Multi-objective optimization method and system for gas reservoir CO2 injection extraction and storage

Abstract

The invention belongs to the technical field of energy exploitation, and particularly relates to a multi-objective optimization method and system for gas reservoir injection, extraction and storage, and the method comprises the steps: firstly obtaining regulation and control parameter variables of extraction and storage, and calling an improved sampling algorithm for sampling combination to generate sample data; calling CMG data files for batch modification, extracting well names and DATE commands from the original dat file, simulating by combining sample parameters, generating a new scheduling block, inserting the new scheduling block into the last DATE command of the original file, generating a new CMG file, and exporting and merging CSV after processing; constructing a multi-output regression prediction model, and training the model by using a CSV file; then all regulation and control parameter variable combinations are extracted, data are exported, the trained model is called for prediction, and a parameter full combination is obtained; and finally, constructing a multi-objective optimization model, and exporting a Pareto optimal solution as a result by taking maximization of cumulative yield and storage efficiency as an objective. According to the invention, the problems of low efficiency and inaccurate precision of collaborative operation of sealing and natural gas exploitation in the prior art can be solved.

Description

Technical Field

[0001] This invention belongs to the field of energy extraction technology, and particularly relates to a gas reservoir injection method. A multi-objective optimization method and system for extraction and storage. Background Technology

[0002] Driven by both "dual carbon" goals and the upgrading of energy demand, Synergistic operations of geological storage and natural gas extraction have become a core direction for achieving low-carbon and efficient development in the energy sector. Currently, CMG numerical simulators (such as GEM composition models) are the core tools in this field, capable of constructing detailed models covering reservoir geological characteristics and fluid phase changes, simulating extraction efficiency and storage effects under different injection parameters, and providing support for engineering scheme design; simultaneously, Although flooding technology can improve gas reservoir recovery and achieve Storage is possible, but the existing technology system has multiple shortcomings and relies on manual labor. The optimization of parameters, numerical simulation and data processing for natural gas storage and extraction are cumbersome and consume a lot of manpower and resources, resulting in low efficiency and difficulty in meeting actual production and environmental protection needs.

[0003] Firstly, when determining engineering parameters, existing methods mostly rely on human experience to set parameters one by one. However, this approach is not only inefficient but also ignores the relationship between parameters, making it difficult to ensure that the globally optimal solution is obtained.

[0004] Next, in the CMG data file processing stage, the current practice largely relies on manual modification of the ".dat" parameters file by file. This modification process is tedious, complex, and prone to errors. Furthermore, the modification process lacks standardized records, resulting in poor traceability. Moreover, parameter setting errors are often only discovered after the simulation is completed and the results are reviewed.

[0005] Finally in In predicting the development potential of gas-driven reservoirs, traditional numerical simulation methods rely on specialized and expensive component simulators, resulting in extremely low computational efficiency and making it difficult to quickly and accurately assess actual development potential. Furthermore, current optimization objectives often overemphasize economic benefits or maximum storage capacity, severely neglecting other factors. The significant threat that leakage poses to the sealing effect greatly reduces the feasibility of the optimized results in practical engineering applications. These shortcomings collectively restrict the efficiency and accuracy of collaborative operations, urgently requiring breakthroughs in innovative technologies. Summary of the Invention

[0006] The technical problem solved by this invention is to provide a gas reservoir injection method. A multi-objective optimization method and system for extraction and storage is proposed to solve the problems in existing technologies. The combined operation of gas storage and extraction suffers from low efficiency and inaccurate precision.

[0007] The basic solution provided by this invention: a gas reservoir injection Multi-objective optimization methods for extraction and storage include: S1: Obtain The control parameter variables for extraction and storage are sampled and combined using a preset sampling algorithm to generate sample data. S2: Use the CMG data file batch modification tool to extract all well names and DATE commands from the original .dat file, verify the effectiveness of the target control wells, and locate the position of the last DATE command; read the control parameter variables from the sample data and perform injection-production ratio, Concentration calculation; then, based on the CMG syntax rules, a new scheduling block is generated and inserted into the last DATE command position of the original dat file. After generating a new complete CMG data file, it is imported into the CMG software for simulation. The CSV file is exported from the CMG running result file and merged. S3: Construct a multi-output regression prediction model. After preprocessing the exported merged CSV file, generate a training set, a validation set, and a test set. Input the training set into the multi-output regression prediction model for training. Validate and test the trained multi-output regression prediction model using the validation set and the test set until the trained multi-output regression prediction model is obtained. S4: Extract All control parameters for extraction and storage are combined and exported as file data. A trained multi-output regression prediction model is then used to predict the file data. All combinations of control parameters and variables in extraction and storage; S5: Use a multi-objective optimization algorithm to construct a multi-objective optimization model. The entire combination of control parameters and variables in extraction and storage is input into a multi-objective optimization model, and the maximization parameter is selected based on the optimization objective. Cumulative output and maximization Storage efficiency, deriving the Pareto optimal solution, as The results of multi-objective optimization of extraction and storage.

[0008] Furthermore, S1 includes: S1-1: For the acquisition The control parameters for extraction and storage are divided into equally probable sub-intervals according to their variable types, including discrete variables and continuous variables. S1-2: Allocate sampling quotas to all split points of each variable according to the total number of samples, and establish a sampling index list for each variable; S1-3: Randomly shuffle the sampling index list of each variable, randomly select a position from the random index list of each variable, extract the corresponding split point value to form a sampling combination, until the preset sample size is reached; S1-4: Use the improved Antlion optimization algorithm to extract the optimal boundary reinforcement combination of the extreme values ​​of each variable, and after performing a repeatability check on the sampling combination generated in S1-3, combine the sampling combination to generate the final variable sampling combination result.

