Carbon dioxide mixed gas throughput parameter dynamic optimization method based on multi-target interaction

Abstract

The invention belongs to the technical field of tight oil reservoir huff-puff development, and particularly discloses a carbon dioxide mixed gas huff-puff parameter dynamic optimization method based on multi-target interaction, which is used for solving the problem of how to supplement stratum energy and improve the crude oil recovery rate in the middle and later periods of current tight oil reservoir development. Comprising the steps of selecting a target block for design, and collecting a time sequence prediction sample set; the method comprises the steps that sample data is subjected to standardization processing, a data set is divided, a TFT neural network is adopted to construct a TFT prediction proxy model, and the TFT prediction proxy model is used for rapid prediction of the recovery ratio, the burying amount and the NPV; and determining an objective function, constructing a multi-objective optimization model, and establishing a CO2 mixed gas huff and puff effect dynamic optimization model based on an NSGA-III multi-objective optimization algorithm. According to the method, the production efficiency is effectively improved, the risk and uncertainty are reduced, and balance optimization among multiple targets is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of tight oil reservoir throughput development, and in particular relates to a method for dynamically optimizing carbon dioxide mixed gas throughput parameters based on multi-objective interaction. Background Art

[0002] Most of the existing tight oil development strategies have certain limitations. Among them, although the initial production of depletion production is high, the production capacity decline rate is generally high, the formation energy will be quickly depleted, resulting in low recovery rate and low economic benefits; due to the poor porosity and permeability of tight reservoirs, the fluidity of oil and gas is poor, making it difficult to achieve water injection development; although gas drive development is effective quickly, it is prone to early gas channeling, resulting in low gas utilization. Gas injection is a general strategy to solve the serious problem of gas channeling in conventional gas drive, and it is also an effective way to improve the recovery rate of crude oil in tight oil reservoirs.

[0003] By summarizing previous studies, CO2 mixed gas huff and puff is an efficient development method to improve the recovery of crude oil in tight oil reservoirs, and the recovery effect is higher than that of single gas huff and puff. However, the reasonable development technology policy boundary of CO2 mixed gas huff and puff to improve the recovery of crude oil in the process of tight oil development is unclear. Therefore, dynamic optimization of CO2 mixed gas huff and puff development factors is the key to the adjustment of huff and puff development in tight oil reservoirs. Summary of the invention

[0004] The purpose of the present invention is to provide a method for dynamically optimizing carbon dioxide mixed gas throughput parameters based on multi-objective interaction, which effectively solves the problem of how to supplement formation energy and improve crude oil recovery in the middle and late stages of current tight oil reservoir development.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a dynamic optimization method for carbon dioxide mixture gas throughput parameters based on multi-objective interaction, comprising the following steps: S1, selecting a target block for design, and collecting a time series prediction sample set; S2, standardizing the sample data, dividing the data set, using the TFT neural network to construct a TFT prediction agent model, and using the TFT prediction agent model to quickly predict the recovery rate, storage volume and NPV; S3, determining the objective function, constructing a multi-objective optimization model, and establishing a dynamic optimization model for the CO2 mixture gas throughput effect based on the NSGA-Ⅲ multi-objective optimization algorithm, ensuring that a reasonable and efficient working system is matched at different time steps, and realizing the dynamic optimization of the CO2 mixture gas throughput parameters of tight oil reservoirs.

[0006] Furthermore, in step S2, the sample data standardization process uses the minimum and maximum standardization method to normalize each feature:

[0007] ;

[0008] in, is the normalized value, is the actual value of the feature factor in the data set, is the maximum value of the characteristic factor in the original data, It is the minimum value of the characteristic factor in the original data.

[0009] The normalized raw data is divided into training set, validation set and test set according to the ratio of 6:2:2, which are used for training the TFT prediction agent model, tuning the hyperparameters of the TFT neural network and performance evaluation of the TFT prediction agent model respectively.

[0010] Furthermore, in step S2, components of the TFT neural network include input features, a variable selection network, a static covariate encoder, a sequence-to-sequence layer, a gated residual network, a multi-head attention mechanism, a temporal fusion decoder, and a quantile output.

[0011] Regarding input features: TFT neural network divides input variables into static variables and dynamic variables, among which static variables are divided into static continuous variables and static discrete variables, and dynamic variables include dynamic time-varying variables and dynamic time-invariant variables. Dynamic time-varying variables refer to variables whose characteristics change over time and cannot be inferred, including recovery rate, storage volume and net present value; dynamic time-invariant variables are variables applied to the TFT prediction agent model based on reasoning, including months and cumulative gas injection volume.

[0012] Furthermore, in step S2, the functions of the variable selection network include: selecting relevant input variables for each time step, and weakening or discarding irrelevant information; the static covariate encoder enables static features to be integrated into the TFT neural network, and adjusts the temporal dynamics by encoding the context vector; the sequence-to-sequence layer adopts two groups of LSTM networks, one of which is used to process past input features and encode past historical inputs; the other is used to process future input features; the gated residual network is a selection framework that uses a gating mechanism to selectively introduce relevant input features through optional jump connections and gated matrix filtering.

[0013] The TFT prediction agent model learns to capture long-term dependencies at different time steps through a multi-head attention mechanism, while focusing on different parts of the input sequence in parallel and increasing the model attention weight for specific features.

