. This represents the predicted geological carbon dioxide sequestration at time Total for the j-th sample in the i-th sample set.

[0105] Based on the geological and production dynamic data of the oil and gas reservoir to be predicted, a numerical simulation model of carbon dioxide sequestration for the reservoir is constructed. This includes: constructing a geological model of the reservoir using geological modeling software; importing the geological model into reservoir numerical simulation software; performing historical fitting of the reservoir's production dynamics using production dynamic data to obtain a numerical simulation model for reservoir development; and conducting a numerical simulation of carbon dioxide sequestration based on this development model to obtain the final carbon dioxide sequestration numerical simulation model.

[0106] Based on the range of carbon dioxide injection parameters for the oil and gas reservoir to be predicted, a small-sample dataset is constructed using sampling methods to train a multi-feature embedding carbon dioxide geological storage prediction model. The small-sample dataset refers to a dataset that uses geological and production dynamic data of the oil and gas reservoir to be predicted as its basis, and a combination of carbon dioxide injection parameters obtained through sampling methods based on the range of injection parameter values ​​as its input data. The output data is the total cumulative carbon dioxide geological storage volume for the entire region obtained through numerical simulation using the numerical simulation model of carbon dioxide storage in the oil and gas reservoir to be predicted. The size of the small-sample dataset varies depending on the range of injection parameter values.

[0107] Furthermore, the steps for determining a small sample dataset include:

[0108] S12. Within the range of carbon dioxide injection parameters, a sampling method is used to perform one sampling, and each sampling forms a set of carbon dioxide injection parameter combinations. Sampling methods include, but are not limited to, importance sampling, Latin hypercube sampling, Markov chain Monte Carlo sampling, random sampling, and stratified sampling. The injection parameters include, but are not limited to: carbon dioxide injection volume, carbon dioxide injection rate, carbon dioxide injection time, carbon dioxide injection pressure, size of the injected carbon dioxide slug, water / gas injection ratio, and well shut-in time.

[0109] S13. Based on the set number of samplings, form a carbon dioxide injection parameter sample set, call the carbon dioxide sequestration numerical simulation model of the oil and gas reservoir to be predicted to perform batch numerical simulation, set the carbon dioxide injection parameters according to the carbon dioxide injection parameter sample set to perform simulation, and obtain the cumulative carbon dioxide geological sequestration in the whole area.

[0110] S14. Construct a small sample dataset, DataSet-1, using each set of carbon dioxide injection parameters, geological data, production dynamics data, and the total cumulative carbon dioxide geological sequestration volume obtained from numerical simulation to train a multi-feature embedded carbon dioxide geological sequestration volume prediction model.

[0111] Data augmentation methods are used to expand the constructed small sample dataset to obtain a sample dataset for training the multi-feature embedding carbon dioxide geological reserve prediction model. Specifically, the small sample dataset is expanded as follows: First, a set of small perturbation data is generated. Then, this set of small perturbation data is added to the input carbon dioxide injection parameters and geological parameters. Furthermore, small perturbation data is added to each set of carbon dioxide injection parameters in S12. The order of magnitude of the small perturbation data is much smaller than two orders of magnitude of the original data.

[0112] Data augmentation methods include, but are not limited to, data with small perturbations conforming to Gaussian, uniform, or Laplace distributions. The mathematical reasoning formulas can all be expressed as:

[0113] (10)

[0114] in, This is the data after adding a small perturbation. It is a small sample of data constructed in S14. These are three different methods of representation: when i=1, the data is represented by a small perturbation following a Gaussian distribution; when i=2, the data is represented by a small perturbation following a uniform distribution; and when i=3, the data is represented by a small perturbation following a Laplace distribution. In other words, after generating the small perturbation data, it is added to the small sample data to obtain a new sample set.

[0115] S21. Add small perturbation data that conforms to a Gaussian distribution. It follows a Gaussian distribution. Randomly perturbed data, i.e., data with a mean of 0 and a variance of 0. The random value. The probability density function is as follows:

[0116] (11)

[0117] After generating random small perturbation data that follows a Gaussian distribution, it is added to the small sample data constructed in S13 to generate a sample set DataSet-2.