[0009] Furthermore, the optimal boundary reinforcement combination for extracting the extreme values ​​of each variable by calling the improved antlion optimization algorithm in S1-4 is specifically as follows: Obtain each control parameter variable extreme values and build Class boundary target vector; Initialize the antlion colony and the ant colony, and... The class boundary target vector is added to the antlion population; The fitness function is constructed based on proximity to the boundary and engineering feasibility objectives, and its expression is:

[0010]

[0011]

[0012] in, For boundary-leading terms, This is a constraint on the feasibility of the project. For individual ants, Let k be the boundary target vector of the kth class. The distance is defined as Hamming distance, and the engineering constraints are preset based on the physical laws of gas reservoirs. For the penalty function; and The weighting coefficients are adjusted according to optimization requirements. and All are greater than 0, and their sum is 1; Project feasibility constraints Other representations in the text indicate that if x has one constraint that does not satisfy all engineering constraints; The ant's movement direction is adjusted by a boundary attraction factor to perform boundary-guided walking, expressed as follows:

[0013]

[0014] in, Indicates in In the iteration, the ant's new position on the i-th control parameter variable, Indicates the first In the iteration, the ant's current position on the i-th control parameter variable is... Indicates the first The step size of the random walk in the iteration. For the first Generated in generation iteration Uniformly distributed random numbers Let i be the current value of the ant in the t-th generation on the i-th variable. For the boundary attraction term, Let be the i-th extreme value of the boundary target vector that the ant individual is currently closest to; The attraction intensity coefficient, ; For individual ants The boundary target vector corresponding to the current minimum Hamming distance. Let x be the ant's current position and the optimal boundary target vector. Hamming distance, , which is the distance decay factor, representing that the closer an ant is to the boundary, the stronger the attraction, and the farther away it is, the weaker the attraction. After each iteration, based on the boundary guiding term Update the antlion trap radius, mark the individual with the best fitness in the population as an elite antlion, and force the retention of at least one individual that meets the criteria. An elite individual with an extreme value for a single variable; After the maximum number of iterations is reached, the best individuals are extracted from the final elite antlion population as the boundary reinforcement combination, and a repeatability judgment is performed on the sampled combination generated in S1-3. Finally, the final sample data of the random sampled combination and the deduplicated boundary reinforcement combination are output.

[0015] Furthermore, S2 includes: S2-1: Read the contents of the original dat file, extract all well names and DATE commands using regular expressions, verify the existence of wells with control parameter variables by using the extracted well names, and locate the position of the last DATE command in the original dat file; S2-2: Read the regulation parameter variables from the sample data and perform... Concentration calculation, injection-production ratio calculation, and according to CMG syntax rules, combine all monthly DATE commands for the start and end dates of injection wells, PRODUCER and OPERATE rate control statements for production wells, INJECTOR and INCOMP fluid concentration configurations for injection wells, and corresponding OPERATE statements to generate a new scheduling block; S2-3: Truncate the original dat file to the last DATE command, insert the generated new scheduling block, then concatenate the remaining content of the original dat file to form a complete new file, and generate unique filenames according to the preset file naming rules to batch generate CMG data files and save them to the specified directory.

[0016] Furthermore, S2 also includes: S2-4: After importing the batch-modified CMG data files into the CMG software for simulation, export the simulation results as a CSV file. Perform data cleaning on the exported CSV file, including using preset regular expression matching formats to match filenames and convert them to a standard format, handling invalid values, filtering data from the CSV file according to business rules, and calculating... The core metrics of storage efficiency; among them The storage efficiency expression is:

[0017] in, express Storage efficiency, Indicates cumulative injection Number of moles Indicates cumulative output Number of moles; if injected If it is 0, then Storage efficiency is 0; S2-5: Merge the CVS files after data cleaning into a single file, retaining the header of the first file, and appending only data rows to subsequent CSV files to generate a merged CVS file containing all samples.

[0018] Furthermore, S3 includes: S3-1: Constructing a multi-output regression prediction model based on an improved stacked ensemble algorithm; S3-2: After preprocessing the merged CVS files, divide them into training, validation, and test sets in a ratio of 7:1.5:1.5; S3-3: Input the training set into the multi-output regression prediction model for training, specifically: Based on the input training set, the first... Feature vector of each sample Input into two base models: Random Forest model expression:

[0019] XGBoost model expression:

[0020] Among them, for the training set Feature vector of each sample Random Forest Model Output predicted value vector XGBoost model It will output a vector of predicted values. , In the random forest model Forecast of cumulative output, In the random forest model Forecast of cumulative output, In the random forest model Predicted value of mole fraction; In the XGBoost model Forecast of cumulative output, In the XGBoost model Forecast of cumulative output, In the XGBoost model Predicted value of mole fraction; The prediction results of the two base models are concatenated into a meta-feature vector. The expression is:

[0021] Metamodel Meta-features As input, output the final predicted value. The expression is:

[0022] in, These are the final predicted results for the three dependent variables; express The final forecast of cumulative production, express The final forecast of cumulative production, express The final prediction result of the mole fraction; The meta-model is optimized by minimizing the loss function on the training set, expressed as:

[0023] in, For the first The true value of the t-th dependent variable in a sample, where n is the total number of samples. For the first The predicted value of the t-th dependent variable in a sample. Representing the metamodel; To impose a business scope constraint on the prediction results, the expression is:

[0024] in Let t represent the minimum value of the t-th dependent variable. Let t be the maximum value of the dependent variable. For the first The predicted value of the t-th dependent variable in a sample; S3-4: Based on the training results, call the preset indicators for evaluation, and perform validation and testing based on the validation set and test set until a multi-output regression prediction model that meets the evaluation indicators is output.

[0025] Furthermore, S5 includes: S5-1: Based on the complete combination of input control parameter variables, extract the injection rate, injection-to-enzyme ratio, and other parameters based on the date index. concentration, Gas mole fraction; S5-2: Call The algorithm constructs a multi-objective optimization model, calling the variable influence coefficient formula to evaluate injection rate, injection-to-enzyme ratio, and other variables. concentration, Influence coefficient analysis of gas mole fraction; S5-3: Injection rate and injection-production ratio based on influence coefficient analysis concentration, In a multi-objective optimization model, the gas mole fraction is used as the optimization target, and the output of the optimization objective is to maximize the gas mole fraction. Cumulative output and maximization Storage efficiency.

[0026] Furthermore, in S5-2, the variable influence coefficient formula is used to apply to the injection rate, injection-to-production ratio, concentration, The influence coefficient analysis of gas mole fraction is as follows: Analysis of the influence coefficient of injection rate:

[0027] in, Indicates the influence coefficient of injection rate. As the injection rate weight, The average injection rate. The standard deviation of the injection rate; Analysis of the influence coefficient of injection-production ratio:

[0028] in, This represents the influence coefficient of the injection-production ratio. For the injection-production ratio weighting, The injection-production ratio is the average. The standard deviation of the injection-production ratio; Concentration influence coefficient analysis includes respectively for Production analysis and Sequestration Analysis: Concentration Production Analysis:

[0029] Concentration Sequestration Analysis:

[0030] in, express Concentration The impact coefficient of output express Concentration The impact coefficient of sealing for Concentration weight, for The average concentration, for Standard deviation of concentration; Analysis of the influence coefficient of gas mole fraction:

[0031] in, express The influence coefficient of gas mole fraction for Mean of mole fraction denoted as the standard deviation of the mole fraction.

