A Method and System for Constructing an Integrated Operation Model for Gas Storage Based on Digital Twins

Abstract

This invention discloses a method and system for constructing an integrated operation model for gas storage facilities based on digital twins, relating to the field of integrated operation model construction for gas storage facilities. The method includes: constructing single-well IPR and VLP curves; using a combination of forward and reverse pressure regulation to output a set of pressure values ​​for each node after pressure stabilization, forming a preliminary pressure steady-state scheme; constructing an economic value assessment index function to evaluate the total daily operating cost of the system under a given gas production rate; constructing a comprehensive optimization model based on a neural network, outputting benchmark values ​​for injection and production rates of all wells and target values ​​for each node; and verifying the stability of the comprehensive optimization model based on the neural network based on the output benchmark values ​​for injection and production rates of all wells and target values ​​for each node. The advantages of this invention are: it deeply integrates the physical laws of gas storage facility injection and production, system pressure balance, and economic objectives, achieving a rapid and reliable closed loop from pressure steady-state optimization to economically optimal decision-making.

Description

Technical Field

[0001] This invention relates to the field of building an integrated operation model for gas storage facilities, specifically to a method and system for building an integrated operation model for gas storage facilities based on digital twins. Background Technology

[0002] With the accelerated transformation of the global energy structure, natural gas, as a clean and efficient transitional energy source, is increasingly in demand for peak shaving and supply guarantee. Underground gas storage facilities, as key buffers and regulating facilities in the natural gas supply chain, undertake multiple functions such as seasonal peak shaving, emergency reserves, and strategic security. The core of gas storage operation lies in the injection-production cycle, that is, injecting surplus natural gas in summer and producing it in winter to meet peak demand. This process involves a complex multiphase flow system of gas reservoir, wellbore, and pipeline network, and its operation and control face multiple challenges, such as dynamic changes in reservoir formation pressure, and the influence of reservoir properties, well structure, fluid properties, and other factors on single-well productivity; energy losses due to gravity and friction during wellbore lifting; and the need for the surface pipeline network to meet the pressure and flow demands of downstream users. How to scientifically allocate the injection and production volume of each well, and maximize economic benefits (reduce energy consumption and processing costs) while ensuring the safe and stable system pressure (avoiding overpressure or underpressure leading to equipment damage or gas supply interruption), is the core objective of the refined and intelligent operation of gas storage facilities. Traditional scheduling methods, which rely primarily on human experience and simplified steady-state models, struggle to accurately depict the dynamic coupling relationships within a system and are unable to quickly find optimal solutions under varying operating conditions. This has become a bottleneck restricting the efficient and safe operation of gas storage facilities. Against this backdrop, the concept of "digital twins," which integrates high-fidelity physical simulation, intelligent optimization algorithms, and visual interactive technologies, offers a new approach to constructing a next-generation integrated operation model for gas storage facilities.

[0003] Most existing technologies treat gas reservoir seepage, wellbore flow, and pipeline transmission as independent links, using simplified models for analysis (such as production allocation based solely on IPR curves or design using VLP curves alone). They lack a mechanism for real-time coupling with surface pipeline hydraulic models. This fragmentation makes it impossible to accurately simulate the complete pressure transmission chain from the reservoir to the user end, and even more impossible to effectively analyze the dynamic balance relationship of pressure at each node of the system, causing optimization schemes to fail in actual coupled systems. Secondly, existing methods lack a framework for deeply integrating the physical rules of gas storage injection and production with the powerful fitting capabilities of neural networks. Therefore, it is difficult to construct an intelligent model that both conforms to physical laws and can quickly find the optimal solution. Ultimately, this results in the inability to reliably and efficiently conduct comprehensive evaluation and optimization of the stability and economy of system pressure operation. Summary of the Invention

[0004] To address the aforementioned technical issues, this paper provides a method and system for constructing an integrated operation model of a gas storage facility based on digital twins. This technical solution solves the problems mentioned in the background art, namely, the lack of a mechanism for real-time coupling with the hydraulic model of the ground pipeline network and the lack of a framework for deeply integrating the physical rules of gas storage injection and production with the powerful fitting ability of neural networks.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for constructing an integrated operation model for gas storage facilities based on digital twins, comprising:

[0007] Based on the current average formation pressure of a single well, calculate its corresponding bottom hole flowing pressure and wellhead pressure, and construct the IPR and VLP curves for the single well;

[0008] Based on the IPR curve and VLP curve family, a combination of forward and reverse pressure regulation is used to output the set of pressure values ​​of each node after pressure stabilization, forming a preliminary pressure steady-state scheme.

[0009] Obtain intermediate pressure data during the positive pressure regulation process, construct an economic value assessment index function, and assess the total daily operating cost of the system under a given gas production rate;

[0010] With the goals of pressure reduction compliance and economic optimization, a comprehensive optimization model based on neural networks is constructed to output the baseline values ​​of injection and recovery volumes for all wells and the target values ​​for each node.

[0011] Based on the output of all well injection and production baseline values ​​and the target values ​​of each node, the stability of the neural network-based comprehensive optimization model is verified by combining forward and reverse pressure regulation.

[0012] Preferably, the step of calculating the corresponding bottomhole flowing pressure and wellhead pressure based on the current formation average pressure of a single well, and constructing the single-well IPR and VLP curves specifically includes:

[0013] Based on the formation pressure monitoring data of the gas storage facility, the current average formation pressure of each individual well is obtained;

[0014] Obtain the wellbore structural parameters, current fluid properties, and preset bottomhole flowing pressure sequence for each single well;

[0015] The structural parameters include at least: tubing inner diameter, vertical depth, and well inclination angle; the current fluid properties include at least: gas-liquid mixture density, multiphase flow friction coefficient, and mixture velocity.

[0016] Based on the current formation average pressure and the preset bottom hole flowing pressure sequence, and based on Darcy's law and gas seepage theory, the gas production sequence of each single well is calculated, and the IPR curve of each single well is constructed.

[0017] Based on the calculated gas production sequence, for each value in the preset bottom hole flowing pressure sequence, the corresponding wellhead pressure is calculated based on the multiphase pipe flow mechanical energy conservation equation, and a family of VLP curves for a single well is constructed.

[0018] Preferably, the step of using a combination of forward and reverse pressure regulation based on the IPR curve and VLP curve family to output the set of pressure values ​​at each node after pressure stabilization, forming a preliminary pressure steady-state scheme, specifically includes:

[0019] Based on the geological data of the gas storage facility, the well structure data of each individual well, and historical production data, the initial given gas production rate for each individual well is set.

[0020] Using the given gas production rate of each individual well as input, and combining the IPR curve and VLP curve family of each individual well, the coordinated bottom hole flowing pressure and coordinated wellhead pressure of each well under the current gas production rate are determined.

[0021] Starting with the coordinated wellhead pressure of each individual well, and based on the multiphase horizontal pipe flow pressure drop formula, segmented hydraulic calculations are performed sequentially along the fluid flow direction until the system terminal pressure for positive pressure regulation is obtained.

[0022] Starting from the set target pressure at the end of the system, and based on the multiphase horizontal pipe flow pressure drop formula, reverse hydraulic calculation is performed to calculate the theoretical pressure values ​​corresponding to each upstream node in order to meet the target pressure at the end.