[0118] S22. Add small perturbation data that conforms to a uniform distribution. It follows a uniform distribution Random, small perturbation data, i.e., values ​​ranging from Random values ​​between. The probability density function is as follows: where the parameter 'a' will be automatically adjusted according to the specific data size rather than a fixed value. For example, if in a set of carbon dioxide injection parameters, the carbon dioxide injection pressure is on the order of 10... 1 Then the parameter a can be -0.05 to 0.05.

[0119] (12)

[0120] After generating random small perturbation data that follows a uniform distribution, it is added to the small sample data constructed in S13 to generate a binary sample set DataSet-3.

[0121] S23. Add small perturbation data that conforms to a Laplace distribution. It is random, small-perturbation data that follows a Laplace distribution. The probability density function is as follows:

[0122] (13)

[0123] In the formula, μ is the mean parameter of the Laplace distribution, and β is the scale parameter of the Laplace distribution. After generating random small perturbation data, it is added to the small sample data constructed in S13 to generate a three-class sample set DataSet-4.

[0124] S24. Combining the small sample dataset DataSet-1 constructed in S14, the first-class sample dataset DataSet-2 constructed in S21 with added small perturbation data conforming to a Gaussian distribution, the second-class sample dataset DataSet-3 constructed in S22 with added small perturbation data conforming to a uniform distribution, and the third-class sample dataset DataSet-4 constructed in S23 with added small perturbation data conforming to a Laplace distribution, DataSet-1, DataSet-2, DataSet-3, and DataSet-4 are merged in a unified manner along the vertical concatenation of the sample sets to construct a sample dataset for training the multi-feature embedding carbon dioxide geological storage prediction model.

[0125] S31. Establish a multi-feature embedded carbon dioxide geological storage prediction model. The model consists of a multi-head attention layer, a residual layer, a fully connected layer, and a regularization layer. The settings of each layer are based on the prediction task of the cumulative carbon dioxide geological storage of the oil and gas reservoir to be predicted in this application.

[0126] S311. Specifically, the multi-head attention layer enables the prediction model to simultaneously focus on information from different locations in the multi-dimensional input data, used to describe the weighted average of the multi-dimensional input data.

[0127] (14)

[0128] In the formula, Query is the query matrix, Key is the keyword matrix, Value is the value matrix, and head is the head matrix. i Let h be the number of attention heads, i.e., h are the i-th attention heads. `Attention(*)` is a general representation of a single attention head, containing three parameters: Query, Key, and Value. `softmax(*)` is a function that, after processing, results in a sequence data sum of 1. O Let W be the weight matrix. Query W Key and W Value For learnable parameters, Let d be the input sequence, d be the dimension of the weight matrix, and T be the length of the sequence. Multihead(*) is a multi-head attention layer.

[0129] Specifically, the input sequence With learnable parameters W Query W Key and W Value After data mapping, we obtain Query, Key, and Value. We calculate the attention of a single attention head using softmax(*). We then concatenate the h attention heads using Concat and multiply them with a matrix to calculate the multi-head attention of the multi-head attention.

[0130] S312. Specifically, the residual layer mainly adds a structure that goes directly from the input to the output between layers, and adds this structure to the output between layers.

[0131] Between layers, including but not limited to multi-head attention layers and regularization layers, regularization layers and fully connected layers, and multi-head attention layers and fully connected layers.

[0132] S313. Specifically, the fully connected layer flattens the multidimensional data after it has been processed by other layers, that is, it transforms it into a long vector.

[0133] S314. Specifically, the regularization layer can transform each feature of the multidimensional input data into sequence data with a mean of 0 and a variance of 1.

[0134] S315. Specifically, an encoder is constructed based on the multi-head attention layer of S311, the residual layer of S312, and the regularization layer of S313. The encoders are connected to each other using the residual layer of S312.

[0135] S316. Specifically, a decoder is constructed based on the multi-head attention layer of S311, the residual layer of S312, and the regularization layer of S313. Decoders are connected to each other using the residual layer of S312.

[0136] S317. Specifically, based on the encoder in S315, the decoder in S316, the softmax activation function, and the linear output layer, the multi-feature embedded carbon dioxide geological storage prediction model proposed in this application is constructed to meet the prediction task of the cumulative carbon dioxide geological storage of the oil and gas reservoir to be predicted in the whole area proposed in this application.