[0014] The temporal fusion decoder consists of two layers of gated residual networks and a multi-head attention mechanism layer; the first layer of the gated residual network combines the output information of the sequence-to-sequence layer with the information from the static features, and then performs a standard self-attention mechanism on the input information at all time steps through the multi-head attention mechanism layer. Finally, the output of the multi-head attention mechanism layer and the remaining connections are passed to another layer of the gated residual network for nonlinear processing.

[0015] Further, in step S2, the TFT neural network determines the prediction interval of the possible target value range of each prediction layer through quantile prediction, and generates the prediction interval using quantile regression to obtain information about the prediction certainty; the quantile prediction is generated by a linear transformation of the output of the time fusion decoder.

[0016] The TFT prediction agent model trains the TFT neural network by jointly minimizing the quantile loss of all quantile outputs:

[0017] ;

[0018] ;

[0019] in, is the total loss function, is the quantile loss function, For the The actual observed value at each time step, For the model time steps in quantile The predicted value under The model classifies samples according to input features. The estimated value of To include Samples training data, For different time steps The output quantile under The prediction data set, , is the maximum value of the time step.

[0020] Normalized quantile loss is used for validation and testing of the TFT prediction proxy model:

[0021] ;

[0022] in, is the forecast time span, is the maximum length of the time series, represents the domain value of the test sample, represents the normalized quantile loss, Represents the model at a point in time About time point Quantile The predicted value of .

[0023] Further, in step S2, the gated residual network is defined as:

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] in, is the input variable parameter, is the static variable vector from the static selection network, With width multiplication factor The gated residual network of is the linear unit activation function, To implement the standard normalization function, is a gated linear unit, is the sigmoid activation function, is the weight matrix of the first transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the second transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the third transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the first linear transformation in GLU, is the weight matrix of the second linear transformation in GLU, For The corresponding bias vector, For The corresponding bias vector, represents the element-wise product, is the width multiplication factor, is the output of the first transformation layer, For the function The processed output, for The input variable parameters.

[0029] Furthermore, in step S3, the multi-objective optimization model includes three factors: objective function, optimization variables and constraints; the objective function is to maximize the recovery rate, maximize the storage volume and maximize the NPV; the optimization variables include the throughput development method and the production system of each horizontal well, wherein the production system includes the mixed gas ratio, gas injection rate, gas injection time, well shut-in time, oil production rate, flowback time and throughput rounds, and quantitative injection is set in the gas injection stage of the throughput well, and quantitative production is set in the flowback stage; regarding the constraints: the key parameters of each throughput well, including daily gas injection volume and daily oil production, do not exceed the maximum working capacity upper limit of each single well, and are not lower than zero.

[0030] Furthermore, there is a correlation between the throughput parameters of each well under different throughput development modes. represents the middle horizontal well, and Represents two edge wells, and the constraint expressions are as follows.

[0031] When the three wells are in synchronous huff and puff, the huff and puff cycles are the same, with only differences in the gas mixture ratio, gas injection rate, oil production rate, and huff and puff rounds:

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] in, Indicates The ratio of mixed gas injected into horizontal wells, Indicates The gas injection rate of the horizontal well, Indicates The oil production rate of the horizontal well, , , Respectively , and The injection time, , , Respectively , and The stewing time, , , Respectively , and The return time, Indicates The throughput cycles of the horizontal wells, and Respectively represent the minimum and maximum value of the mixed gas ratio, and represent the minimum and maximum values ​​of the gas injection rate, respectively. and represent the minimum and maximum oil production rates, respectively. and Represent the minimum and maximum values ​​of the gas injection time, and Respectively represent the minimum and maximum values ​​of the soaking time. and Represent the minimum and maximum values ​​of the return time, and Represent the maximum and minimum values ​​of the throughput round respectively.

[0040] When asynchronous huff and puff is used, the gas injection time, well shut-in time and flowback time of the three wells are the same at each time step. Only the mixed gas ratio, gas injection rate, oil production rate and huff and puff rounds are different:

[0041] ;

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] ;

[0048] in, and They represent the minimum and maximum values ​​of each time step respectively. The gas injection time, shut-off time and flowback time of the three wells are the same in each time step.

[0049] When the well is pumping, In depletion development, there is only one variable, oil production rate; and The throughput cycles are the same, only the mixed gas ratio, gas injection rate, oil production rate and throughput rounds are different:

[0050] , ;

[0051] , ;

[0052] ;

[0053] ;

[0054] ;

[0055] ;

[0056] , ;

[0057] in, express The ratio of the injected mixed gas, express The ratio of the injected mixed gas, express The injection speed, express The gas injection rate.

[0058] Furthermore, in step S3, the TFT prediction agent model established in step S2 is coupled, and the TFT prediction agent model is used to quickly predict the recovery factor, storage volume and NPV. Secondly, the NSGA-Ⅲ multi-objective optimization algorithm is called to dynamically control the throughput development mode and the throughput working system of each well to ensure that an efficient and reasonable working system is matched at different time steps, so as to achieve dynamic optimization of the CO2 mixed gas oil recovery and storage effect in tight oil reservoirs.