[0032] Furthermore, the optimization target output in S5-3 is to maximize Cumulative output and maximization Storage efficiency is specifically as follows: The optimization objective is to maximize The cumulative output expression is:

[0033] Maximize the optimization objective The storage efficiency expression is:

[0034] in, For the final calculation Cumulative production optimization target value, express The baseline value for cumulative production. For the final calculation Storage efficiency optimization target value for Storage efficiency benchmark, Avoid results that are 0 or negative.

[0035] A type of gas reservoir injection A multi-objective optimization system for extraction and storage, applying the above-mentioned gas reservoir injection... The multi-objective optimization method for extraction and storage includes a sample data generation module, a CMG processing module, a prediction model generation and training module, a variable combination module, and a multi-objective optimization module, wherein: Sample data generation module: used to acquire The control parameter variables for extraction and storage are sampled and combined using a preset sampling algorithm to generate sample data. CMG processing module: Used to call the CMG data file batch modification tool, extract all well names and DATE commands from the original .dat file, verify the effectiveness of the target control wells, and locate the position of the last DATE command; read control parameter variables from sample data, and perform injection-production ratio, Concentration calculation; then, based on the CMG syntax rules, a new scheduling block is generated and inserted into the last DATE command position of the original dat file. After generating a new complete CMG data file, it is imported into the CMG software for simulation. The CSV file is exported from the CMG running result file and merged. Prediction Model Generation and Training Module: This module is used to build a multi-output regression prediction model. It preprocesses the exported merged CSV file to generate training, validation, and test sets. The training set is then input into the multi-output regression prediction model for training. The training multi-output regression prediction model is validated and tested using the validation and test sets until the trained multi-output regression prediction model is obtained. Variable combination module: used for extraction All control parameters for extraction and storage are combined and exported as file data. A trained multi-output regression prediction model is then used to predict the file data. All combinations of control parameters and variables in extraction and storage; Multi-objective optimization module: Used to call multi-objective optimization algorithms to build multi-objective optimization models. The entire combination of control parameters and variables in extraction and storage is input into a multi-objective optimization model, and the maximization parameter is selected based on the optimization objective. Cumulative output and maximization Storage efficiency, deriving the Pareto optimal solution, as The results of multi-objective optimization of extraction and storage.

[0036] The principles and advantages of this invention are as follows: The technical solution of this application constructs a multi-objective optimization framework by integrating experimental design, numerical simulation, machine learning, and intelligent optimization algorithms; specifically, firstly, sampling processing combined with an improved antlion optimization algorithm is used to generate a sample of control parameter combinations, and CMG numerical simulation is used to obtain dynamic data of gas reservoir development, providing a high-quality dataset for model training; subsequently, a multi-output regression prediction model based on an improved stacking algorithm is used to learn parameters and development indicators (…). Yield, The mapping relationship (of storage efficiency) is used to replace the time-consuming numerical simulation process; finally, the NSGA-II algorithm is used to search for the Pareto optimal solution in the output space of the prediction model, thus achieving... Cumulative output and The core of collaborative optimization of storage efficiency is to solve the complexity and coupling of multi-objective decision-making in gas reservoir development through "data-driven modeling + intelligent optimization search".

[0037] The advantages of this application are: it significantly improves gas reservoir injection. This method optimizes the efficiency and scientific basis of data extraction and storage. By combining CMG numerical simulation with an improved antlion optimization algorithm, the physical reliability of the data is ensured while significantly reducing the computational cost of the optimization process. The NSGA-II algorithm effectively explores the nonlinear relationship between control parameters and dual objectives, and the generated Pareto optimal solution set presents the optimal solution under different objective weights, providing flexible selection space for engineering practice. Compared with traditional single-objective optimization or empirical decision-making, this method can improve optimization accuracy with the same computational resources, while simultaneously achieving... Recovery rate and The synergistic improvement in the sequestration rate has significant economic and environmental benefits. Attached Figure Description

[0038] Figure 1 This is a flowchart of an embodiment of the present invention; Figure 2 This is a functional block diagram of an embodiment of the present invention. Detailed Implementation

[0039] The following detailed description illustrates the specific implementation method: The implementation examples are basically as follows Figure 1 As shown: A gas reservoir injection Multi-objective optimization methods for extraction and storage include: S1: Obtain The control parameter variables for extraction and storage are sampled and combined using a preset sampling algorithm to generate sample data; wherein, S1 includes: S1-1: For the acquisition The control parameters for extraction and storage are divided into equally probable sub-intervals according to their variable types, including discrete variables and continuous variables. S1-2: Allocate sampling quotas to all split points of each variable according to the total number of samples, and establish a sampling index list for each variable; S1-3: Randomly shuffle the sampling index list of each variable, randomly select a position from the random index list of each variable, extract the corresponding split point value to form a sampling combination, until the preset sample size is reached; S1-4: Use the improved Antlion optimization algorithm to extract the optimal boundary reinforcement combination of the extreme values ​​of each variable, and after performing a repeatability check on the sampling combination generated in S1-3, combine the sampling combination to generate the final variable sampling combination result.

[0040] In this embodiment, the controllable parameter variables include injection rate, Concentration, injection-to-production ratio, and production rate. For example, for injection rate, the variable type is continuous, with a value range of 100,000-300,000 and a step size of 10,000. The concentration variable is a discrete variable with values ​​of 0.6, 0.7, 0.8, 0.9, and 1.0; the injection-to-production ratio variable is a discrete variable with values ​​of 0.6, 0.7, 0.8, 0.9, and 1.0; and the production rate variable is a continuous variable with values ​​determined based on the total sample size.

[0041] The values ​​of the various control parameter variables given above are for illustrative purposes only, and are also intended to better represent the numerical examples of different variable types in the control parameter variables. Subsequently, sample data was generated. To better and more comprehensively cover the value range of the control parameter variables, the sampling process first divided each variable into equally probable sub-intervals according to its value range. The sub-intervals differed for different variable types. Specifically: For discrete variables, no additional partitioning is performed; the set of discrete values ​​is used directly as the partitioning result. For example... If the concentration is a discrete variable with values ​​of 0.6, 0.7, 0.8, 0.9, and 1.0, then it should be divided directly by commas. For continuous variables, a linear partitioning method is used. For example, based on a set number of partitions (at least 2), the np.linspace function is used to divide the variable into segments. Equally spaced dividing points are generated within the interval. With a minimum value of 0, a maximum value of 10, and a dividing number of 5, this will generate... This is the result of the division.

[0042] After the variables are divided, sampling quotas are allocated to all the split points of each variable according to the total number of samples. The calculation method is to divide the total number of samples by the number of split points to obtain the quota. If the division is not even, the remainder is randomly allocated to the first few split points.