[0137] S32. Define the input and output data of the multi-feature embedded carbon dioxide geological sequestration prediction model. Transform the geological data and production dynamic data of the oil and gas reservoir to be predicted into a physical feature layer embedded prediction model. Use the physical feature layer and injection parameters as input data and the cumulative carbon dioxide geological sequestration of the whole area as output data.

[0138] The physical feature layer is embedded into a multi-feature embedding model for predicting carbon dioxide geological reserves, specifically including:

[0139] S321. Read the sample dataset constructed in S24 and extract the geological data and production dynamic data of the oil and gas reservoirs to be predicted from the sample dataset.

[0140] S322. A geological encoder is used to encode the geological data of the oil and gas reservoir to be predicted, extract the geological data features, and obtain the geological data feature sequence.

[0141] S323. A production encoder is used to encode the production dynamics data of the oil and gas reservoir to be predicted, extract the features of the production dynamics data, and obtain the feature sequence of the production dynamics data.

[0142] S324. A trend encoder is used to encode the trend features in the geological data and production dynamic data of the oil and gas reservoir to be predicted, and to extract the trend feature sequence.

[0143] S325. Value encoders and position encoders are used to encode the sequence features in the geological data and production dynamic data of the oil and gas reservoir to be predicted, and to extract the parameter feature sequence.

[0144] Specifically, the encoding methods corresponding to geological encoders, production encoders, trend encoders, value encoders, and position encoders include, but are not limited to: sequence number encoding, average encoding, Helmert encoding, binary encoding, one-hot encoding, tag encoding, and frequency encoding.

[0145] S326. Construct a physical feature layer based on geological data characteristics, production dynamic data characteristics, trend characteristics, and sequence characteristics. Use this physical feature layer and injection parameters as inputs to a multi-feature embedded carbon dioxide geological sequestration prediction model.

[0146] S33. Based on the defined input and output data, train a multi-feature embedding carbon dioxide geological sequestration prediction model, and optimize the hyperparameters in the model using an optimization algorithm. Hyperparameter optimization methods include, but are not limited to, covariance matrix adaptive evolution algorithm, grid search, random search, Bayesian optimization, gradient descent, etc.

[0147] A multi-feature embedding model for predicting carbon dioxide geological reserves was trained using S33, with a preset training iteration count of 1000. The model was then optimized based on evaluation metrics, retaining the best-performing multi-feature embedding model and its corresponding hyperparameters. Evaluation metrics included, but were not limited to, CCSR, CCSME, and CCSMA.

[0148] The specific calculation methods for CCSR, CCSME, and CCSMA are as follows:

[0149] (36)

[0150] (37)

[0151] (38)

[0152] Specifically, the overall predictive performance of the CCSR observation multi-feature embedding carbon dioxide geological sequestration prediction model, whether the CCSME observation multi-feature embedding carbon dioxide geological sequestration prediction model converges, and whether the CCSMA observation multi-feature embedding carbon dioxide geological sequestration prediction model stops training.

[0153] The system calls upon the best-performing multi-feature embedded carbon dioxide geological sequestration prediction model retained by S33, inputting geological data, production dynamics data, and injection parameter combinations for the oil and gas reservoir to be predicted. The output is the cumulative carbon dioxide geological sequestration for the entire reservoir area. This includes:

[0154] S41. Determine whether the oil and gas reservoir to be predicted and the control oil and gas reservoir are in the same block.

[0155] S42. If yes, input the geological data, production dynamics data and injection parameter combination scheme of the oil and gas reservoir to be predicted, and output the total cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir to be predicted.

[0156] S43. If not, obtain the geological data, production dynamic data, and injection parameter value ranges of the oil and gas reservoir to be predicted. Construct a geological model and a numerical simulation model of carbon dioxide sequestration for the oil and gas reservoir to be predicted. Using the importance sampling method, set 5% of the original sampling frequency in S12 to perform secondary sampling of the carbon dioxide injection parameters, and perform numerical simulation for each set of injection parameters for the oil and gas reservoir to be predicted to obtain the total geological carbon dioxide sequestration of the entire area. Add small data perturbations to the carbon dioxide injection parameters, geological parameters, and total geological carbon dioxide sequestration of the entire area to construct a sample set.