[0059] Compared with the prior art, the beneficial technical effects of the present invention are: (1) the TFT prediction agent model established by the present invention trains the TFT neural network by screening the data set, which is conducive to more accurate fitting and prediction of the throughput effect; the dynamic optimization model of the CO2 mixed gas throughput effect established by the present invention can optimize the processing of multiple development goals with conflicting interests (storage volume, recovery rate and NPV), and combine with the TFT prediction agent model to obtain the storage volume, recovery rate and NPV optimization plan based on the benchmark plan; it is conducive to maximizing the storage volume, recovery rate and NPV, and can provide guidance for the optimal design of the CO2 mixer throughput development plan, which has positive significance for reservoir development and management.

[0060] (2) The dynamic optimization model of CO2 mixed gas throughput effect established in the present invention effectively improves production efficiency and reduces risks and uncertainties. In addition, the model comprehensively considers the three development objectives of storage volume, recovery rate and NPV, and achieves balanced optimization among multiple objectives. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is the relationship curve between the loss value of the TFT neural network and the number of iterations in Example 1.

[0062] Figure 2 It is a comparison chart of the predicted value and the true value of the recovery factor of the TFT prediction agent model after training in Example 1.

[0063] Figure 3 This is a comparison chart of the predicted value and the actual value of the buried storage volume by the TFT prediction agent model after training in Example 1.

[0064] Figure 4 It is a comparison chart of the predicted value and the true value of NPV by the trained TFT prediction agent model in Example 1.

[0065] Figure 5 It is a comparison chart of the average formation pressure changing over time under the production capacity scheme and the basic scheme in Example 1.

[0066] Figure 6 It is a comparison chart of the recovery rate change trend over time under the capacity scheme and the basic scheme in Example 1.

[0067] Figure 7 It is a comparison chart of the change trend of structural storage volume over time under the storage scheme in Example 1 and the basic scheme.

[0068] Figure 8 It is a comparison chart of the time-varying trends of bound gas storage volume under the storage scheme in Example 1 and the basic scheme. DETAILED DESCRIPTION

[0069] Example 1: This example takes the actual data of a tight sandstone reservoir as an example, and extracts the target well group from the actual heterogeneous geological model of a block. By coupling the tight oil reservoir TFT prediction agent model with the multi-objective optimization algorithm, an agent-assisted multi-objective optimization process is established, and the maximum recovery rate, maximum storage volume and maximum net present value (NPV) are selected as the objective function. The multi-level parameter and multi-objective dynamic optimization of CO2 mixed gas throughput is carried out to achieve dynamic regulation of the working system of each well, and form a set of solutions that consider different development goals.

[0070] Based on the above data, a dynamic optimization method of carbon dioxide mixed gas throughput parameters based on multi-objective interaction is used to jointly optimize well location and injection and production parameters, including the following steps: S1, select the target block for design and collect the time series prediction sample set.

[0071] (1) Selection of target blocks: The target blocks are selected mainly based on the remaining oil storage characteristics and seepage characteristics of the blocks, while taking into account the influence of faults and special structures. A 1800-square-meter target block is selected from the geological model. An actual model of 1510 m was built, and three horizontal wells were designed and established.

[0072] (2) Constructing a time series prediction sample set: The time series data involved in this embodiment is set as the throughput development mode, as well as the mixed gas ratio, gas injection rate, oil production rate, gas injection time, well shut-in time, flowback time of each well, and the crude oil recovery rate, CO2 storage volume, and NPV that change with time as output variables. A set of production parameters is set to be used every 5 years, and the time interval is set to record a set of sample data every 6 months. The setting of the throughput development parameters is mainly based on the actual production conditions of the oil field site to constrain the parameter range. Based on the Monte Carlo sampling method, the specific variable values ​​are extracted within the setting range of each parameter, and the throughput development parameters of each well are assigned separately. Based on the numerical simulation model of the target block reservoir, 600 groups of different production plans are randomly selected and set, and the time step is set to 6 months. A total of 18,000 data samples under different production times and different production systems can be obtained, which constitute the time series prediction sample set required by this embodiment.

[0073] S2. Standardize the sample data, divide the data set, use the TFT neural network to build the TFT prediction agent model, and use the TFT prediction agent model to quickly predict the recovery factor, storage volume and NPV.

[0074] (1) Sample data standardization: Since the unprocessed sample data has differences in units and magnitudes, in order to ensure the efficiency and accuracy of the TFT prediction agent model training process, the sample data needs to be standardized. This embodiment uses the minimum and maximum standardization method to normalize each feature:

[0075] ;

[0076] in, is the normalized value, is the actual value of the feature factor in the data set, is the maximum value of the characteristic factor in the original data, It is the minimum value of the characteristic factor in the original data.

[0077] (2) Design of TFT neural network: The design of TFT neural network aims to use canonical components to effectively represent various input types, including static features that do not change over time, past information, and constructed features of future known inputs. The components of TFT neural network include input features, variable selection network, static covariate encoder, sequence-to-sequence layer, gated residual network, multi-head attention mechanism, time fusion decoder, and quantile output.

[0078] I. Input features: When the TFT prediction agent model is used for time series problem prediction, the input variables are divided into two categories: static variables and dynamic variables. Static variables are divided into static continuous variables and static discrete variables, and dynamic variables include dynamic time-varying variables and dynamic time-invariant variables. Among them, dynamic time-varying variables refer to variables whose characteristics change over time and cannot be inferred, such as recovery rate, storage volume, net present value and other characteristics; dynamic time-invariant variables are variables that can be applied to the TFT prediction agent model based on reasoning, such as months, cumulative gas injection volume and other characteristics. Dynamic time-varying variables provide sensitivity to changes in time series and can capture key features in the data that evolve over time. In contrast, dynamic time-invariant variables provide the TFT prediction agent model with features that remain stable in the time dimension, allowing the TFT prediction agent model to better understand factors that are not related to time.