[0023] Based on the system design pressure and instrument accuracy, a set of preset allowable pressure deviation thresholds for each node is established.

[0024] If the absolute value of the deviation between the pressure of each node obtained by forward pressure regulation and the theoretical pressure value of the corresponding node obtained by reverse pressure regulation is greater than the corresponding deviation threshold, it is determined that the system pressure has not reached equilibrium. The gas production of each single well is dynamically adjusted through feedback control algorithm. Based on the new gas production allocation after adjustment, the process of determining coordinated pressure, forward and reverse pressure regulation is re-executed until the pressure deviation of all key nodes meets the threshold requirement. If not, it is determined that the system pressure has reached equilibrium and the theoretical pressure value of each node is directly output.

[0025] Record the final gas production distribution of each well after iterative convergence, coordinate the bottom hole flowing pressure, coordinate the wellhead pressure, and the theoretical pressure values ​​of each node to form a preliminary pressure steady-state scheme and updated IPR and VLP curves of each well.

[0026] Preferably, the step of acquiring intermediate pressure data during the positive pressure regulation process, constructing an economic value assessment index function, and assessing the total daily operating cost of the system under a given gas production volume specifically includes:

[0027] Based on the forward pressure regulation process, intermediate pressure data during the forward pressure regulation process are obtained, including: formation pressure, bottom hole flowing pressure, and wellhead pressure;

[0028] Based on intermediate pressure data, an economic value assessment index function is constructed.

[0029] The economic value assessment index function consists of production cost, processing cost, compressor energy consumption cost, and penalty for constraint violation;

[0030] Based on the economic value assessment index function, the total daily operating cost of the system is calculated under a given gas production volume.

[0031] Preferably, the construction of a comprehensive optimization model based on neural networks, with the objectives of pressure reduction compliance and economic optimization, and the output of all well injection-production baseline values ​​and target values ​​for each node, specifically includes:

[0032] In the offline phase, based on the constructed single-well IPR and VLP curves, and using a combination of forward and reverse pressure regulation, batch numerical simulations were conducted using the controlled variable method.

[0033] The data after numerical simulation is preprocessed and a training sample set is constructed. The preprocessing includes at least: filtering, interpolation, one-hot encoding of non-numerical features, and normalization.

[0034] The training sample set is divided into a training set, a validation set, and a test set according to an 8:1:1 allocation method.

[0035] The sample set input features include at least: the current formation average pressure of each single well, well structure parameters, current fluid properties, preset bottom hole flow pressure sequence, pipeline parameters, target total gas production of the gas storage facility, and target pressure at the end of the system.

[0036] From the sample set, we select the steady-state pressure scheme that satisfies the pressure drop constraint and minimizes the total daily operating cost of the system. We then extract the pressure values ​​of each node and the gas production of a single well corresponding to the scheme, and use them together as the output labels of the sample set.

[0037] With the goal of minimizing the deviation between the pressure value of the output label and the node pressure value predicted by the model, a pressure drop compliance loss function is constructed based on the mean square error formula.

[0038] With the goal of minimizing the deviation between the output gas production and the model-predicted single-well gas production, an economically optimal loss function is constructed based on the mean square error formula.

[0039] Based on the pressure reduction compliance loss function and the economic optimal loss function, the total loss function is obtained by weighted summation, and its weights are obtained through cross-validation of the validation set.

[0040] Based on a multilayer perceptron network architecture, the Adam optimizer is used to train the system with the goal of minimizing the total loss function using a training set.

[0041] After training is completed, the model parameters that perform best on the validation set during the entire training process are saved as the final comprehensive optimization model based on the neural network.

[0042] Based on the trained neural network-based comprehensive optimization model, the system outputs the baseline values ​​of injection and recovery volumes for all wells and the target values ​​for each node by inputting the actual collected input features.

[0043] Furthermore, this solution proposes a system for constructing an integrated operation model of gas storage facilities based on digital twins, used to implement the aforementioned method for constructing an integrated operation model of gas storage facilities based on digital twins, including:

[0044] The physical model module is used to calculate the corresponding bottomhole flowing pressure and wellhead pressure based on the current formation average pressure of a single well, and to construct the IPR and VLP curves of the single well; based on the IPR curve and VLP curve family, it outputs the set of pressure values ​​of each node after pressure stabilization by combining forward and reverse pressure regulation, forming a preliminary pressure steady-state scheme; it obtains the intermediate pressure data during the forward pressure regulation process, constructs an economic value assessment index function, and assesses the total daily operating cost of the system under a given gas production rate;

[0045] The integrated model and verification module is used to construct a neural network-based integrated optimization model with the goals of pressure reduction compliance and economic optimization, and output the injection and recovery volume benchmark values ​​of all wells and the target values ​​of each node; based on the output of all well injection and recovery volume benchmark values ​​and the target values ​​of each node, the stability of the neural network-based integrated optimization model is verified by combining positive and negative pressure regulation.

[0046] Preferably, the physical model module includes:

[0047] The IPR and VLP curve unit is used to calculate the corresponding bottom hole flowing pressure and wellhead pressure based on the current formation average pressure of a single well, and to construct the IPR and VLP curves of a single well.

[0048] The positive and negative pressure regulating unit is used to output a set of pressure values ​​of each node after stabilization by combining positive and negative pressure regulation according to the IPR curve and VLP curve family, forming a preliminary pressure steady state scheme.

[0049] An economic evaluation unit is used to acquire intermediate pressure data during the positive pressure regulation process, construct an economic value evaluation index function, and evaluate the total daily operating cost of the system under a given gas production volume.

[0050] Preferably, the integrated model and verification module include:

[0051] The integrated model unit is used to construct a comprehensive optimization model based on neural networks with the goals of pressure reduction compliance and economic optimization, and outputs the baseline values ​​of injection and recovery volume for all wells and the target values ​​for each node.

[0052] The model verification unit is used to verify the stability of the neural network-based comprehensive optimization model based on the output of all well injection and production benchmark values ​​and the target values ​​of each node, using a combination of forward and reverse pressure regulation.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] This invention proposes a method for constructing an integrated operation model for gas storage facilities based on digital twins. By constructing single-well IPR and VLP curves, a forward and reverse pressure regulation coupling mechanism, and an economic value assessment function, an integrated decision-making system of "physical simulation—intelligent optimization—closed-loop verification" is built, achieving simultaneous optimization of system pressure dynamic balance and economic optimality. This scheme ensures that the pressure at each node strictly meets physical constraints and pipeline network requirements throughout the entire system through bidirectional pressure regulation iteration, fundamentally guaranteeing the physical feasibility and system stability of the operation scheme. Furthermore, a comprehensive evaluation model based on neural network training is introduced, which can quickly output well injection-production allocation and node pressure setting schemes that approximate the global optimum, solving the problem of excessively long calculation times in traditional high-precision simulation optimization. First, it addresses the difficulty in supporting real-time scheduling. Second, this solution constructs a comprehensive economic value function that includes production costs, processing costs, compressor energy consumption, and safety penalties, deeply embedding economic objectives into the optimization process. This overcomes the shortcomings of existing technologies, which often sacrifice economic efficiency for pressure balance or vice versa, due to a single optimization objective. Furthermore, a model verification closed loop based on physical simulation is designed, and the neural network output scheme is re-substituted into the high-fidelity twin system for pressure balance and economic verification. This achieves a reliable enhancement mechanism of "intelligent recommendation - physical verification - feedback correction," which not only leverages the efficient optimization capabilities of the neural network but also eliminates unreasonable or physically incompatible results, thereby improving decision-making efficiency while ensuring the safety and reliability of the solution. Attached Figure Description

[0055] Figure 1 This is a flowchart of the method for constructing an integrated operation model of a gas storage facility based on digital twins according to the present invention.