[0157] Using the reservoir geological data and production dynamic data to be optimized as the physical feature layer embedding prediction model, the physical feature layer and injection parameters are used as input data, and the cumulative carbon dioxide geological reserves of the entire area are used as output data. A multi-feature embedding carbon dioxide geological reserves prediction model is then trained and applied to predict the cumulative carbon dioxide geological reserves of the entire oil and gas reservoir.

[0158] The following specific examples will be used to verify this embodiment, such as... Figure 2 The method in this embodiment includes:

[0159] Step 1: Based on the geological and production dynamic data of the oil and gas reservoir to be predicted, establish a geological model corresponding to the actual reservoir, and construct a numerical simulation model of carbon dioxide sequestration in the oil and gas reservoir to be predicted using numerical simulation software. The parameters used are shown in Table 1.

[0160] Table 1. Basic Parameters of the Numerical Simulation Model for Carbon Dioxide Geological Storage

[0161]

[0162] Step 2: Based on the numerical simulation model of carbon dioxide sequestration in the oil and gas reservoir to be predicted, construct a small sample dataset for training a multi-feature embedded carbon dioxide geological sequestration prediction model. The range of injection parameter values ​​is shown in Table 2, and some injection parameter combination schemes formed by importance sampling are shown in Table 3. A small sample dataset refers to a dataset with fewer than 200 samples per sampling.

[0163] Table 2. Range of Carbon Dioxide Injection Parameters

[0164]

[0165] Table 3 Examples of some carbon dioxide injection parameter combinations

[0166]

[0167] Step 3: Use data augmentation methods to expand the constructed small sample dataset to obtain an expanded sample dataset for training the multi-feature embedding carbon dioxide geological sequestration prediction model. Partial data of the expanded sample dataset is shown in Table 4.

[0168] Table 4. Partial Data Illustration of the Sample Dataset

[0169]

[0170] Step 4: Establish a multi-feature embedded carbon dioxide geological sequestration prediction model: Geological data and production dynamic data of the oil and gas reservoir to be predicted are used as the physical feature layer embedding prediction model. The combination scheme of physical feature layer and carbon dioxide injection parameters is used as input data, and the cumulative carbon dioxide geological sequestration of the entire region is used as output data. The multi-feature embedded carbon dioxide geological sequestration prediction model is trained, and the hyperparameters in the multi-feature embedded carbon dioxide geological sequestration prediction model are optimized using a covariance matrix adaptive evolution strategy. The multi-feature embedded carbon dioxide geological sequestration prediction model with the best prediction performance is selected to predict the cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir. A 45° intersection plot of the predicted and actual carbon dioxide geological sequestration values ​​is shown below. Figure 5 As shown, the goodness of fit for the perturbation is 96.86%.

[0171] Step 5: Use the best-performing multi-feature embedded carbon dioxide geological sequestration prediction model. Input the geological data and production dynamics data of the oil and gas reservoir to be predicted. Output the cumulative carbon dioxide geological sequestration of the entire reservoir area. The cumulative carbon dioxide geological sequestration of the entire area is 3098.53 × 10⁻⁶. 4 m 3 .

[0172] like Figure 6 In one exemplary embodiment, a carbon dioxide geological sequestration prediction device is provided, comprising:

[0173] Importance Sampling Module 202: Used to construct different combinations of injection parameters based on the value range of each injection parameter determined by the parameter loading module.

[0174] Numerical simulation module 303: Used to conduct reservoir numerical simulation in the oil and gas reservoir to be predicted. It uses the parameter combination constructed by module 202 to perform numerical simulation and obtain the cumulative geological carbon dioxide sequestration of the whole area.

[0175] Data expansion module 404: Based on the data expansion process, it expands the total geological carbon dioxide sequestration of the whole region obtained by the numerical simulation module, calculates the data under three different small perturbation conditions, and merges them with the results obtained by module 303 into a sample set.