[0043] The sampling index list for each variable is then randomly shuffled. The sampling index list contains the quota number of each partition point to ensure that the sampling position of each partition point is randomly distributed and to avoid systematic errors.

[0044] Next, a position is randomly selected from the random index list of each variable, and the corresponding split point value is extracted to form a sampling combination. Here, the combination is stored in a set to ensure uniqueness. If a duplicate combination is generated, resampling is performed until the specified sample size is reached.

[0045] After generating the above random sampling combinations, the validity of the sample data cannot be guaranteed at this point. Therefore, a boundary reinforcement combination is added. The boundary reinforcement combination is generated using an improved antlion optimization algorithm, specifically: Obtain each control parameter variable extreme values and build Class boundary target vector; for discrete variables, the minimum / maximum value in the value set can be directly selected; for continuous variables, the extreme value is determined according to the gas reservoir physical constraints and engineering upper limit. Initialize the antlion colony and the ant colony, and... The class boundary target vector is added to the antlion population; The fitness function is constructed based on proximity to the boundary and engineering feasibility objectives, and its expression is:

[0046]

[0047]

[0048] in, For boundary-leading terms, For individual ants, Let k be the boundary target vector of the kth class. For Hamming distance, These are engineering feasibility constraints, which are preset based on the physical laws of gas reservoirs. They transform engineering experience and physical laws into something that the algorithm can recognize. This is a penalty function used to filter out infeasible parameter combinations in engineering, such as well leakage or pressure imbalance caused by excessive injection rate. The more severe the violation of the constraint, the larger the penalty value. and The weighting coefficients are adjusted according to optimization requirements. and All are greater than 0, and their sum is 1. For example, the priority boundary is... If the value is between 0.6 and 0.8, and the priority project is feasible, then... Take 0.7-0.9; Project feasibility constraint. Other representations in this application indicate that if x does not satisfy any of the engineering constraints; in this application, Hamming distance is introduced into the fitness function, where Hamming distance is used to measure the relationship between individual x and the boundary target vector. The degree of difference; the smaller the difference, the closer the individual is to the target boundary. The specific expression process is as follows: Define dimension matching function :

[0049] Indicates the precision threshold; Individual x and the boundary target vector of the kth class Hamming distance:

[0050] ,when This indicates that the individual perfectly matches the boundary target vector. This indicates a complete mismatch. This represents the total number of variables.

[0051] The ant's movement direction is adjusted by a boundary attraction factor to perform boundary-guided walking, expressed as follows:

[0052]

[0053] in, Indicates in In the iteration, the ant's new position on the i-th control parameter variable, Indicates the first In the iteration, the ant's current position on the i-th control parameter variable is... Indicates the first The step size of the random walk in the iteration. For the first Generated in generation iteration Uniformly distributed random numbers Let i be the current value of the ant in the t-th generation on the i-th variable. For the boundary attraction term, Let be the i-th extreme value of the boundary target vector that the ant individual is currently closest to; The attraction intensity coefficient, This is used to avoid over-guided local optima; For individual ants The boundary target vector corresponding to the current minimum Hamming distance. Let x be the ant's current position and the optimal boundary target vector. Hamming distance, , which is the distance decay factor, representing that the closer an ant is to the boundary, the stronger the attraction, and the farther away it is, the weaker the attraction. After each iteration, based on the boundary guiding term The expression to update the antlion trap radius is:

[0054] in, The initial trap radius, For the t-th generation of the current antlion, For the t-th generation antlion population; thus, when the boundary guiding term The smaller the antlion, the smaller the trap radius, and the easier it is to capture ants near the boundary, thus enhancing the local search in the boundary area; After each iteration, the individual with the best fitness in the population is marked as an elite antlion, specifically the top 5%-10% of individuals with the lowest fitness values, and at least one individual meeting the fitness requirement is forcibly retained. An elite individual with an extreme value for one variable ensures that the algorithm does not deviate from the boundary reinforcement target.

[0055] After the maximum number of iterations is reached, the optimal individual is extracted from the final elite antlion population as the boundary reinforcement combination. The optimal individual in this application satisfies the following two conditions: Condition 1: Includes Extreme values ​​of one variable; Condition 2: This indicates that the project is feasible.

[0056] Finally, the sampling combination generated in S1-3 is subjected to a repeatability test, and the final sample data of the random sampling combination and the deduplication boundary reinforcement combination are output.

[0057] Thus, in addition to the sampling combinations obtained by random sampling, the generation of sample data in this application also ensures the inclusion of boundary reinforcement combinations, ensuring that the sample data covers the entire region of the parameter variables.

[0058] S2: Use the CMG data file batch modification tool to extract all well names and DATE commands from the original .dat file, verify the effectiveness of the target control wells, and locate the position of the last DATE command; read the control parameter variables from the sample data and perform injection-production ratio, Concentration calculation; then, based on CMG syntax rules, a new scheduling block is generated and inserted into the last DATE command position of the original dat file to generate a new complete CMG data file. This file is then imported into CMG software for simulation. A CSV file is exported from the CMG execution result file and merged. S2 includes: S2-1: Read the contents of the original dat file, extract all well names and DATE commands using regular expressions, verify the existence of wells with control parameter variables by using the extracted well names, and locate the position of the last DATE command in the original dat file; S2-2: Read the regulation parameter variables from the sample data and perform... Concentration calculation, injection-production ratio calculation, and according to CMG syntax rules, combine all monthly DATE commands for the start and end dates of injection wells, PRODUCER and OPERATE rate control statements for production wells, INJECTOR and INCOMP fluid concentration configurations for injection wells, and corresponding OPERATE statements to generate a new scheduling block; S2-3: Truncate the original dat file to the last DATE command, insert the generated new scheduling block, then concatenate the remaining content of the original dat file to form a complete new file, and generate unique filenames according to the preset file naming rules to generate CMG data files in batches and save them to the specified directory. S2-4: After importing the batch-modified CMG data files into the CMG software for simulation, export the simulation results as a CSV file. Perform data cleaning on the exported CSV file, including using preset regular expression matching formats to match filenames and convert them to a standard format, handling invalid values, filtering data from the CSV file according to business rules, and calculating... The core metrics of storage efficiency; among them The storage efficiency expression is:

[0059] in, express Storage efficiency, Indicates cumulative injection Number of moles Indicates cumulative output Number of moles; if injected If it is 0, then Storage efficiency is 0; S2-5: Merge the CVS files after data cleaning into a single file, retaining the header of the first file, and appending only data rows to subsequent CSV files to generate a merged CVS file containing all samples.