[0079] II. Variable Selection Network (VSN): The main function of the variable selection network is to select relevant input variables for each time step. By strengthening the key control factors for the prediction task, VSN can effectively improve the prediction performance of the TFT prediction agent model. At the same time, VSN also has the ability to weaken or discard irrelevant information, thereby optimizing the representation of input features and making the TFT prediction agent model more focused on accurate prediction of the target.

[0080] This strategy of introducing variable selection networks helps improve the expressive power of TFT neural networks when processing time series data. By automatically learning and selecting input variables, the TFT forecasting agent model can adapt to different tasks and data characteristics more flexibly. The role of VSN is similar to an intelligent filter, which enables the TFT forecasting agent model to capture key information more accurately and improve the overall forecasting performance when facing complex time series forecasting problems.

[0081] III. Static Covariate Encoder (SCE): Compared with other time series prediction architectures, the TFT neural network effectively integrates information from static metadata. The introduction of the static covariate encoder allows static features to be integrated into the entire network structure, and the temporal dynamics are adjusted by encoding the context vector. This integration makes the TFT neural network more adaptable and can make full use of static information for prediction while considering temporal dynamics.

[0082] IV. Sequence-to-sequence layer: This part uses two groups of LSTM networks, one for processing past input features and encoding past historical inputs, and the other for processing future input features, acting as a decoder. By locally processing known input features and observed input features, appropriate inductive bias is provided for the temporal ordering of inputs, thereby improving the performance of the next layer based on the temporal attention architecture.

[0083] V. Gated Residual Network (GRN): Gated Residual Network is a selection framework that uses a gating mechanism to selectively introduce relevant input features through optional skip connections and gated matrix filtering. GRN consists of dense layers followed by a gating mechanism that selectively offsets values ​​to remove unnecessary information. They can be defined as:

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] in, is the input variable parameter, is the static variable vector from the static selection network, With width multiplication factor The gated residual network of is the linear unit activation function, To implement the standard normalization function, is a gated linear unit, is the sigmoid activation function, is the weight matrix of the first transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the second transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the third transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the first linear transformation in GLU, is the weight matrix of the second linear transformation in GLU, For The corresponding bias vector, For The corresponding bias vector, represents the element-wise product, is the width multiplication factor, is the output of the first transformation layer, For the function The processed output, for The input variable parameters.

[0089] VI. Multi-head attention mechanism: Based on the multi-head attention mechanism in the Transformer model, the TFT prediction agent model also learns to capture long-term dependencies at different time steps through the multi-head attention mechanism. The TFT prediction agent model has improved it by introducing multiple attention heads, so that the TFT prediction agent model can focus on different parts of the input sequence in parallel, thereby better capturing the associated information in the time series. By increasing the attention weight of the TFT prediction agent model for specific features, the importance of key features for the prediction results is emphasized. This not only improves the modeling ability of the TFT prediction agent model for long-term dependencies, but also enhances the interpretability of the prediction results of the TFT prediction agent model.

[0090] VII. Temporal Fusion Decoder (TFD): The temporal fusion decoder consists of two layers of gated residual networks and a multi-head attention mechanism layer. In this structure, the first layer of gated residual network combines the output information of the sequence-to-sequence layer with the information from the static features to achieve the fusion of past and static features; the multi-head attention mechanism layer then performs a standard self-attention mechanism on the input information at all time steps, so that the TFT prediction agent model can better capture the long-term dependencies within the sequence; finally, the output of the multi-head attention mechanism and the remaining connections are passed to another layer of gated residual network for nonlinear processing.

[0091] VIII. Quantile Output: Unlike other prediction models that give a single point estimate, the TFT neural network determines the prediction intervals for the range of possible target values ​​at each prediction layer through quantile predictions. Quantile regression is used to generate prediction intervals to obtain more information about the certainty of the prediction. Quantile predictions are generated through a linear transformation of the output of the temporal fusion decoder. The TFT prediction proxy model trains the TFT neural network by jointly minimizing the quantile loss of all quantile outputs:

[0092] ;

[0093] ;

[0094] in, is the total loss function, is the quantile loss function, For the The actual observed value at each time step, For the model time steps in quantile The predicted value under The model classifies samples according to input features. The estimated value of To include Samples training data, For different time steps The output quantile under The prediction data set, , is the maximum value of the time step.

[0095] Normalized quantile loss is used for validation and testing of the TFT prediction proxy model:

[0096] ;

[0097] in, is the forecast time span, is the maximum length of the time series, represents the domain value of the test sample, represents the normalized quantile loss, Represents the model at a point in time About time point Quantile The predicted value of .

[0098] (3) Dataset division: Based on the TFT neural network time series algorithm, the tight oil reservoir CO2 mixed gas throughput time series prediction sample set established in this embodiment is trained and trained. The original data is divided into a training set, a validation set, and a test set according to a ratio of 6:2:2, which are used for the training of the TFT prediction agent model, the tuning of the TFT neural network hyperparameters, and the performance evaluation of the TFT prediction agent model, respectively.