[0176] Physical feature layer embedding module 505: used to extract geological data features, production dynamic data features, trend features and sequence features of the geological data and production dynamic data of the oil and gas reservoirs to be predicted in the sample set, and to construct a physical feature layer with the above features as input to the multi-feature embedding carbon dioxide geological sequestration prediction model.

[0177] Model training module 606: Used to train a multi-feature embedding carbon dioxide geological sequestration prediction model. Input data includes physical feature layers and injection parameters embedded from geological data and production dynamics data. Output data is the cumulative carbon dioxide geological sequestration of the entire region. Simultaneously, an adaptive evolution strategy using the covariance matrix is ​​employed to optimize the hyperparameters of the multi-feature embedding carbon dioxide geological sequestration prediction model, retaining the model with the best prediction performance.

[0178] Prediction Module 707: Used to predict the cumulative geological carbon dioxide sequestration of the entire oil and gas reservoir. It calls the carbon dioxide geological sequestration prediction model with the best prediction effect through multi-feature embedding, inputs geological data, production dynamic data and injection parameter combination scheme, and obtains the predicted data of the cumulative geological carbon dioxide sequestration of the entire oil and gas reservoir.

[0179] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting carbon dioxide geological reserves.

[0180] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0181] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0182] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0183] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0184] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0185] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0186] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the geological reserves of carbon dioxide, characterized in that, include: Determine whether the oil and gas reservoir to be predicted and the control oil and gas reservoir belong to the same block; If so, the geological data, production dynamics data, and injection parameter combination scheme of the oil and gas reservoir to be predicted are input into the optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model to obtain the total cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir to be predicted. The optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model is obtained by using the geological data, production dynamics data, and injection parameter combination scheme of the control oil and gas reservoir as input and the total cumulative carbon dioxide geological sequestration of the control oil and gas reservoir as output. The hyperparameters in the multi-feature embedded carbon dioxide geological sequestration prediction model are optimized by using an optimization algorithm, and the multi-feature embedded carbon dioxide geological sequestration prediction model is obtained after multiple trainings. If not, obtain the geological data, production dynamics data, and injection parameter range of the oil and gas reservoir to be predicted; Based on the geological data and production dynamic data of the oil and gas reservoir to be predicted, a numerical simulation model of carbon dioxide sequestration in the oil and gas reservoir to be predicted is constructed. Based on the geological and production dynamic data of the oil and gas reservoir to be predicted, a numerical simulation model of carbon dioxide sequestration in the reservoir is constructed, including: Based on the geological data of the oil and gas reservoir to be predicted, a geological model of the oil and gas reservoir to be predicted is constructed using geological modeling software; The geological model of the oil and gas reservoir to be predicted is imported into the reservoir numerical simulation software. The production dynamics of the oil and gas reservoir to be predicted are historically fitted using production dynamic data to obtain the numerical simulation model for the development of the oil and gas reservoir to be predicted. Numerical simulation of carbon dioxide sequestration in the predicted oil and gas reservoir is carried out based on the numerical simulation model of the development of the predicted oil and gas reservoir, and the numerical simulation model of carbon dioxide sequestration in the predicted oil and gas reservoir is obtained. A sampling is conducted within the range of carbon dioxide injection parameters for the oil and gas reservoir to be predicted to obtain multiple sets of carbon dioxide injection parameter combinations for the oil and gas reservoir to be predicted; the carbon dioxide injection parameters include, but are not limited to: carbon dioxide injection volume, carbon dioxide injection rate, carbon dioxide injection time, carbon dioxide injection pressure, size of the injected carbon dioxide slug, water / gas injection ratio, and well shut-in time. Multiple sets of injection parameter combinations for the oil and gas reservoirs to be predicted are input into the numerical simulation model of carbon dioxide sequestration of the oil and gas reservoirs to be predicted, and the total cumulative geological carbon dioxide sequestration of the whole area corresponding to different injection parameter combinations for the oil and gas reservoirs to be predicted is obtained. Using the geological data and production dynamic data of the oil and gas reservoir to be predicted, as well as the injection parameter combination scheme of the oil and gas reservoir to be predicted, as input, and the total cumulative carbon dioxide geological sequestration of the whole area corresponding to the injection parameter combination scheme of the oil and gas reservoir to be predicted as output, the optimal control oil and gas reservoir carbon dioxide geological sequestration prediction model is transferred and trained to obtain the prediction model of carbon dioxide geological sequestration of the oil and gas reservoir to be predicted. The geological data, production dynamic data, and injection parameter combination scheme of the oil and gas reservoir to be predicted are input into the prediction model of carbon dioxide geological sequestration of the oil and gas reservoir to be predicted, so as to obtain the cumulative carbon dioxide geological sequestration of the entire oil and gas reservoir.