[0060] In this embodiment, the generated sample data is modified in batches using CMG data files. Specifically, the batch modification process consists of the following three stages: File parsing phase: First, the original .dat file content is read, and all well names and DATE commands are extracted using regular expressions. The extracted well names are used to verify the existence of production and injection wells for the specified target optimization, and to locate the position of the last DATE command in the original .dat file, which serves as the insertion point for subsequent new scheduling blocks. Wherein: The regular expression to extract well names is: ^\s*WELL\s+(['\"])([^'\"]+)\1\s*$ ^: Matches the beginning of a string (or each line) (when combined with the re.MULTILINE flag, it matches the beginning of each line).

[0061] \s*: Matches zero or more whitespace characters (space, tab, etc.).

[0062] WELL: Matches the string "WELL" exactly (case sensitive).

[0063] \s+: Matches one or more whitespace characters.

[0064] (['\"]): Matches a single quote ' or a double quote " and captures it as the first group (for subsequent quotations).

[0065] ([^'\"]+): Matches one or more unquoted characters ([^'\"] represents any character except ' and "), and captures the second group (i.e., the hash name).

[0066] \1: Refer to the content matched in the first group (ensure that the quotation marks before and after the well name are consistent; if the previous one is a single quotation mark, then the single quotation mark here should also be matched).

[0067] \s*: Matches zero or more whitespace characters.

[0068] $: Matches the end of a string (or each line) (when combined with the re.MULTILINE flag, it matches the end of each line).

[0069] The regular expression to extract the DATE command is: ^\s*DATE\s+(\d+)\s+(\d+)\s+ ^: Matches the beginning of each line (in combination with the re.MULTILINE flag).

[0070] \s*: Matches zero or more whitespace characters.

[0071] DATE: Matches the string "DATE" exactly (case sensitive).

[0072] \s+: Matches one or more whitespace characters.

[0073] (\d+): Matches one or more digits (years), capturing as the first group.

[0074] \s+: Matches one or more whitespace characters.

[0075] (\d+): Matches one or more digits (months), capturing as the second group.

[0076] \s+: Matches one or more whitespace characters.

[0077] Parameter processing and content generation stage: Obtain production rate, injection rate, etc. from sample data. Concentration and other parameters are used for Concentration calculation and injection-production ratio calculation, among which, Concentration simulation calculation is done by subtracting 1 Concentration was obtained, and the injection-production ratio was calculated using simulation (injection rate). 2) (Production rate) 3) Once obtained, a new scheduling block is generated according to the CMG syntax rules, such as all monthly DATE commands from the start date (July 2024) to the end date (September 2025), PRODUCER and OPERATE rate control statements for production wells, INJECTOR and INCOMP fluid concentration configurations for injection wells (e.g., Concentration and The concentration) and the corresponding OPERATE statement generate a new scheduling block; in this application, Concentration calculations are used for generating new scheduling blocks, and injection-to-production ratios are used for file naming in the subsequent file reorganization phase. File reassembly phase: The original .dat file is truncated to the last DATE command, a newly generated scheduling block is inserted, and the remaining content of the original .dat file is then concatenated to form a complete new file, and the reassembly is performed according to the following criteria: time range + injection rate + injection-to-collection ratio + ... The "concentration" rule generates unique filenames, and the sampling ratio in the filename is replaced with "p" to replace the decimal point, for example, 1.5 is replaced with 1p5. Concentration is presented as an integer percentage, for example 0.3 replaced with 30, thereby transforming these discrete parameters into identifiable file features and compliant CMG command parameters to achieve batch generation.

[0078] After importing the batch-modified CMG data files into the CMG software for simulation, the simulation results are exported as CSV files. These CSV files then require parameter extraction, cleaning, and merging. Specifically, during processing, key parameters, such as injection rate and injection-to-production ratio, are extracted from the filename or first line. Concentration is parsed using regular expressions to match the format of the numerical value. For example, the regular expression could be `CMG_(\d+)-(\d+)_(\d+)([az]*)_(\d+p\d+)-( Next, the data in the CSV file is read, the numeric columns are converted to a standard format and invalid values ​​are handled, and then the data is filtered according to business rules, such as retaining data after July 2024. mole fraction Records were taken at 0.4, and core metrics were calculated. The core metrics were: Storage efficiency, expressed as:

[0079] in, express Storage efficiency, Indicates cumulative injection Number of moles Indicates cumulative output Number of moles; if injected If it is 0, then Storage efficiency is 0; Finally, a merge process is performed to generate a merged file containing all samples, which facilitates subsequent batch analysis.

[0080] S3: Construct a multi-output regression prediction model. Preprocess the exported merged CSV file to generate training, validation, and test sets. Input the training set into the multi-output regression prediction model for training. Validate and test the trained multi-output regression prediction model using the validation and test sets until a fully trained multi-output regression prediction model is obtained. S3 includes: S3-1: Constructing a multi-output regression prediction model based on an improved stacked ensemble algorithm; S3-2: After preprocessing the merged CVS files, divide them into training, validation, and test sets in a ratio of 7:1.5:1.5; S3-3: Input the training set into the multi-output regression prediction model for training, specifically: Based on the input training set, the first... Feature vector of each sample Input into two base models: Random Forest model expression:

[0081] XGBoost model expression:

[0082] Among them, for the training set Feature vector of each sample Random Forest Model Output predicted value vector XGBoost model It will output a vector of predicted values. , In the random forest model Forecast of cumulative output, In the random forest model Forecast of cumulative output, In the random forest model Predicted value of mole fraction; In the XGBoost model Forecast of cumulative output, In the XGBoost model Forecast of cumulative output, In the XGBoost model Predicted value of mole fraction; The prediction results of the two base models are concatenated into a meta-feature vector. The expression is:

[0083] Metamodel Meta-features As input, output the final predicted value. The expression is:

[0084] in, These are the final predicted results for the three dependent variables; express The final forecast of cumulative production, express The final forecast of cumulative production, express The final prediction result of the mole fraction; The meta-model is optimized by minimizing the loss function on the training set, expressed as:

[0085] in, For the first The true value of the t-th dependent variable in a sample, where n is the total number of samples. For the first The predicted value of the t-th dependent variable in a sample. Representing the metamodel; To impose a business scope constraint on the prediction results, the expression is:

[0086] in Let t represent the minimum value of the t-th dependent variable. Let t be the maximum value of the dependent variable. For the first The predicted value of the t-th dependent variable in a sample; S3-4: Based on the training results, call the preset indicators for evaluation, and perform validation and testing based on the validation set and test set until a multi-output regression prediction model that meets the evaluation indicators is output.