[0056] Figure 2 This is a flowchart illustrating the construction of single-well IPR and VLP curves according to the present invention.

[0057] Figure 3This invention employs a combination of forward and reverse pressure regulation to output a set of pressure values ​​for each node after pressure stabilization, as shown in the flowchart below.

[0058] Figure 4 The flowchart for constructing a comprehensive optimization model based on neural networks in this invention outputs the baseline values ​​of injection and recovery volumes for all wells and the target values ​​for each node. Detailed Implementation

[0059] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0060] Reference Figure 1 As shown, a method for constructing an integrated operation model for gas storage facilities based on digital twins includes:

[0061] Based on the current average formation pressure of a single well, calculate its corresponding bottom hole flowing pressure and wellhead pressure, and construct the IPR and VLP curves for the single well;

[0062] Based on the IPR curve and VLP curve family, a combination of forward and reverse pressure regulation is used to output the set of pressure values ​​of each node after pressure stabilization, forming a preliminary pressure steady-state scheme.

[0063] Obtain intermediate pressure data during the positive pressure regulation process, construct an economic value assessment index function, and assess the total daily operating cost of the system under a given gas production rate;

[0064] With the goals of pressure reduction compliance and economic optimization, a comprehensive optimization model based on neural networks is constructed to output the baseline values ​​of injection and recovery volumes for all wells and the target values ​​for each node.

[0065] Based on the output of all well injection and production baseline values ​​and the target values ​​of each node, the stability of the neural network-based comprehensive optimization model is verified by combining forward and reverse pressure regulation.

[0066] To achieve safe, stable, economical, and intelligent integrated operation of a gas storage facility, it is first necessary to establish an accurate physical characterization of the production behavior of individual wells. Therefore, this scheme, based on the current average formation pressure of a single well, and combining Darcy's seepage theory and the conservation equation of mechanical energy for multiphase pipe flow, calculates the bottom-hole flowing pressure and wellhead pressure respectively, constructing an IPR curve characterizing the formation's gas supply capacity and a VLP curve family characterizing the wellbore's transport capacity. This provides an accurate foundation of single-well physical characteristics for subsequent system-level analysis and optimization, avoiding global scheme failure due to distortion of the single-well model. Secondly, based on clarifying the physical characteristics of individual wells, it is necessary to solve the problem of pressure coordination and balance throughout the system. This scheme, based on the constructed single-well IPR and VLP curve families, addresses this issue. This method employs a combination of forward and reverse pressure regulation for system-level iterative optimization. Forward calculations yield the "upper limit of reservoir pressure supply capacity," while reverse calculations yield the "lower limit of pipeline demand pressure." The gas production of each well is dynamically adjusted based on node pressure deviations until the pressure at all key nodes in the system reaches dynamic equilibrium. This outputs a physically feasible preliminary pressure steady-state scheme that satisfies pipeline constraints, addressing the limitation of traditional fragmented analysis in ensuring global system pressure matching. However, a scheme that only satisfies pressure balance may not be economically optimal. To maximize operational efficiency, this scheme introduces an economic dimension based on pressure balance, taking into account factors such as formation pressure, bottomhole flowing pressure, and wellhead pressure obtained during forward pressure regulation. By using key intermediate data, an economic value assessment index function is constructed that integrates production costs, processing costs, compressor energy consumption costs, and safety constraint penalties. This function can quantitatively evaluate the total daily operating cost of the system under any given gas production allocation scheme, providing a clear economic target and quantitative benchmark for subsequent optimization. In summary, to quickly and accurately find the globally optimal solution that simultaneously satisfies pressure balance and economic optimization in a complex feasible solution space, this scheme constructs a comprehensive optimization model based on neural networks. This model uses high-quality data generated from the aforementioned physical simulation as training samples to learn the complex mapping relationship from system operating conditions (such as target total gas volume and terminal pressure) to optimal gas production allocation and node pressure distribution. The trained model can quickly output... By balancing pressure reduction compliance and economic optimization, the baseline values ​​for injection and production volumes of all wells and the target values ​​for each node have been improved, overcoming the bottlenecks of low computational efficiency and difficulty in online application of traditional optimization methods. At the same time, based on the recommended scheme output by the neural network model, the scheme is re-substituted into a high-fidelity physical simulation system based on IPR / VLP curves and the principle of forward and reverse pressure regulation for closed-loop verification. By verifying whether the system pressure can quickly converge to equilibrium, whether the economic cost is close to the optimum, and whether all safety constraints are strictly met, the stability, feasibility, and safety of the model output scheme are ultimately verified, forming a reliable decision-making closed loop of "intelligent rapid optimization - physical high-fidelity verification". This ultimately ensures the high credibility and effectiveness of the integrated operation model of gas storage in practical applications.

[0067] Reference Figure 2 As shown, the construction of single-well IPR and VLP curves specifically includes:

[0068] Based on the formation pressure monitoring data of the gas storage facility, the current average formation pressure of each individual well is obtained;

[0069] Obtain the wellbore structural parameters, current fluid properties, and preset bottomhole flowing pressure sequence for each single well;

[0070] The structural parameters include at least: tubing inner diameter, vertical depth, and well inclination angle; the current fluid properties include at least: gas-liquid mixture density, multiphase flow friction coefficient, and mixture velocity.

[0071] Based on the current formation average pressure and the preset bottom hole flowing pressure sequence, and based on Darcy's law and gas seepage theory, the gas production sequence of each single well is calculated, and the IPR curve of each single well is constructed.

[0072] Based on the calculated gas production sequence, for each value in the preset bottom hole flowing pressure sequence, the corresponding wellhead pressure is calculated based on the multiphase pipe flow mechanical energy conservation equation, and a family of VLP curves for a single well is constructed.

[0073] This can be explained by the fact that pressure is the core driving force for gas flow in a single well of a gas storage facility (the pressure difference between the formation and the bottom of the well, and between the bottom of the well and the wellhead, respectively drives gas from the reservoir into the wellbore and from the wellbore to the wellhead). Pressure parameters are easily monitored on-site. Therefore, this scheme constructs single-well IPR and VLP curves by combining pressure as the starting point with Darcy's law and the multiphase pipe flow mechanical energy conservation equation. The IPR and VLP curves are crucial for the refined operation and control of the gas storage facility. The IPR curve characterizes the gas supply capacity from the formation to the wellbore, clarifies the production capacity boundary and variation law under different formation pressures, and provides a basis for single-well production capacity evaluation and multi-well production allocation. The VLP curve characterizes the flow characteristics of the fluid within the wellbore, quantifies the influence of friction, fluid properties, etc., on the flow, establishes the correlation between bottom-hole flowing pressure and wellhead pressure, and provides support for process parameter design and flow bottleneck identification. The coupling of these two curves can solve the relationship between formation energy supply and wellbore energy transmission. Matched and coordinated gas production points provide core quantitative basis for optimizing single-well operating procedures, controlling injection-production balance, and regulating overall operation. Among them, the preset bottom-hole flowing pressure sequence is the core input data for constructing the single-well IPR and VLP curves. Its value range needs to be determined in conjunction with the single-well production boundary: specifically, starting from the current formation average pressure (the initial energy benchmark for single-well production), it gradually decreases to the abandonment pressure of the single well (the minimum economic production pressure set according to the gas storage development plan, i.e., the bottom-hole pressure when the single-well production revenue cannot cover the cost). By selecting several sampling points within this range (such as dividing according to equal pressure difference), a continuous bottom-hole flowing pressure sequence is formed. The setting of this sequence ensures that the IPR curve can cover the entire production range of the single well (from maximum production capacity to abandoned production capacity), and also provides the VLP curve with pressure input that matches the actual production conditions. It is the basis for the complete characterization of the flow characteristics of a single well by the family of curves.