2. The method for predicting carbon dioxide geological reserves according to claim 1, characterized in that, The multi-feature embedded carbon dioxide geological sequestration prediction model includes a physical feature layer and a multi-feature embedded prediction layer connected in sequence.

3. The method for predicting carbon dioxide geological reserves according to claim 2, characterized in that, The physical feature layer includes: a geological data feature sequence extraction branch, a production dynamics data feature sequence extraction branch, a parameter feature sequence extraction branch, a trend feature sequence extraction branch, and a splicing layer; the output ends of the production dynamics data feature sequence extraction branch, the geological data feature sequence extraction branch, the parameter feature sequence extraction branch, and the trend feature sequence extraction branch are all connected to the input end of the splicing layer; The geological data feature sequence extraction branch includes a geological encoder; the input of the geological data feature sequence extraction branch is the geological data of the oil and gas reservoir to be predicted. The feature sequence extraction branch of the production dynamic data includes a production encoder; the input of the feature sequence extraction branch of the production dynamic data is the production dynamic data of the oil and gas reservoir to be predicted. The parameter feature sequence extraction branch includes a parallel value encoder and a position encoder; the input of the parameter feature sequence extraction branch is the geological data and production dynamic data of the oil and gas reservoir to be predicted. The trend feature sequence extraction branch includes a trend encoder; the input to the trend feature sequence extraction branch is the geological data and production dynamic data of the oil and gas reservoir to be predicted. The production dynamic data feature sequence extraction branch and the geological data feature sequence extraction branch extract the overall features of all parameters in the production dynamic data and geological data, respectively. The parameter feature sequence extraction branch extracts the individual features of any parameter in the production dynamic data and geological data, and the trend feature sequence extraction branch extracts the trend features of any parameter in the production dynamic data and geological data. The splicing layer is formed by splicing the production dynamic data feature sequence extraction branch, the geological data feature sequence extraction branch, the parameter feature sequence extraction branch, and the trend feature sequence extraction branch in a horizontal splicing manner. The output end of the splicing layer is connected to the multi-feature embedding prediction layer. The horizontal splicing method aligns the heads of the production dynamic data feature sequence extraction branch, the geological data feature sequence extraction branch, the parameter feature sequence extraction branch, and the trend feature sequence extraction branch, and splices the aligned extraction branches in a left-to-right order to obtain the splicing layer.

4. The method for predicting carbon dioxide geological reserves according to claim 2, characterized in that, The multi-feature embedding prediction layer includes: The branches at time t, time t+1, the first fully connected layer, and the activation function are connected in sequence. The input terminals of the branch at time t and the branch at time t+1 are both connected to the output terminal of the physical feature layer. The input terminals of the branch at time t and the branch at time t+1 are also used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted.

5. The method for predicting carbon dioxide geological reserves according to claim 4, characterized in that, The branch at time t includes: a first multi-head attention layer, a first residual layer, a first regularization layer, a second fully connected layer, a second residual layer, and a second regularization layer connected in sequence; The branch at time t+1 includes: a second multi-head attention layer, a third residual layer, a third regularization layer, a third fully connected layer, a fourth residual layer, and a fourth regularization layer connected in sequence; The input terminals of the first multi-head attention layer and the first regularization layer are used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted and the data at time t in the physical feature sequence; the physical feature sequence is obtained by splicing the production dynamic data feature sequence, geological data feature sequence, parameter feature sequence and trend feature sequence; The output of the first regularization layer is also connected to the input of the second regularization layer; The output of the second regularization layer is also connected to the input of the third regularization layer; The input end of the second multi-head attention layer is used to input the injection parameter combination scheme of the oil and gas reservoir to be predicted and the data at time t+1 in the physical feature sequence; The output of the third regularization layer is also connected to the input of the fourth regularization layer; The output of the fourth regularization layer is also connected to the input of the first fully connected layer.