[0087] In this embodiment, for the multi-output regression prediction model, the stacked ensemble algorithm is improved to enhance the accuracy of the prediction results. In addition, multi-output regression prediction models including neural networks, random forests, Transformers, and XGBoost can also be used.

[0088] For the training metrics of the multi-output regression prediction model, evaluation metrics such as MSE and RMSE are used for evaluation until the multi-output regression prediction model with the best prediction results is obtained.

[0089] S4: Extract All control parameters for extraction and storage are combined and exported as file data. A trained multi-output regression prediction model is then used to predict the file data. All combinations of control parameters and variables in extraction and storage; In this embodiment, the example of the full combination of variables is as follows: the date range is from August 1, 2024 to January 1, 2034, and the step size is 1 month; the injection rate is set with a starting value of 100,000, an ending value of 300,000, and a step size of 10,000; the injection-to-pickup ratio is set with a starting value of 0.6, an ending value of 1.0, and a step size of 0.1. The concentration inputs are 0.6, 0.7, 0.8, 0.9, and 1. After combining all the above variables, 71,820 combined data points are obtained. These combined data points are exported as a variable combination table. The optimal multi-output regression prediction model is used for prediction. After obtaining the prediction results, data cleaning is performed to handle invalid dates, correct unreasonable prediction values, etc., to obtain the cleaned prediction results.

[0090] S5: Call The algorithm constructs a multi-objective optimization model, and The entire combination of control parameters and variables in extraction and storage is input into a multi-objective optimization model, and the maximization parameter is selected based on the optimization objective. Cumulative output and maximization Storage efficiency, deriving the Pareto optimal solution, as The results of multi-objective optimization of extraction and storage; among which, S5 includes: S5-1: Based on the complete combination of input control parameter variables, extract the injection rate, injection-to-enzyme ratio, and other parameters based on the date index. concentration, Gas mole fraction; S5-2: Call The algorithm constructs a multi-objective optimization model, calling the variable influence coefficient formula to evaluate injection rate, injection-to-enzyme ratio, and other variables. concentration, Influence coefficient analysis of gas mole fraction; S5-3: Injection rate and injection-production ratio based on influence coefficient analysis concentration, In a multi-objective optimization model, the gas mole fraction is used as the optimization target, and the output of the optimization objective is to maximize the gas mole fraction. Cumulative output and maximization Storage efficiency.

[0091] In S5-2, the formula for the influence coefficient of variables is used to apply to the injection rate, injection-to-production ratio, concentration, The influence coefficient analysis of gas mole fraction is as follows: Analysis of the influence coefficient of injection rate:

[0092] in, Indicates the influence coefficient of injection rate. As the injection rate weight, The average injection rate. The standard deviation of the injection rate; Analysis of the influence coefficient of injection-production ratio:

[0093] in, This represents the influence coefficient of the injection-production ratio. For the injection-production ratio weighting, The injection-production ratio is the average. The standard deviation of the injection-production ratio; Concentration influence coefficient analysis includes respectively for Production analysis and Sequestration Analysis: Concentration Production Analysis:

[0094] Concentration Sequestration Analysis:

[0095] in, express Concentration The impact coefficient of output express Concentration The impact coefficient of sealing for Concentration weight, for The average concentration, for Standard deviation of concentration; Analysis of the influence coefficient of gas mole fraction:

[0096] in, express The influence coefficient of gas mole fraction for Mean of mole fraction denoted as the standard deviation of the mole fraction.

[0097] In S5-3, the optimization objective is to maximize the output. Cumulative output and maximization Storage efficiency is specifically as follows: The optimization objective is to maximize The cumulative output expression is:

[0098] Maximize the optimization objective The storage efficiency expression is:

[0099] in, For the final calculation Cumulative production optimization target value, express The baseline value for cumulative production. For the final calculation Storage efficiency optimization target value for Storage efficiency benchmark, Avoid results that are 0 or negative.

[0100] Therefore, the technical solution of this application is as follows: 1. It can achieve fully automated operation, integrating functions such as sampling, batch modification of CMG data files, CSV data processing and merging, multi-output regression prediction modeling and NSGA-II multi-objective optimization. From parameter design to optimization decision-making, there is no need for a lot of manual intervention, which greatly reduces labor costs, significantly improves work efficiency, reduces human operation errors, and ensures the accuracy of results. 2. Supports multi-objective collaborative optimization, generating Pareto optimal frontier solutions through the NSGA-II algorithm, which can... Finding a balance between conflicting objectives such as storage efficiency and cumulative methane production solves the problem that traditional technologies struggle to balance multiple objectives, providing a scientific basis for parameter optimization in actual production; 3. It integrates a variety of advanced algorithms and models, including improved sampling algorithms (ensuring uniform parameter samples and coverage of boundaries) and multiple multi-output regression prediction models (accurately capturing data time-series characteristics and complex nonlinear relationships to improve prediction accuracy). It is suitable for multivariate scenarios in complex reservoir environments, providing diversified technical options for different needs and enhancing the applicability and reliability of the system. 4. Possesses comprehensive data processing capabilities, using CSV data cleaning tools to process invalid data, convert numerical types, and calculate storage efficiency. Combined with the result verification mechanism of multi-output regression models (such as correcting non-negative yields and limiting the range of mole fractions), it ensures data quality and the rationality of prediction results, providing high-quality data support for subsequent optimization decisions. This system can realize gas reservoir injection. Synergistic optimization of gas storage and extraction not only improves the efficiency of gas extraction but also enhances... The energy storage effect provides strong technical support for the sustainable development of the energy industry and helps to achieve a win-win situation for energy extraction and environmental protection.

[0101] like Figure 2 As shown, in another embodiment of this example, a gas reservoir injection system is also included. A multi-objective optimization system for extraction and storage is applied to the above-mentioned gas reservoir injection... Multi-objective optimization methods for extraction and storage include: It includes a sample data generation module, a CMG processing module, a prediction model generation and training module, a variable combination module, and a multi-objective optimization module, among which: Sample data generation module: used to acquire The control parameter variables for extraction and storage are sampled and combined using a preset sampling algorithm to generate sample data. CMG processing module: Used to call the CMG data file batch modification tool, extract all well names and DATE commands from the original .dat file, verify the effectiveness of the target control wells, and locate the position of the last DATE command; read control parameter variables from sample data, and perform injection-production ratio, Concentration calculation; then, based on the CMG syntax rules, a new scheduling block is generated and inserted into the last DATE command position of the original dat file. After generating a new complete CMG data file, it is imported into the CMG software for simulation. The CSV file is exported from the CMG running result file and merged. Prediction Model Generation and Training Module: This module is used to build a multi-output regression prediction model. It preprocesses the exported merged CSV file to generate training, validation, and test sets. The training set is then input into the multi-output regression prediction model for training. The training multi-output regression prediction model is validated and tested using the validation and test sets until the trained multi-output regression prediction model is obtained. Variable combination module: used for extraction All control parameters for extraction and storage are combined and exported as file data. A trained multi-output regression prediction model is then used to predict the file data. All combinations of control parameters and variables in extraction and storage; Multi-objective optimization module: Used to call multi-objective optimization algorithms to build multi-objective optimization models. The entire combination of control parameters and variables in extraction and storage is input into a multi-objective optimization model, and the maximization parameter is selected based on the optimization objective. Cumulative output and maximization Storage efficiency, deriving the Pareto optimal solution, as The results of multi-objective optimization of extraction and storage.