[0099] (4) Hyperparameter tuning: The hyperparameter optimization process of the TFT neural network is a key step, which involves selecting and adjusting the neural network structure and various parameters during its training process to maximize the performance of the TFT prediction agent model. Hyperparameter optimization is performed through random search, and the number of iterations is set to 200. The final hyperparameter configuration of the TFT neural network after tuning is shown in Table 1.

[0100] Table 1 TFT neural network hyperparameter configuration

[0101]

[0102] (5) Forecast effect evaluation: The evaluation method uses the mean absolute percentage error ( ), mean square error ( ) and mean absolute error ( ) are used to evaluate and analyze the prediction effect of the TFT prediction proxy model for tight oil reservoirs. This evaluation index is used to measure the degree of dispersion between the actual value calculated by numerical simulation and the predicted value of the TFT prediction proxy model. Through these indicators, the accuracy of the TFT prediction proxy model can be quantified and its degree of fit to the actual observed values ​​in the validation set can be evaluated. Different evaluation indicators have different focuses, and the specific calculation process is as follows:

[0103] ;

[0104] ;

[0105] ;

[0106] in, For the The actual value, For the Prediction values, is the number of actual values ​​in the validation set.

[0107] The smaller the values ​​of the three evaluation indicators are, the smaller the error between the predicted value of the TFT prediction agent model and the actual observed value calculated by numerical simulation is, which proves that the performance ability of the established TFT prediction agent model is stronger.

[0108] The prediction effect evaluation and analysis results of the TFT prediction agent model in this embodiment are shown in Table 2. The relationship curve between the loss value of the TFT neural network obtained through training and the number of iterations is shown in Table 2. Figure 1 As shown, it shows that the TFT neural network obtained through training has good convergence and can predict errors.

[0109] Table 2 Evaluation indicators of prediction effect of TFT prediction agent model

[0110]

[0111] From the total of 600 production plans in the time series prediction sample set, 120 production plan test sets were selected, and the sample data at the end time were used to conduct a detailed comparison of the prediction effect. The comparison between the predicted value and the actual value of the recovery factor, storage volume and NPV is shown in Figure 2. Figure 2 , Figure 3 and Figure 4 .Depend on Figure 2 to Figure 4It can be seen that during the TFT prediction proxy model test, the model can better fit the actual recovery rate, storage volume and NPV calculated by numerical simulation, among which the recovery rate has the best fitting effect, followed by NPV and storage volume. The error between the actual value and the predicted value is within the allowable reasonable range, indicating that the TFT prediction proxy model established in this embodiment can effectively predict the oil displacement-storage effect under different production systems, and at the same time proves that the use of TFT neural network can achieve accurate prediction of CO2 mixed gas throughput effect.

[0112] The production system data of a certain scheme in the test set was selected, and the TFT prediction agent model based on the TFT neural network was used to predict its recovery factor, storage volume and NPV, and compared with the numerical simulation calculation results. The comparison results show that the prediction results of the recovery factor are very close to the numerical simulation calculation results, indicating that the TFT prediction agent model has a high ability to restore the actual production trend; for the two targets of storage volume and NPV, the model also shows a high prediction accuracy, and the error is always kept within an acceptable range. In summary, the error between the prediction results of the TFT prediction agent model and the numerical simulation calculation results is relatively small, indicating that the prediction effect of the TFT prediction agent model is within the ideal range, with high credibility, and can replace numerical simulation calculations to achieve rapid response to different goals.

[0113] S3. Determine the objective function, construct a multi-objective optimization model, and establish a dynamic optimization model for the CO2 mixed gas throughput effect based on the NSGA-Ⅲ multi-objective optimization algorithm to ensure that a reasonable and efficient working system is matched at different time steps and realize the dynamic optimization of the CO2 mixed gas throughput parameters in tight oil reservoirs.

[0114] (1) Problem Analysis: This example uses the maximum recovery factor, maximum storage volume, and maximum NPV as the objectives to perform multi-objective dynamic optimization on the development parameter combination that changes every 5 years during the tight oil reservoir huff-and-puff development process. The 15-year production process is divided into three stages: early production, mid-term, and late production, and three sets of production system solutions that change over time are optimized.

[0115] The three objective functions of recovery rate, storage capacity and NPV reflect the oil recovery effect, storage effect and overall economic effect of the huff and puff process respectively. Recovery rate is the main indicator of oil recovery effect in the huff and puff process, storage capacity is related to the storage effect of the huff and puff process, and the calculation of NPV not only reflects the economic benefits of the huff and puff process, but also reflects the gas utilization rate of the CO2 mixed gas to a certain extent. The calculation formula of net present value NPV is as follows:

[0116] ;

[0117] in, is the net present value, 10 7Yuan; is the production time, year, ; is the discount rate; is the cumulative oil production, m 3 ; is the cumulative gas volume, m 3 ; is the current crude oil price, yuan / m 3 ; is the current gas injection cost, yuan / m 3 .

[0118] (2) Establishment of optimization mathematical model: When optimizing the parameters of CO2 mixed gas throughput, due to the contradictory relationship between recovery rate, NPV and storage volume, in order to obtain the best development effect, this embodiment achieves the highest benefit by optimizing the three objectives simultaneously and independently. It can be seen that the optimization process is a multi-objective and multi-variable engineering problem. The optimization mathematical model established for this problem mainly includes three factors, namely objective function, optimization variables and constraints.