6. The method for predicting carbon dioxide geological reserves according to claim 1, characterized in that, Before determining whether the predicted oil and gas reservoir and the control oil and gas reservoir belong to the same block, the following steps are also included: Obtain geological data, production dynamics data, and injection parameter ranges for the reference oil and gas reservoir; Based on the geological and production dynamic data of the reference oil and gas reservoir, a numerical simulation model of carbon dioxide sequestration of the reference oil and gas reservoir was constructed. Secondary sampling was conducted within the range of carbon dioxide injection parameters to obtain multiple sets of carbon dioxide injection parameter combinations for reference oil and gas reservoirs. Multiple sets of injection parameter combinations for the control oil and gas reservoirs were input into the numerical simulation model of carbon dioxide sequestration for the control oil and gas reservoirs to obtain the total cumulative geological carbon dioxide sequestration in the whole area corresponding to different injection parameter combinations for the control oil and gas reservoirs. A small sample dataset was constructed using the control oil and gas reservoir injection parameter combination scheme as input and the total cumulative carbon dioxide geological sequestration in the whole area corresponding to different control oil and gas reservoir injection parameter combination schemes as output. The small sample dataset is augmented using small perturbation data to obtain an augmented sample dataset. The augmentation process includes, but is not limited to, small perturbation data conforming to Gaussian, uniform, or Laplace distributions. The small sample dataset and the augmented sample dataset are then concatenated vertically to obtain a training dataset. The vertical concatenation method aligns the left sides of the augmented sample dataset and the small sample dataset, and the aligned datasets are stacked sequentially from top to bottom to obtain the training dataset. Using geological and production dynamic data of the reference oil and gas reservoir, as well as the injection parameter combination scheme of the reference oil and gas reservoir in the training dataset as input, and the total cumulative carbon dioxide geological sequestration of the entire region in the training dataset as output, the carbon dioxide geological sequestration prediction model with multi-feature embedding is trained multiple times and the hyperparameters in the multi-feature embedding carbon dioxide geological sequestration prediction model are optimized using optimization algorithms. The evaluation index of the carbon dioxide geological sequestration prediction model of the reference oil and gas reservoir obtained in each training is determined, and the optimal carbon dioxide geological sequestration prediction model of the reference oil and gas reservoir is obtained based on the evaluation index. Evaluation metrics include, but are not limited to, goodness of fit to perturbation, error in the storage process under small perturbation, and average absolute error of geological storage volume under small perturbation. The goodness of fit of the perturbation is: The error in the sealing process under minor disturbances is: The mean absolute error of geological reserves under minor disturbances is: In the formula, CCSR is the goodness of fit to the perturbation, CCSME is the error in the sealing process under small perturbations, CCSMA is the mean absolute error of geological sealing under small perturbations, N is the number of samples in the training dataset, i is the sample set number, j is the sample number in a certain sample set, and N j Let be the total number of samples in the i-th sample set. This represents the predicted geological carbon dioxide sequestration amount for the j-th sample in the i-th sample set at time t. The numerical simulation results of the geological carbon dioxide sequestration at time t for the j-th sample in the i-th sample set are as follows. The numerical simulation results of carbon dioxide geological sequestration for the j-th sample in the i-th sample set are the arithmetic mean of the carbon dioxide injection time, where num is the total number of samples, Total is the carbon dioxide injection time, t is the single time step of the carbon dioxide injection time, and f(ξ) is the arithmetic mean of the carbon dioxide injection time. i Let be the distribution function corresponding to the small perturbation data in the i-th sample set. The numerical simulation results of the geological carbon dioxide sequestration at time Total for the j-th sample in the i-th sample set are shown. This represents the predicted geological carbon dioxide sequestration at time Total for the j-th sample in the i-th sample set.

7. The method for predicting carbon dioxide geological reserves according to claim 6, characterized in that, The number of samples in a single sampling is 5 to 10% of the number of samples in a double sampling.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for predicting the geological reserves of carbon dioxide according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for predicting the geological reserves of carbon dioxide as described in any one of claims 1-7.

Images