[0102] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A gas reservoir injection A multi-objective optimization method for extraction and storage, characterized by: include: S1: Obtain The control parameter variables for extraction and storage are sampled and combined using a preset sampling algorithm to generate sample data. S2: Use the CMG data file batch modification tool to extract all well names and DATE commands from the original .dat file, verify the effectiveness of the target control wells, and locate the position of the last DATE command; read the control parameter variables from the sample data and perform injection-production ratio, Concentration calculation; then, based on the CMG syntax rules, a new scheduling block is generated and inserted into the last DATE command position of the original dat file. After generating a new complete CMG data file, it is imported into the CMG software for simulation. The CSV file is exported from the CMG running result file and merged. S3: Construct a multi-output regression prediction model. After preprocessing the exported merged CSV file, generate a training set, a validation set, and a test set. Input the training set into the multi-output regression prediction model for training. Validate and test the trained multi-output regression prediction model using the validation set and the test set until the trained multi-output regression prediction model is obtained. S4: Extract All control parameters for extraction and storage are combined and exported as file data. A trained multi-output regression prediction model is then used to predict the file data. All combinations of control parameters and variables in extraction and storage; S5: Use a multi-objective optimization algorithm to construct a multi-objective optimization model. The entire combination of control parameters and variables in extraction and storage is input into a multi-objective optimization model, and the maximization parameter is selected based on the optimization objective. Cumulative output and maximization Storage efficiency, deriving the Pareto optimal solution, as The results of multi-objective optimization of extraction and storage.

2. A gas reservoir injection system according to claim 1 A multi-objective optimization method for extraction and storage, characterized by: S1 includes: S1-1: For the acquisition The control parameters for extraction and storage are divided into equally probable sub-intervals according to their variable types, including discrete variables and continuous variables. S1-2: Allocate sampling quotas to all split points of each variable according to the total number of samples, and establish a sampling index list for each variable; S1-3: Randomly shuffle the sampling index list of each variable, randomly select a position from the random index list of each variable, extract the corresponding split point value to form a sampling combination, until the preset sample size is reached; S1-4: Use the improved Antlion optimization algorithm to extract the optimal boundary reinforcement combination of the extreme values ​​of each variable, and after performing a repeatability check on the sampling combination generated in S1-3, combine the sampling combination to generate the final variable sampling combination result.

3. A gas reservoir injection system according to claim 2 A multi-objective optimization method for extraction and storage, characterized by: The optimal boundary reinforcement combination for extracting the extreme values ​​of each variable in S1-4 by calling the improved antlion optimization algorithm is as follows: Obtain each control parameter variable extreme values and build Class boundary target vector; Initialize the antlion colony and the ant colony, and... The class boundary target vector is added to the antlion population; The fitness function is constructed based on proximity to the boundary and engineering feasibility objectives, and its expression is: in, For boundary-leading terms, This is a constraint on the feasibility of the project. For individual ants, Let k be the boundary target vector of the kth class. The distance is defined as Hamming distance, and the engineering constraints are preset based on the physical laws of gas reservoirs. For the penalty function; and The weighting coefficients are adjusted according to optimization requirements. and All are greater than 0, and their sum is 1; Project feasibility constraints Other representations in the text indicate that if x has one constraint that does not satisfy all engineering constraints; The ant's movement direction is adjusted by a boundary attraction factor to perform boundary-guided walking, expressed as follows: in, Indicates in In the iteration, the ant's new position on the i-th control parameter variable, Indicates the first In the iteration, the ant's current position on the i-th control parameter variable is... Indicates the first The step size of the random walk in the iteration. For the first Generated in generation iteration Uniformly distributed random numbers Let i be the current value of the ant in the t-th generation on the i-th variable. For the boundary attraction term, Let be the i-th extreme value of the boundary target vector that the ant individual is currently closest to; The attraction intensity coefficient, ; For individual ants The boundary target vector corresponding to the current minimum Hamming distance. Let x be the ant's current position and the optimal boundary target vector. Hamming distance, , which is the distance decay factor, representing that the closer an ant is to the boundary, the stronger the attraction, and the farther away it is, the weaker the attraction. After each iteration, based on the boundary guiding term Update the antlion trap radius, mark the individual with the best fitness in the population as an elite antlion, and force the retention of at least one individual that meets the criteria. An elite individual with an extreme value for a single variable; After the maximum number of iterations is reached, the best individuals are extracted from the final elite antlion population as the boundary reinforcement combination, and a repeatability judgment is performed on the sampled combination generated in S1-3. Finally, the final sample data of the random sampled combination and the deduplicated boundary reinforcement combination are output.

4. A gas reservoir injection system according to claim 1 A multi-objective optimization method for extraction and storage, characterized by: S2 includes: S2-1: Call the CMG data file batch modification tool, read the contents of the original dat file, extract all well names and DATE commands through regular expressions, verify the existence of wells with control parameter variables by using the extracted well names, and locate the position of the last DATE command in the original dat file; S2-2: Read the regulation parameter variables from the sample data and perform... Concentration calculation, injection-production ratio calculation, and according to CMG syntax rules, combine all monthly DATE commands for the start and end dates of injection wells, PRODUCER and OPERATE rate control statements for production wells, INJECTOR and INCOMP fluid concentration configurations for injection wells, and corresponding OPERATE statements to generate a new scheduling block; S2-3: Truncate the original dat file to the last DATE command, insert the generated new scheduling block, then concatenate the remaining content of the original dat file to form a complete new file, and generate unique filenames according to the preset file naming rules to batch generate CMG data files and save them to the specified directory.