[0074] The expression for calculating the bottom hole flowing pressure of each single well is as follows:

[0075] ,

[0076] In the formula, For the first The current average formation pressure of the well. For the first Bottom-hole flowing pressure of a well The laminar flow coefficient is a comprehensive resistance term obtained by fitting historical well test data from a single well when describing linear seepage based on Darcy's law. The turbulence coefficient is an additional resistance term obtained by fitting historical well test data from a single well when Darcy's law is extended to nonlinear seepage (turbulence). For the first Gas production per well;

[0077] Through a pre-set bottom hole pressure sequence The corresponding gas production sequence can be calculated using this expression. In the calculation process of generating the IPR curve, the bottom hole flowing pressure is used as the input independent variable, and the corresponding gas production is obtained by solving the expression, which is then used as the output dependent variable, thereby generating the IPR curve data set for that well. ,in, The set number of bottom hole flowing pressure sampling points;

[0078] The expression for calculating the wellhead pressure of each single well is as follows:

[0079] ,

[0080] In the formula, For the first Wellhead pressure of the well Let be the variable of integration, and let the integration range begin from the bottom of the well ( ) to the wellhead ( ), For the first Vertical depth of the well For the first Density of the gas-liquid mixture in the well. It is the acceleration due to gravity. For the first Multiphase flow friction coefficient of wellhead For the first The flow rate of the mixture in the well. For the first The inner diameter of the tubing string in the wellhead;

[0081] The density expression for the gas-liquid mixture is:

[0082] ,

[0083] In the formula, For the first Liquid phase density of the well. For the first The gas phase density of the well. For the first The liquid holdup of a well, which is the volume fraction of the liquid phase on the cross section of the flowing tubing, is a function of the gas production rate, tubing diameter, inclination angle, and fluid properties, and needs to be determined by a multiphase flow model (such as Hagedorn-Brown).

[0084] The expression for the flow rate of the mixture is:

[0085] ,

[0086] In the formula, For the first The cross-sectional area of ​​the tubing string in the wellhead;

[0087] This expression is based on the multiphase pipe flow mechanical energy conservation equation, obtained by integrating along the vertical depth of the wellbore, and is derived from the calculated gas production sequence. According to each gas production A corresponding set can be determined. , Iterate through the preset bottom hole flowing pressure sequence For each of them Based on the wellhead pressure expression for each individual well, the corresponding wellhead pressure is obtained. This allows for the generation of that gas production. The following VLP curve data set: By selecting multiple different gas production values ​​and repeating the above calculation process, a family of VLP curves with gas production as a parameter can be obtained, which can be used to systematically analyze the wellbore flow characteristics under different production conditions.

[0088] By constructing VLP curves for different given gas production values, a family of VLP curves is formed, which fully characterizes the flow characteristics of the wellbore under different production intensities. In the subsequent system optimization process, the corresponding VLP curve will be selected based on the initial given gas production value to participate in the calculation, and the overall pressure balance and optimization of the system will be achieved by iteratively updating the gas production value.

[0089] Reference Figure 3 As shown, the method of combining forward and reverse pressure regulation to output the set of pressure values ​​for each node after pressure stabilization specifically includes:

[0090] Based on the geological data of the gas storage facility, the well structure data of each individual well, and historical production data, the initial given gas production rate for each individual well is set.

[0091] Using the given gas production rate of each individual well as input, and combining the IPR curve and VLP curve family of each individual well, the coordinated bottom hole flowing pressure and coordinated wellhead pressure of each well under the current gas production rate are determined.

[0092] Starting with the coordinated wellhead pressure of each individual well, and based on the multiphase horizontal pipe flow pressure drop formula, segmented hydraulic calculations are performed sequentially along the fluid flow direction until the system terminal pressure for positive pressure regulation is obtained.

[0093] Starting from the set target pressure at the end of the system, and based on the multiphase horizontal pipe flow pressure drop formula, reverse hydraulic calculation is performed to calculate the theoretical pressure values ​​corresponding to each upstream node in order to meet the target pressure at the end.

[0094] Based on the system design pressure and instrument accuracy, a set of preset allowable pressure deviation thresholds for each node is established.

[0095] If the absolute value of the deviation between the pressure of each node obtained by forward pressure regulation and the theoretical pressure value of the corresponding node obtained by reverse pressure regulation is greater than the corresponding deviation threshold, it is determined that the system pressure has not reached equilibrium. The gas production of each single well is dynamically adjusted through feedback control algorithm. Based on the new gas production allocation after adjustment, the process of determining coordinated pressure, forward and reverse pressure regulation is re-executed until the pressure deviation of all key nodes meets the threshold requirement. If not, it is determined that the system pressure has reached equilibrium and the theoretical pressure value of each node is directly output.

[0096] Record the final gas production distribution of each well after iterative convergence, coordinate the bottom hole flowing pressure, coordinate the wellhead pressure, and the theoretical pressure values ​​of each node to form a preliminary pressure steady-state scheme and updated IPR and VLP curves of each well.

[0097] This can be explained by the fact that during the gas storage and injection process, the fluid produced (or injected) from the gas reservoir needs to reach the site via the wellbore and pipeline network. The pressure at each stage must meet the matching relationship of "gas reservoir pressure supply capacity → wellbore delivery capacity → pipeline network carrying capacity". Secondly, the downstream pipeline network pressure requirement is a rigid constraint (e.g., requiring a terminal pressure ≥ 6.0 MPa). If only forward pressure regulation is performed, a situation may arise where "gas production meets the target but terminal pressure is insufficient"; if only reverse pressure regulation is performed, a situation may arise where "pressure meets the target but gas reservoir production capacity is insufficient". Two-way calculation is the only method that can simultaneously meet both the "gas production target" and the "pipeline network constraint". Therefore, this scheme, by combining forward and reverse pressure regulation, allows the model to achieve the desired total gas production after a given target. The theoretical pressure values ​​of each node are derived forward, and the theoretical pressure values ​​of each node are derived backward by setting the target pressure at the end. Specifically, the "upper limit of theoretical pressure on the gas reservoir side" is obtained by forward pressure regulation, and the "lower limit of theoretical pressure on the pipeline side" is obtained by reverse pressure regulation. Thus, the theoretical pressure values ​​of each node are determined bidirectionally according to production needs, ensuring the dynamic pressure balance of a single well. Therefore, through iterative optimization, the pressure of each node output by this model is located within the intersection of the gas reservoir pressure supply capacity envelope and the pipeline pressure demand envelope, achieving the steady-state global optimum of single-well pressure under hard constraints. The nodes include: gas reservoir side nodes (such as gas reservoir formation, single well bottom, single well wellhead) and pipeline side nodes (such as gas gathering station, distribution valve group, system end).