[0119] I. Objective function: The core of the optimization problem of CO2 mixed gas throughput oil recovery and storage effect in tight oil reservoirs is the multi-objective, multi-round and dynamic optimization of throughput development parameters. In the process of throughput development, high production capacity does not necessarily mean high storage effect and economic benefit. Adjusting the proportion of injected mixed gas and other throughput development parameters may have different degrees of impact on production capacity, storage effect and economic benefit. In order to ensure the increase of CO2 storage while efficiently exploiting crude oil, this embodiment establishes a multi-objective optimization mathematical model for CO2 mixed gas throughput parameters in tight oil reservoirs with maximizing recovery rate, maximizing storage volume and maximizing NPV as objective functions. The three objectives are interrelated, which is challenging for the selection and adjustment of throughput development parameters.

[0120] II. Optimization variables: In the optimization problem of CO2 mixed gas huff-and-puff oil recovery and storage effect, the optimization variables involved include the huff-and-puff development mode and the production system of each horizontal well, where the production system includes seven parameters, namely, the ratio of mixed gas, gas injection rate, gas injection time, well shut-in time, oil production rate, flowback time and huff-and-puff rounds. Set the huff-and-puff well to have quantitative injection in the gas injection phase and quantitative production in the flowback phase.

[0121] III. Constraints: In the problem of throughput parameter optimization, the constraint range of the change of the horizontal well working system needs to be considered. The key parameters such as the daily gas injection volume and daily oil production of each throughput well must follow the actual operation restrictions of the oil field site, that is, these parameters must not exceed the maximum working capacity of each single well, nor be lower than zero. At the same time, there is a certain correlation between the throughput parameters of each well under different throughput development methods. represents the middle horizontal well, and Represents two edge wells, and the constraint expressions are as follows.

[0122] ① When the three wells are operated synchronously, the three wells have the same operation cycle, and only the mixed gas ratio, gas injection rate, oil production rate and operation rounds are different:

[0123] ;

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] ;

[0130] in, Indicates The ratio of mixed gas injected into horizontal wells, Indicates The gas injection rate of the horizontal well, Indicates The oil production rate of the horizontal well, , , Respectively , and The injection time, , , Respectively , and The stewing time, , , Respectively , and The return time, Indicates The throughput cycles of the horizontal wells, and Respectively represent the minimum and maximum value of the mixed gas ratio, and represent the minimum and maximum values ​​of the gas injection rate, respectively. and represent the minimum and maximum oil production rates, respectively. and Represent the minimum and maximum values ​​of the gas injection time, and Respectively represent the minimum and maximum values ​​of the soaking time. and Represent the minimum and maximum values ​​of the return time, and Represent the maximum and minimum values ​​of the throughput round respectively.

[0131] ② When the three wells are in asynchronous huff-and-puff mode, the gas injection time, well shut-off time and flowback time at each time step are the same. Only the mixed gas ratio, gas injection rate, oil production rate and huff-and-puff rounds are different:

[0132] ;

[0133] ;

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] ;

[0139] in, and They represent the minimum and maximum values ​​of each time step respectively. The gas injection time, shut-off time and flowback time of the three wells are the same in each time step.

[0140] ③When the well is pumping, In depletion development, there is only one variable, oil production rate; and The throughput cycles are the same, only the mixed gas ratio, gas injection rate, oil production rate and throughput rounds are different:

[0141] , ;

[0142] , ;

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] , ;

[0148] in, express The ratio of the injected mixed gas, express The ratio of the injected mixed gas, express The injection speed, express The gas injection rate.

[0149] (3) Based on the NSGA-Ⅲ multi-objective optimization algorithm, a dynamic optimization model for the CO2 mixed gas throughput effect is established.

[0150] The TFT prediction agent model established in step S2 is coupled, and the TFT prediction agent model is used to quickly predict the recovery factor, storage volume and NPV. Secondly, the NSGA-Ⅲ multi-objective optimization algorithm is called to dynamically adjust the throughput development mode and the throughput working system of each well, aiming to ensure that an efficient and reasonable working system can be matched at different time steps, so as to achieve dynamic optimization of the CO2 mixed gas oil recovery and storage effect in tight oil reservoirs.

[0151] In order to implement the NSGA-Ⅲ multi-objective optimization algorithm, the throughput parameters are optimized and designed based on the established multi-objective optimization mathematical model of tight oil reservoir throughput parameters. Generally speaking, the value range of the cross coefficient is usually 0.40-0.99, and the value range of the coefficient of variation is usually 0.0001-0.1. The selection of the cross coefficient and the coefficient of variation takes into account their role in the performance of the algorithm. In this embodiment, after hyperparameter tuning, this embodiment selects a cross coefficient of 0.45, a coefficient of variation of 0.05, a population size of 200, and a genetic generation of 300 as the parameters of NSGA-Ⅲ. The reasonable selection of this parameter helps to efficiently find high-quality throughput parameter optimization solutions.

[0152] For the target well group of tight oil reservoirs selected in this embodiment, when the recovery factor is large, the storage volume is usually large, while the NPV is usually small. Under high economic benefits, it is impossible to meet good oil recovery benefits and storage effects, which means that there is a contradictory relationship between the three objective functions. In this case, the NSGA-III optimization algorithm will produce a set of non-dominated solutions.