5. A gas reservoir injection system according to claim 4 A multi-objective optimization method for extraction and storage, characterized by: S2 further includes: S2-4: After importing the batch-modified CMG data files into the CMG software for simulation, export the simulation results as a CSV file. Perform data cleaning on the exported CSV file, including using preset regular expression matching formats to match filenames and convert them to a standard format, handling invalid values, filtering data from the CSV file according to business rules, and calculating... The core metrics of storage efficiency; among them The storage efficiency expression is: in, express Storage efficiency, Indicates cumulative injection Number of moles Indicates cumulative output Number of moles; if injected If it is 0, then Storage efficiency is 0; S2-5: Merge the CVS files after data cleaning into a single file, retaining the header of the first file, and appending only data rows to subsequent CSV files to generate a merged CVS file containing all samples.

6. A gas reservoir injection system according to claim 1 A multi-objective optimization method for extraction and storage, characterized by: S3 includes: S3-1: Constructing a multi-output regression prediction model based on an improved stacked ensemble algorithm; S3-2: After preprocessing the merged CVS files, divide them into training, validation, and test sets in a ratio of 7:1.5:1.5; S3-3: Input the training set into the multi-output regression prediction model for training, specifically: Based on the input training set, the first... Feature vector of each sample Input into two base models: Random Forest model expression: XGBoost model expression: Among them, for the training set Feature vector of each sample Random Forest Model Output predicted value vector XGBoost model It will output a vector of predicted values. , In the random forest model Forecast of cumulative output, In the random forest model Forecast of cumulative output, In the random forest model Predicted value of mole fraction; In the XGBoost model Forecast of cumulative output, In the XGBoost model Forecast of cumulative output, In the XGBoost model Predicted value of mole fraction; The prediction results of the two base models are concatenated into a meta-feature vector. The expression is: Metamodel Meta-features As input, output the final predicted value. The expression is: in, These are the final predicted results for the three dependent variables; express The final forecast of cumulative production, express The final forecast of cumulative production, express The final prediction result of the mole fraction; The meta-model is optimized by minimizing the loss function on the training set, expressed as: in, For the first The true value of the t-th dependent variable in a sample, where n is the total number of samples. For the first The predicted value of the t-th dependent variable in a sample. Representing the metamodel; To impose a business scope constraint on the prediction results, the expression is: in Let t represent the minimum value of the t-th dependent variable. Let t be the maximum value of the dependent variable. For the first The predicted value of the t-th dependent variable in a sample. This represents the forecast value after adjustment for business scope constraints; S3-4: Based on the training results, call the preset indicators for evaluation, and perform validation and testing based on the validation set and test set until a multi-output regression prediction model that meets the evaluation indicators is output.

7. A gas reservoir injection system according to claim 1 A multi-objective optimization method for extraction and storage, characterized by: S5 includes: S5-1: Based on the complete combination of input control parameter variables, extract the injection rate, injection-to-enzyme ratio, and other parameters based on the date index. concentration, Gas mole fraction; S5-2: Call The algorithm constructs a multi-objective optimization model, calling the variable influence coefficient formula to evaluate injection rate, injection-to-enzyme ratio, and other variables. concentration, Influence coefficient analysis of gas mole fraction; S5-3: Injection rate and injection-production ratio based on influence coefficient analysis concentration, In a multi-objective optimization model, the gas mole fraction is used as the optimization target, and the output of the optimization objective is to maximize the gas mole fraction. Cumulative output and maximization Storage efficiency.

8. A gas reservoir injection system according to claim 7 A multi-objective optimization method for extraction and storage, characterized by: The formula for the influence coefficient of variables in S5-2 is used to apply to injection rate, injection-to-enzyme ratio, concentration, The influence coefficient analysis of gas mole fraction is as follows: Analysis of the influence coefficient of injection rate: in, Indicates the influence coefficient of injection rate. As the injection rate weight, The average injection rate. The standard deviation of the injection rate; Analysis of the influence coefficient of injection-production ratio: in, This represents the influence coefficient of the injection-production ratio. For the injection-production ratio weighting, The injection-production ratio is the average. The standard deviation of the injection-production ratio; Concentration influence coefficient analysis includes respectively for Production analysis and Sequestration Analysis: Concentration Production Analysis: Concentration Sequestration Analysis: in, express Concentration The impact coefficient of output express Concentration The impact coefficient of sealing for Concentration weight, for The average concentration, for Standard deviation of concentration; Analysis of the influence coefficient of gas mole fraction: in, express The influence coefficient of gas mole fraction for Mean of mole fraction denoted as the standard deviation of the mole fraction.

9. A gas reservoir injection system according to claim 8 A multi-objective optimization method for extraction and storage, characterized by: The optimization objective output in S5-3 is to maximize... Cumulative output and maximization Storage efficiency is specifically as follows: The optimization objective is to maximize The cumulative output expression is: Maximize the optimization objective The storage efficiency expression is: in, For the final calculation Cumulative production optimization target value, express The baseline value for cumulative production. For the final calculation Storage efficiency optimization target value for Storage efficiency benchmark, Avoid results that are 0 or negative.

10. A gas reservoir injection A multi-objective optimization system for extraction and storage, applying the gas reservoir injection method described in any one of claims 1-9 above. A multi-objective optimization method for extraction and storage, characterized by: It includes a sample data generation module, a CMG processing module, a prediction model generation and training module, a variable combination module, and a multi-objective optimization module, among which: Sample data generation module: used to acquire The control parameter variables for extraction and storage are sampled and combined using a preset sampling algorithm to generate sample data. CMG processing module: Used to call the CMG data file batch modification tool, extract all well names and DATE commands from the original .dat file, verify the effectiveness of the target control wells, and locate the position of the last DATE command; read control parameter variables from sample data, and perform injection-production ratio, Concentration calculation; then, based on the CMG syntax rules, a new scheduling block is generated and inserted into the last DATE command position of the original dat file. After generating a new complete CMG data file, it is imported into the CMG software for simulation. The CSV file is exported from the CMG running result file and merged. Prediction Model Generation and Training Module: This module is used to build a multi-output regression prediction model. It preprocesses the exported merged CSV file to generate training, validation, and test sets. The training set is then input into the multi-output regression prediction model for training. The training multi-output regression prediction model is validated and tested using the validation and test sets until the trained multi-output regression prediction model is obtained. Variable combination module: used for extraction All control parameters for extraction and storage are combined and exported as file data. A trained multi-output regression prediction model is then used to predict the file data. All combinations of control parameters and variables in extraction and storage; Multi-objective optimization module: Used to call multi-objective optimization algorithms to build multi-objective optimization models. The entire combination of control parameters and variables in extraction and storage is input into a multi-objective optimization model, and the maximization parameter is selected based on the optimization objective. Cumulative output and maximization Storage efficiency, deriving the Pareto optimal solution, as The results of multi-objective optimization of extraction and storage.

Images