[0098] The initial given gas production rate for each individual well can be estimated based on the unobstructed flow rate, the proportion of controlled geological reserves, or the historical best gas production rate proportion of the previous cycle. This initial given gas production rate is intended to provide a reasonable starting point for subsequent system allocation, and its accuracy will be optimized and corrected by a combination of forward and reverse pressure regulation.

[0099] The determination of the coordinated bottomhole flowing pressure and coordinated wellhead pressure for each well at the current gas production rate specifically includes: based on the already acquired VLP curve family... Each VLP curve corresponds to the mapping relationship between bottom hole flowing pressure and wellhead pressure under different gas production rates. At the same time, the IPR curve of each individual well is the mapping relationship between gas production and bottom hole flowing pressure. For a given preset gas production rate, the VLP curve corresponding to the preset gas production rate is obtained by interpolating on the VLP curve family (such as linear interpolation or spline interpolation). This curve represents the correspondence between bottom hole flowing pressure and wellhead pressure when the gas production rate is the preset gas production rate. Based on the IPR curve of the individual well, the bottom hole flowing pressure corresponding to the preset gas production rate is read. By substituting the bottom hole flowing pressure corresponding to the preset gas production rate into the VLP curve corresponding to the preset gas production rate, the wellhead pressure corresponding to the preset gas production rate is obtained. The bottom hole flowing pressure corresponding to the preset gas production rate is recorded as the coordinated bottom hole flowing pressure, and the corresponding wellhead pressure is recorded as the coordinated wellhead pressure.

[0100] The formula for the pressure drop in multiphase horizontal pipe flow is:

[0101] ,

[0102] In the formula, For upstream nodes to downstream nodes The Total pressure drop of the pipeline section The integral variable is along the length of the pipe segment, with the integration range starting from the upstream end ( ) to downstream end ( ), For the first The length of the pipe section For the first The angle between the section of pipe and the horizontal plane. In the first Density of gas-liquid mixture under flow conditions in a section of pipeline For the first Multiphase flow friction coefficient of a section of pipeline For the first The flow rate of the mixture within the section of the pipe, For the first The inner diameter of the section of pipe;

[0103] The multiphase horizontal pipe flow pressure drop formula is the fundamental physical model for hydraulic calculations of gathering and transmission pipeline networks. Its function is to quantitatively calculate the total pressure drop of any pipeline segment under given gas production, fluid properties, and pipeline geometric parameters. This pressure drop consists of two parts: gravity pressure drop and friction pressure drop, which respectively reflect the effects of topographic relief and flow resistance. In forward pressure regulation, this expression is used to solve for the downstream pressure from the known upstream pressure; in reverse pressure regulation, it is used to solve for the required upstream pressure from the downstream target pressure. By applying this expression sequentially along the pipeline network topology, the pressure distribution of the entire system can be simulated and the target can be reversed. It is a key calculation tool for completing system pressure balance analysis.

[0104] The target pressure at the end of the system is determined based on downstream user contracts, minimum pressure requirements for safe operation of the pipeline network, or instructions from the superior dispatcher. As an optional implementation method, the target pressure at the end usually needs to meet the minimum operating pressure requirements of the downstream pipeline network. Its specific value can be given by the design documents or operating procedures of the gas storage facility. For example, for a system with a design operating pressure of 6.0-8.0 MPa, its target pressure at the end can be set to 6.0 MPa.

[0105] The deviation threshold is determined based on the minimum controllable pressure difference between the system design pressure and the instrument accuracy. A recommended method for determining this threshold is to set... In the formula, The deviation threshold, For safety margin, the value can range from 1.5 to 3. This represents the maximum permissible absolute value of the pressure transmitter's error.

[0106] The method for dynamically adjusting the gas production of each well through feedback control algorithm is recommended to use a PID algorithm. Specifically, the absolute value deviation between the pressure at each node obtained by forward pressure regulation and the theoretical pressure at the corresponding node obtained by reverse pressure regulation is used as the deviation value of the PID algorithm. The proportional, integral, and derivative terms in the PID algorithm are added together to obtain the gas production adjustment amount for each well in this iteration. This adjustment amount is added to the current preset gas production amount to generate a new test gas production amount for the next iteration calculation. The new test gas production amount is then used to re-execute the process of determining the coordinated pressure, forward pressure regulation, and reverse pressure regulation until the absolute value of the wellhead pressure deviation of all individual wells is less than the deviation threshold. At this point, the system is determined to have reached a pressure balance state, and the iteration stops.

[0107] At this point, for each single well, its final gas production simultaneously satisfies: the flow rate that the formation can provide under the current formation pressure, determined by the IPR curve, and the flow rate that the wellbore can lift under the current wellhead pressure, determined by the VLP curve. That is, its operating point is located at the intersection of the IPR and VLP curves. The preliminary pressure steady-state scheme is the physical feasibility basis for subsequent economic optimization and real-time control.

[0108] The process of acquiring intermediate pressure data during the positive pressure regulation process, constructing an economic value assessment index function, and evaluating the total daily operating cost of the system under a given gas production volume specifically includes:

[0109] Based on the forward pressure regulation process, intermediate pressure data during the forward pressure regulation process are obtained, including: formation pressure, bottom hole flowing pressure, and wellhead pressure;

[0110] Based on intermediate pressure data, an economic value assessment index function is constructed.

[0111] The economic value assessment index function consists of production cost, processing cost, compressor energy consumption cost, and penalty for constraint violation;

[0112] Based on the economic value assessment index function, the total daily operating cost of the system is calculated under a given gas production volume.

[0113] This can be explained by combining single-well IPR with VLP curves and forward and reverse pressure regulation, using theoretical physical rules to maintain stable and balanced pressure during the gas storage injection and production process. However, in actual gas storage injection and production, pressure stability and balance cannot be considered alone; economic optimization is also a crucial factor. Perfect pressure stability and balance during gas storage injection and production cannot be guaranteed at the expense of significant economic losses. Therefore, an economically optimal constraint needs to be added. This scheme constructs an economic value assessment index function based on intermediate pressure data from the forward pressure regulation process, combined with production costs, processing costs, compressor energy consumption costs, and constraint violation penalties during gas storage injection and production. This function serves as the optimization objective, with pressure balance as the constraint. Ultimately, this yields an injection and production scheme that satisfies both steady-state pressure requirements and economic feasibility, thus improving the engineering feasibility of the technical solution.

[0114] The functional expression for the economic value assessment index is as follows:

[0115] ,

[0116] In the formula, The total number of individual wells. For the first Water production rate of the well , , , , , Coefficients for various costs or penalties. Here is the compressor power calculation function, where, Due to import pressure, Due to export pressure, Total gas volume For the first The upper limit of gas production of a well. For the first The upper limit of wellhead pressure for each well is a fixed constraint constant that is pre-set based on the rated capacity of single well equipment and the safety operation specifications of gas storage facilities. Its specific value can usually be determined in combination with the actual working conditions on site (such as the pressure bearing capacity of the tubing string, the rated parameters of the surface equipment, etc.).