[0153] The optimization result analysis of this embodiment will take the synchronous throughput initial design scheme of CO2 mixed gas throughput parameters as the benchmark, and select three different throughput schemes that match the benchmark scheme in terms of recovery rate, storage volume, and net present value from the Pareto optimal solution set, as the "production capacity scheme", "storage scheme", and "economic scheme" respectively. By comparing the different objective function values ​​of each optimization scheme with the basic scheme, the optimal solution with the same storage volume and NPV as the basic scheme is selected as the production capacity scheme, the optimal solution with the same recovery rate and NPV as the basic scheme is selected as the storage scheme, and the optimal solution with the same recovery rate and storage volume as the basic scheme is selected as the production capacity scheme, and three sets of optimization schemes are generated considering different development goals. The specific result comparison information is shown in Table 3. Reservoir engineers can select the most suitable optimization scheme according to the focus of the target.

[0154] Table 3 Comparison of calculation results between the optimization scheme and the basic scheme

[0155]

[0156] I. Analysis of production plan: Under the same storage volume and NPV, the recovery factor of the production plan is about 3.81% higher than that of the basic plan. The optimized production plan working system is shown in Table 4. The target well group coordinates the injection and production throughput relationship between wells at different levels in different production stages, so that each throughput well can maximize its development potential.

[0157] Table 4 Capacity plan work system

[0158]

[0159] The average formation pressure and recovery rate change trends over time under the production plan and the basic plan are as follows Figure 5 and Figure 6 In the initial stage of production, both the basic plan and the capacity plan adopt the throughput method of synchronous throughput, and the average formation pressure gradually decreases. In the middle and late stages of production, the capacity plan design adopts the throughput method of asynchronous throughput, which effectively raises the formation pressure, thereby better replenishing the formation energy and improving the oil recovery rate.

[0160] II. Storage Scheme Analysis: Under the same recovery factor and NPV, the storage volume of the storage scheme is about 10.15% higher than that of the basic scheme. The optimized storage scheme working system is shown in Table 5. By coordinating the injection-production throughput relationship between the horizontal wells in the target well group at different production stages, the CO2 storage effect of each throughput well can be maximized while satisfying the constraints.

[0161] Table 5: Storage plan work system

[0162]

[0163] The temporal trends of structural storage and bound gas storage under the storage scheme and the basic scheme are as follows: Figure 7 and Figure 8 As shown. The storage scheme is designed with a throughput method of well separation. Since the middle well is always exhausted during the well separation, the side wells have a displacement effect on the middle well during the well separation, which promotes the flow of crude oil and CO2, allowing the structural storage to accumulate and be relatively concentrated in the formation. The structural storage volume is higher than the storage volume under the basic scheme of synchronous throughput. At the same time, by adjusting the gas injection volume at different production stages, the overall storage volume of the storage scheme has been greatly improved.

[0164] III. Economic plan analysis: Under the same recovery rate and storage volume, the NPV of the economic plan is 0.38 higher than that of the basic plan. 10 7 The optimized economic scheme working system is shown in Table 6. Through dynamic optimization of different production stages, the working system of the economic scheme is determined, that is, to maximize the NPV under given conditions, so that the injected gas can have a larger swept area, and the output can be increased while ensuring higher economic benefits.

[0165] Table 6 Economic plan work system

[0166]

[0167] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction, characterized in that: The following steps are involved: S1, select the target block for design and collect the time series prediction sample set; S2. Standardize the sample data, divide the data set, use the TFT neural network to build a TFT prediction agent model, and use the TFT prediction agent model to quickly predict the recovery factor, buried storage volume and NPV; S3. Determine the objective function, construct a multi-objective optimization model, and establish a dynamic optimization model for the CO2 mixed gas throughput effect based on the NSGA-Ⅲ multi-objective optimization algorithm to ensure that a reasonable and efficient working system is matched at different time steps and realize the dynamic optimization of the CO2 mixed gas throughput parameters in tight oil reservoirs.

2. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 1, characterized in that: In step S2, the sample data standardization process uses the minimum and maximum standardization method to normalize each feature: ； in, is the normalized value, is the actual value of the feature factor in the data set, is the maximum value of the characteristic factor in the original data, It is the minimum value of the characteristic factor in the original data; The normalized raw data is divided into training set, validation set and test set according to the ratio of 6:2:2, which are used for training the TFT prediction agent model, tuning the hyperparameters of the TFT neural network and performance evaluation of the TFT prediction agent model respectively.

3. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 2, characterized in that: In step S2, the components of the TFT neural network include input features, variable selection network, static covariate encoder, sequence-to-sequence layer, gated residual network, multi-head attention mechanism, time fusion decoder and quantile output; Regarding input features: TFT neural network divides input variables into static variables and dynamic variables. Static variables are divided into static continuous variables and static discrete variables, and dynamic variables include dynamic time-varying variables and dynamic time-invariant variables. Dynamic time-varying variables refer to variables whose characteristics change over time and cannot be inferred, including recovery factor, storage volume and net present value; dynamic time-invariant variables are variables applied to the TFT prediction agent model based on inference, including month and cumulative gas injection volume.

4. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 3 is characterized in that: In step S2, the functions of the variable selection network include: selecting relevant input variables for each time step, and weakening or discarding irrelevant information; The static covariate encoder enables the incorporation of static features into the TFT neural network, regulating temporal dynamics by encoding the context vector; The sequence-to-sequence layer uses two groups of LSTM networks, one of which is used to process past input features and encode past historical inputs; the other is used to process future input features; The gated residual network is a selection framework that uses a gating mechanism to selectively introduce relevant input features through optional skip connections and gated matrix filtering; The TFT prediction agent model learns to capture long-term dependencies at different time steps through a multi-head attention mechanism, while focusing on different parts of the input sequence in parallel and increasing the model attention weight for specific features; The temporal fusion decoder consists of two layers of gated residual networks and a multi-head attention mechanism layer; the first layer of the gated residual network combines the output information of the sequence-to-sequence layer with the information from the static features, and then performs a standard self-attention mechanism on the input information at all time steps through the multi-head attention mechanism layer. Finally, the output of the multi-head attention mechanism layer and the remaining connections are passed to another layer of the gated residual network for nonlinear processing.

5. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 4, characterized in that: In step S2, the TFT neural network determines the prediction interval of the possible target value range of each prediction layer through quantile prediction, and generates the prediction interval using quantile regression to obtain information about the prediction certainty; the quantile prediction is generated by the linear transformation of the output of the time fusion decoder; The TFT prediction agent model trains the TFT neural network by jointly minimizing the quantile loss of all quantile outputs: ； ； in, is the total loss function, is the quantile loss function, For the The actual observed value at each time step, For the model time steps in quantile The predicted value under The model classifies samples according to input features. The estimated value of To include Samples training data, For different time steps The output quantile under The prediction data set, , is the maximum value of the time step; Normalized quantile loss is used for validation and testing of the TFT prediction proxy model: ； in, is the forecast time span, is the maximum length of the time series, Represents the domain value of the test sample, represents the normalized quantile loss, Represents the model at a point in time About time point Quantile The predicted value of .

6. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 5, characterized in that: In step S2, the gated residual network is defined as: ； ； ； ； in, is the input variable parameter, is the static variable vector from the static selection network, With width multiplication factor The gated residual network of is the linear unit activation function, To implement the standard normalization function, is a gated linear unit, is the sigmoid activation function, is the weight matrix of the first transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the second transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the third transformation layer in GRN, For The corresponding bias vector, is the weight matrix of the first linear transformation in GLU, is the weight matrix of the second linear transformation in GLU, For The corresponding bias vector, For The corresponding bias vector, represents the element-wise product, is the width multiplication factor, is the output of the first transformation layer, For the function The processed output, for The input variable parameters.

7. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 6, characterized in that: In step S3, the multi-objective optimization model includes three factors: objective function, optimization variables and constraints; The objective function is to maximize recovery, maximize storage volume and maximize NPV; The optimization variables include the throughput development mode and the production system of each horizontal well. The production system includes the mixed gas ratio, gas injection rate, gas injection time, well shut-in time, oil production rate, flowback time and throughput rounds. The throughput well is set to have quantitative injection in the gas injection stage and quantitative production in the flowback stage. Regarding the constraints: the key parameters of each throughput well, including daily gas injection volume and daily oil production, shall not exceed the maximum working capacity of each single well, and shall not be less than zero.

8. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 7, characterized in that: There is a correlation between the throughput parameters of each well under different throughput development methods. represents the middle horizontal well, and Represents two side wells, and the constraint expressions are as follows: When the three wells are in synchronous huff and puff, the huff and puff cycles are the same, with only differences in the gas mixture ratio, gas injection rate, oil production rate, and huff and puff rounds: ； ； ； ； ； ； ； in, Indicates The ratio of mixed gas injected into horizontal wells, Indicates The gas injection rate of the horizontal well, Indicates The oil production rate of the horizontal well, , , Respectively , and The injection time, , , Respectively , and The stewing time, , , Respectively , and The return time, Indicates The throughput cycles of the horizontal wells, and Respectively represent the minimum and maximum value of the mixture ratio, and represent the minimum and maximum values ​​of the gas injection rate, respectively. and represent the minimum and maximum oil production rates, respectively. and Represent the minimum and maximum values ​​of the gas injection time, and Respectively represent the minimum and maximum values ​​of the soaking time. and Represent the minimum and maximum values ​​of the flowback time, and Represent the maximum and minimum values ​​of the throughput round respectively; When asynchronous huff and puff is used, the gas injection time, well shut-in time and flowback time of the three wells are the same at each time step. Only the mixed gas ratio, gas injection rate, oil production rate and huff and puff rounds are different: ； ； ； ； ； ； ； in, and Represent the minimum and maximum values ​​of each time step respectively. The gas injection time, well soaking time and flowback time of the three wells are the same in each time step. When the well is pumping, In depletion development, there is only one variable, oil production rate; and The throughput cycles are the same, only the mixed gas ratio, gas injection rate, oil production rate and throughput rounds are different: ， ； ， ； ； ； ； ； ， ； in, express The ratio of the injected mixed gas, express The ratio of the injected mixed gas, express The injection speed, express The gas injection rate.

9. The method for dynamic optimization of carbon dioxide mixed gas throughput parameters based on multi-objective interaction according to claim 8, characterized in that: In step S3, the TFT prediction agent model established in step S2 is coupled, and the TFT prediction agent model is used to quickly predict the recovery factor, storage volume and NPV. Secondly, the NSGA-Ⅲ multi-objective optimization algorithm is called to dynamically control the throughput development mode and the throughput working system of each well to ensure that an efficient and reasonable working system is matched at different time steps, so as to achieve dynamic optimization of the CO2 mixed gas oil recovery and storage effect in tight oil reservoirs.

Images