[0117] In the expression The corresponding production cost item includes the production costs of all individual wells. The corresponding processing cost item is the gas processing cost based on the total gas volume. The corresponding compressor energy consumption cost item represents the energy consumption cost of the compressor during the injection and extraction process. The corresponding constraint violation penalty items are based on penalties for exceeding gas production limits and pressure limits. and By examining the first The historical production data of the well were fitted using multiple linear regression to represent the energy consumption cost per unit pressure drop and the chemical cost per unit produced fluid, respectively. The unit processing cost is calculated based on the gas processing plant's total annual operating cost and total gas volume contract. Based on the compressor's factory performance curve and the current electricity price contract, the unit energy cost coefficient is determined. and The penalty coefficient set for engineering experience is sufficiently large (e.g.) (at the order of magnitude) to ensure that the optimization algorithm strictly avoids violations numerically.

[0118] Reference Figure 4 As shown, the construction of the comprehensive optimization model based on neural networks, which outputs the baseline values ​​of injection and recovery rates for all wells and the target values ​​for each node, specifically includes:

[0119] In the offline phase, based on the constructed single-well IPR and VLP curves, and using a combination of forward and reverse pressure regulation, batch numerical simulations were conducted using the controlled variable method.

[0120] The data after numerical simulation is preprocessed and a training sample set is constructed. The preprocessing includes at least: filtering, interpolation, one-hot encoding of non-numerical features, and normalization.

[0121] The training sample set is divided into a training set, a validation set, and a test set according to an 8:1:1 allocation method.

[0122] The sample set input features include at least: the current formation average pressure of each single well, well structure parameters, current fluid properties, preset bottom hole flow pressure sequence, pipeline parameters, target total gas production of the gas storage facility, and target pressure at the end of the system.

[0123] From the sample set, we select the steady-state pressure scheme that satisfies the pressure drop constraint and minimizes the total daily operating cost of the system. We then extract the pressure values ​​of each node and the gas production of a single well corresponding to the scheme, and use them together as the output labels of the sample set.

[0124] With the goal of minimizing the deviation between the pressure value of the output label and the node pressure value predicted by the model, a pressure drop compliance loss function is constructed based on the mean square error formula.

[0125] With the goal of minimizing the deviation between the output gas production and the model-predicted single-well gas production, an economically optimal loss function is constructed based on the mean square error formula.

[0126] Based on the pressure reduction compliance loss function and the economic optimal loss function, the total loss function is obtained by weighted summation, and its weights are obtained through cross-validation of the validation set.

[0127] Based on a multilayer perceptron network architecture, the Adam optimizer is used to train the system with the goal of minimizing the total loss function using a training set.

[0128] After training is completed, the model parameters that perform best on the validation set during the entire training process are saved as the final comprehensive optimization model based on the neural network.

[0129] Based on the trained neural network-based comprehensive optimization model, the system outputs the baseline values ​​of injection and recovery volumes for all wells and the target values ​​for each node by inputting the actual collected input features.

[0130] This can be explained by the fact that the IPR curve provides the relationship between single-well production capacity and bottomhole flowing pressure, the VLP curve provides the relationship between single-well bottomhole flowing pressure and wellhead pressure, bidirectional pressure regulation ensures pressure matching between the gas reservoir, wellbore, and pipeline network, and the economic value assessment index function provides a quantitative standard for optimal cost. These four elements together constitute the physical rule logic of batch numerical simulation, providing high-quality training samples (the mapping relationship between input features and output labels) for the neural network model. Based on the powerful fitting ability of the neural network, it learns and trains under the condition of high-quality training samples with physical rules, so that the resulting comprehensive optimization model based on the neural network has good predictive ability and physical rule constraints, thereby ensuring the feasibility of all well injection and production benchmark values ​​and target values ​​at each node output by the model.

[0131] The batch numerical simulation using the controlled variable method specifically includes:

[0132] Using existing oil and gas field engineering numerical simulation platforms or self-built 3D simulation models, a digital twin 3D model is constructed according to design requirements. Through a data interface, IPR and VLP curve equations, bidirectional pressure regulation methods, and economic evaluation functions are embedded into the digital twin model to build a computational kernel. This kernel is driven by automated scripts to systematically traverse a preset operating condition parameter space, generating a high-fidelity simulation dataset and achieving parameter linkage between the theoretical model and the 3D visualization model. The specific selection of the simulation platform (such as using commercial software like Petrel, Pipesim, or a self-developed system) does not affect the core steps and effects of the method. Furthermore, by setting all operating condition parameters that affect the operation of the gas storage facility, and controlling a single... The simulation experiment is conducted using variables. That is, through automated scripts, the theoretical model of this scheme is driven to perform a complete pressure balance optimization calculation based on physical rules for each combination of parameters in the set. Each calculation simulates the entire process from a given initial state to achieving system pressure balance, and records the initial conditions (input) and the final converged optimal gas production allocation and pressure distribution (output). The set of all records constitutes a high-fidelity simulation dataset for training the neural network model. The operating parameters include at least: different formation pressures for each well, well structure parameters, current fluid properties, preset bottom hole flow pressure sequence, pipeline parameters, different target total gas production of gas storage facilities, and different target pressures at the end of the system.

[0133] It should be noted that the pipeline parameters mentioned are based on the actual pipeline parameter values ​​of the gas storage tank during the injection and production process, such as the pipeline inner diameter, cross-sectional area, length, and the angle between the pipeline and the horizontal plane. These parameters can be specifically determined based on the parameters in the theoretical model of this scheme. If it is necessary to set parameters that are not directly involved in the parameters but are related to them in the theoretical model of this scheme, it is only necessary to supplement the physical rules of them and the parameters that affect them, so that the scheme can be operated and implemented within its framework.

[0134] As a preferred example, the basic architecture of the neural network model in this solution is as follows:

[0135] Network architecture: It adopts a multilayer perceptron (MLP), which consists of an input layer, three fully connected hidden layers and an output layer. The number of neurons in the input layer is equal to the dimension of the input features, and the number of neurons in the output layer is equal to the dimension of the output labels.

[0136] Hidden layer configuration: The number of neurons in the three hidden layers are 128, 64 and 32 respectively;

[0137] Activation function: The hidden layer uses the ReLU activation function, and the output layer uses the linear activation function, which is suitable for regression prediction of gas production and pressure values;

[0138] Loss function: The basic function is constructed using the mean squared error loss function, which measures the difference between the model's predicted values ​​and the actual simulation values;

[0139] Optimizer: The Adam optimizer is used, with a piecewise constant decay strategy for the learning rate. The initial learning rate is... After training for 20 epochs, it decays to After 50 epochs, it further decays to ;

[0140] Training parameters: To accommodate GPU memory limitations, the batch size was set to 2, the total number of training epochs was 100, and the experiment was repeated 10 times to evaluate model stability;

[0141] Dataset partitioning: The training sample set is randomly divided into a training set, a validation set, and a test set, with a ratio of 8:1:1;

[0142] Early stopping strategy: During training, an early stopping method is used. If the validation set loss does not decrease for 10 consecutive training rounds, training is automatically terminated to prevent overfitting.

[0143] The verification of the stability of the neural network-based comprehensive optimization model, based on a combination of forward and reverse pressure regulation, using the output benchmark values ​​of injection and recovery volumes for all wells and the target values ​​for each node, specifically includes:

[0144] The input features of the test set are input into a comprehensive optimization model based on a neural network to obtain the predicted well injection recovery baseline value and the target pressure value of each node;

[0145] Starting with the predicted wellhead pressure, perform forward calculations and compare whether the obtained system end pressure is within an acceptable range from the system end target pressure in the input features.

[0146] Using the predicted gas production as a given value, re-execute the complete forward and reverse pressure regulation iteration process and evaluate the following results:

[0147] Determine whether the iteration can converge within a preset number of consecutive rounds;

[0148] Determine whether the deviation between the pressure obtained at each node after convergence and the pressure predicted by the model is less than the deviation threshold.

[0149] Determine whether the distance between the converged point and the intersection of the IPR and VLP curves of each well is less than the preset value;

[0150] Substitute the model-predicted scheme into the economic value assessment index function to calculate its total daily operating cost of the system, and compare it with the optimal cost in the test set label to determine that the deviation between the two should be within an acceptable range.

[0151] The comprehensive optimization model based on neural networks, whose judgment results all meet the design requirements, is used as the integrated operation model for gas storage facilities in practical applications.

[0152] It can be explained that the model validation in this step is performed on the test set to evaluate the stability of the model after training. The validation results can be judged and evaluated based on the design specifications and expert experience. For cases where the validation fails, adjustments are made based on the judgment points of the failure using expert experience. For example, if the iteration fails to converge within a preset number of consecutive rounds, the training parameters of the model are adjusted and retrained. The preset number of rounds, preset values, and acceptable ranges are set based on the accuracy requirements of integrated gas storage operation and expert experience.

[0153] Furthermore, based on the same inventive concept as the aforementioned method for constructing an integrated operation model for gas storage facilities based on digital twins, this solution proposes a system for constructing an integrated operation model for gas storage facilities based on digital twins, comprising:

[0154] The physical model module is used to calculate the corresponding bottomhole flowing pressure and wellhead pressure based on the current formation average pressure of a single well, and to construct the IPR and VLP curves of the single well; based on the IPR curve and VLP curve family, it outputs the set of pressure values ​​of each node after pressure stabilization by combining forward and reverse pressure regulation, forming a preliminary pressure steady-state scheme; it obtains the intermediate pressure data during the forward pressure regulation process, constructs an economic value assessment index function, and assesses the total daily operating cost of the system under a given gas production rate;

[0155] The integrated model and verification module is used to construct a neural network-based integrated optimization model with the goals of pressure reduction compliance and economic optimization, and output the baseline values ​​of injection and recovery volume of all wells and the target values ​​of each node; based on the output baseline values ​​of injection and recovery volume of all wells and the target values ​​of each node, the stability of the neural network-based integrated optimization model is verified by combining positive and negative pressure regulation.

[0156] The physical model module includes:

[0157] The IPR and VLP curve unit is used to calculate the corresponding bottom hole flowing pressure and wellhead pressure based on the current formation average pressure of a single well, and to construct the IPR and VLP curves of a single well.

[0158] The positive and negative pressure regulating unit is used to output a set of pressure values ​​of each node after stabilization by combining positive and negative pressure regulation according to the IPR curve and VLP curve family, forming a preliminary pressure steady state scheme.

[0159] An economic evaluation unit is used to acquire intermediate pressure data during the positive pressure regulation process, construct an economic value evaluation index function, and evaluate the total daily operating cost of the system under a given gas production volume.

[0160] The integrated model and verification module include:

[0161] The integrated model unit is used to construct a comprehensive optimization model based on neural networks with the goals of pressure reduction compliance and economic optimization, and outputs the baseline values ​​of injection and recovery volume for all wells and the target values ​​for each node.

[0162] The model verification unit is used to verify the stability of the neural network-based comprehensive optimization model based on the output of all well injection and production benchmark values ​​and the target values ​​of each node, using a combination of forward and reverse pressure regulation.

[0163] In summary, the advantages of this invention are: it deeply integrates the physical laws of gas storage injection and production, system pressure balance, and economic objectives, achieving a rapid and reliable closed loop from pressure steady-state optimization to economically optimal decision-making.

[0164] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for constructing an integrated operation model for gas storage facilities based on digital twins, characterized in that, include: Based on the current average formation pressure of a single well, calculate its corresponding bottom hole flowing pressure and wellhead pressure, and construct the IPR and VLP curves for the single well; Based on the IPR curve and VLP curve family, a combination of forward and reverse pressure regulation is used to output the set of pressure values ​​of each node after pressure stabilization, forming a preliminary pressure steady-state scheme. Obtain intermediate pressure data during the positive pressure regulation process, construct an economic value assessment index function, and assess the total daily operating cost of the system under a given gas production rate; With the goals of pressure reduction compliance and economic optimization, a comprehensive optimization model based on neural networks is constructed to output the baseline values ​​of injection and recovery volumes for all wells and the target values ​​for each node. Based on the output of all well injection and production baseline values ​​and the target values ​​of each node, the stability of the neural network-based comprehensive optimization model is verified by combining forward and reverse pressure regulation. The aforementioned comprehensive optimization model, based on a neural network and aiming at pressure reduction compliance and economic optimization, outputs benchmark values ​​for injection and recovery rates for all wells and target values ​​for each node, specifically including: In the offline phase, based on the constructed single-well IPR and VLP curves, and using a combination of forward and reverse pressure regulation, batch numerical simulations were conducted using the controlled variable method. The data after numerical simulation is preprocessed and a training sample set is constructed. The preprocessing includes at least: filtering, interpolation, one-hot encoding of non-numerical features, and normalization. The training sample set is divided into a training set, a validation set, and a test set according to an 8:1:1 allocation method. The input features of the sample set should include at least the following: the current formation average pressure of each single well, well structure parameters, current fluid properties, preset bottom hole flow pressure sequence, pipeline parameters, target total gas production of the gas storage facility, and target pressure at the end of the system. From the sample set, we select the steady-state pressure scheme that satisfies the pressure drop constraint and minimizes the total daily operating cost of the system. We then extract the pressure values ​​of each node and the gas production of a single well corresponding to the scheme, and use them together as the output labels of the sample set. With the goal of minimizing the deviation between the pressure value of the output label and the node pressure value predicted by the model, a pressure drop compliance loss function is constructed based on the mean square error formula. With the goal of minimizing the deviation between the output gas production and the model-predicted single-well gas production, an economically optimal loss function is constructed based on the mean square error formula. Based on the pressure reduction compliance loss function and the economic optimal loss function, the total loss function is obtained by weighted summation, and its weights are obtained through cross-validation of the validation set. Based on a multilayer perceptron network architecture, the Adam optimizer is used to train the system with the goal of minimizing the total loss function using a training set. After training is completed, the model parameters that perform best on the validation set during the entire training process are saved as the final comprehensive optimization model based on the neural network. Based on the trained neural network-based comprehensive optimization model, the system outputs the baseline values ​​of injection and recovery volumes for all wells and the target values ​​for each node by inputting the actual collected input features.

2. The method for constructing an integrated operation model of a gas storage facility based on digital twins according to claim 1, characterized in that, The process of calculating the corresponding bottomhole flowing pressure and wellhead pressure based on the current average formation pressure of a single well, and constructing the single-well IPR and VLP curves specifically includes: Based on the formation pressure monitoring data of the gas storage facility, the current average formation pressure of each individual well is obtained; Obtain the wellbore structural parameters, current fluid properties, and preset bottomhole flowing pressure sequence for each single well; The structural parameters include at least: tubing inner diameter, vertical depth, and well inclination angle; the current fluid properties include at least: gas-liquid mixture density, multiphase flow friction coefficient, and mixture velocity. Based on the current formation average pressure and the preset bottom hole flowing pressure sequence, and based on Darcy's law and gas seepage theory, the gas production sequence of each single well is calculated, and the IPR curve of each single well is constructed. Based on the calculated gas production sequence, for each value in the preset bottom hole flowing pressure sequence, the corresponding wellhead pressure is calculated based on the multiphase pipe flow mechanical energy conservation equation, and a family of VLP curves for a single well is constructed.

3. The method for constructing an integrated operation model of a gas storage facility based on digital twins according to claim 2, characterized in that, The process of using a combination of forward and reverse pressure regulation based on the IPR curve and VLP curve family to output the set of pressure values ​​at each node after pressure stabilization, forming a preliminary pressure steady-state scheme, specifically includes: Based on the geological data of the gas storage facility, the well structure data of each individual well, and historical production data, the initial given gas production rate for each individual well is set. Using the given gas production rate of each individual well as input, and combining the IPR curve and VLP curve family of each individual well, the coordinated bottom hole flowing pressure and coordinated wellhead pressure of each well under the current gas production rate are determined. Starting with the coordinated wellhead pressure of each individual well, and based on the multiphase horizontal pipe flow pressure drop formula, segmented hydraulic calculations are performed sequentially along the fluid flow direction until the system terminal pressure for positive pressure regulation is obtained. Starting from the set target pressure at the end of the system, and based on the multiphase horizontal pipe flow pressure drop formula, reverse hydraulic calculation is performed to calculate the theoretical pressure values ​​corresponding to each upstream node in order to meet the target pressure at the end. Based on the system design pressure and instrument accuracy, a set of preset allowable pressure deviation thresholds for each node is established. If the absolute value of the deviation between the pressure of each node obtained by forward pressure regulation and the theoretical pressure value of the corresponding node obtained by reverse pressure regulation is greater than the corresponding deviation threshold, it is determined that the system pressure has not reached equilibrium. The gas production of each single well is dynamically adjusted through feedback control algorithm. Based on the new gas production allocation after adjustment, the process of determining coordinated pressure, forward and reverse pressure regulation is re-executed until the pressure deviation of all key nodes meets the threshold requirement. If not, it is determined that the system pressure has reached equilibrium and the theoretical pressure value of each node is directly output. Record the final gas production distribution of each well after iterative convergence, coordinate the bottom hole flowing pressure, coordinate the wellhead pressure, and the theoretical pressure values ​​of each node to form a preliminary pressure steady-state scheme and updated IPR and VLP curves of each well.

4. The method for constructing an integrated operation model of a gas storage facility based on digital twins according to claim 3, characterized in that, The process of acquiring intermediate pressure data during the positive pressure regulation process, constructing an economic value assessment index function, and evaluating the total daily operating cost of the system under a given gas production volume specifically includes: Based on the forward pressure regulation process, intermediate pressure data during the forward pressure regulation process are obtained, including: formation pressure, bottom hole flowing pressure, and wellhead pressure; Based on intermediate pressure data, an economic value assessment index function is constructed. The economic value assessment index function consists of production cost, processing cost, compressor energy consumption cost, and penalty for constraint violation; Based on the economic value assessment index function, the total daily operating cost of the system is calculated under a given gas production volume.

5. The method for constructing an integrated operation model of a gas storage facility based on digital twins according to claim 4, characterized in that, The verification of the stability of the neural network-based comprehensive optimization model, based on a combination of forward and reverse pressure regulation, using the output benchmark values ​​of injection and recovery volumes for all wells and the target values ​​for each node, specifically includes: The input features of the test set are input into a comprehensive optimization model based on a neural network to obtain the predicted well injection recovery baseline value and the target pressure value of each node; Starting with the predicted wellhead pressure, perform forward calculations and compare whether the obtained system end pressure is within an acceptable range from the system end target pressure in the input features. Using the predicted gas production as a given value, re-execute the complete forward and reverse pressure regulation iteration process and evaluate the following results: Determine whether the iteration can converge within a preset number of consecutive rounds; Determine whether the deviation between the pressure obtained at each node after convergence and the pressure predicted by the model is less than the deviation threshold. Determine whether the distance between the converged point and the intersection of the IPR and VLP curves of each well is less than the preset value; Substitute the model-predicted scheme into the economic value assessment index function to calculate its total daily operating cost of the system, and compare it with the optimal cost in the test set label to determine that the deviation between the two should be within an acceptable range. The comprehensive optimization model based on neural networks, whose judgment results all meet the design requirements, is used as the integrated operation model for gas storage facilities in practical applications.

6. A system for constructing an integrated operation model for gas storage facilities based on digital twins, characterized in that, The method for constructing an integrated operation model of a gas storage facility based on digital twins as described in any one of claims 1-5 includes: The physical model module is used to calculate the corresponding bottomhole flowing pressure and wellhead pressure based on the current formation average pressure of a single well, and to construct the IPR and VLP curves of the single well; based on the IPR curve and VLP curve family, it outputs the set of pressure values ​​of each node after pressure stabilization by combining forward and reverse pressure regulation, forming a preliminary pressure steady-state scheme; it obtains the intermediate pressure data during the forward pressure regulation process, constructs an economic value assessment index function, and assesses the total daily operating cost of the system under a given gas production rate; The integrated model and verification module is used to construct a neural network-based integrated optimization model with the goals of pressure reduction compliance and economic optimization, and output the injection and recovery volume benchmark values ​​of all wells and the target values ​​of each node; based on the output of all well injection and recovery volume benchmark values ​​and the target values ​​of each node, the stability of the neural network-based integrated optimization model is verified by combining positive and negative pressure regulation.

7. The integrated operation model construction system for gas storage based on digital twins according to claim 6, characterized in that, The physical model module includes: The IPR and VLP curve unit is used to calculate the corresponding bottom hole flowing pressure and wellhead pressure based on the current formation average pressure of a single well, and to construct the IPR and VLP curves of a single well. The positive and negative pressure regulating unit is used to output a set of pressure values ​​of each node after stabilization by combining positive and negative pressure regulation according to the IPR curve and VLP curve family, forming a preliminary pressure steady state scheme. An economic evaluation unit is used to acquire intermediate pressure data during the positive pressure regulation process, construct an economic value evaluation index function, and evaluate the total daily operating cost of the system under a given gas production volume.

8. The integrated operation model construction system for gas storage based on digital twins according to claim 7, characterized in that, The integrated model and verification module include: The integrated model unit is used to construct a comprehensive optimization model based on neural networks with the goals of pressure reduction compliance and economic optimization, and outputs the baseline values ​​of injection and recovery volume for all wells and the target values ​​for each node. The model verification unit is used to verify the stability of the neural network-based comprehensive optimization model based on the output of all well injection and production benchmark values ​​and the target values ​​of each node, using a combination of forward and reverse pressure regulation.

Images