Intelligent completion double closed-loop regulation method and system based on double-scale proxy model

Abstract

The application discloses an intelligent completion double closed loop regulation method and system based on a double scale agent model, which comprises a reservoir scale agent model and a near well scale agent model, is respectively established through a neural differential equation, and is used for predicting formation pressure, saturation and well production and well section liquid production and water cut. In an outer large closed loop, cumulative oil production, water cut and net present value are used as optimization targets, formation pressure, water cut and facility capacity are used as constraints, and well production or bottom hole pressure is optimized; in an inner small closed loop, section water cut uniformity, maximum gradient and production deviation are used as targets, and the opening degrees of ICVs are optimized, so that profile stability and water breakthrough delay are realized. The system supports periodic online optimization and sliding window light correction, sets safety constraints and rollback strategies, and guarantees the feasibility and safety of the strategies. The application realizes the collaborative optimization of the reservoir scale and the wellbore scale, takes into account far field interference and near well rectification, realizes real-time production regulation and control, and improves the intelligent completion production increasing effect and economic benefits.

Description

Technical Field

[0001] This invention belongs to the field of petroleum engineering and intelligent reservoir management technology, and in particular relates to an intelligent well completion dual closed-loop control method and system based on a dual-scale surrogate model. Background Technology

[0002] As oilfield development enters its mid-to-late stages, edge and bottom water energy gradually increases, and horizontal wells are widely used as an important means to improve production capacity and utilization. However, in actual production, due to the strong heterogeneity of the reservoir, complex interference between multiple wells, and uneven injection-production-replenishment relationships, horizontal wells are prone to problems such as premature water breakthrough, profile instability, and uneven production in different sections. This leads to a rapid decline in oil well production, resulting in a decrease in overall recovery rate and making it difficult to achieve the expected water control effect.

[0003] In recent years, intelligent well completion (ICV) technology has been gradually applied. ICV allows for the active control of the production fluid and water ratio in each well section by adjusting the opening degree. These technologies provide new means for water-controlled development, but there are still obvious limitations in their application. On the one hand, if only reservoir-scale overall optimization is relied upon, although it can provide multi-well control schemes such as bottom hole pressure and injection-production ratio, it cannot effectively decompose the results to the wellbore profile, making it difficult to guide the specific adjustment of ICV. On the other hand, if only the wellbore or near-well range is focused on, and the ICV opening degree is adjusted through historical fitting or profile analysis, the far-field injection-production relationship and inter-well interference are often ignored, resulting in local improvements but limited overall development effects.

[0004] While traditional numerical simulation techniques can achieve multi-scale joint analysis to some extent, they suffer from massive computational demands and insufficient real-time performance. A relatively sophisticated numerical model often requires several hours or even longer to complete an optimization calculation, making it difficult to meet the needs of online or near-online control in the field. Furthermore, reservoir geological parameters exhibit significant uncertainties, and the state continuously evolves during the displacement process, causing model predictions to gradually deviate from reality, and the optimal strategy quickly becomes invalid. Existing methods generally lack effective online adaptive mechanisms and safe rollback strategies, resulting in insufficient robustness and stability in practical production applications.

[0005] Furthermore, reservoir development exhibits differences in objectives and time scales across different scales. Reservoir-scale optimization typically aims to maximize long-term cumulative recovery or economic benefits, while near-wellbore-scale control emphasizes profile stability and delaying water inrush. Without an effective hierarchical coordination mechanism, these two approaches often operate independently or even conflict, making it difficult to achieve optimal overall system development.

[0006] In summary, existing water control technologies still suffer from problems such as a disconnect between reservoir and wellbore optimization objectives, insufficient numerical simulation efficiency, poor model reliability, and a lack of intelligent collaborative mechanisms. Therefore, a new method is urgently needed to overcome the shortcomings of existing water control strategies, such as reliance on single-scale models, neglect of reservoir-wellbore coupling, and insufficient real-time performance. Summary of the Invention

[0007] The problem to be solved by this invention is to provide an intelligent well completion dual-closed-loop control method and system based on a dual-scale surrogate model. This method can simultaneously establish an efficient and generalizable surrogate model at both the reservoir scale and the near-wellbore scale, and achieve the unification of macroscopic and microscopic objectives through a dual-closed-loop hierarchical control method, thereby improving real-time performance and safety while ensuring water control effect.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: an intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model, comprising the following steps: S1: Obtain the geological and engineering parameters of the target reservoir and block. If production data is available, use the production data for model training and calibration. If no production data is available, generate sample data based on reservoir numerical simulation. Generate wellbore segment water control samples using a local fine model or coupled model. Clean, normalize and time-align the production data or the sample data. S2: Using neural differential equations, multi-well control parameters and static geological parameters are modeled to establish a reservoir-scale surrogate model for predicting formation pressure, saturation distribution, well production and recovery rate. S3: Using bottom hole flowing pressure, ICV segment opening vector and wellbore friction parameters as inputs, a near-well scale surrogate model is established based on neural differential equations, and the outputs well section production distribution, section water cut and pressure drop; S4: On the reservoir-scale proxy model, with the cumulative oil production, water cut and net present value as optimization objectives, constrain the upper and lower limits of formation pressure, the upper limit of water cut and the capacity of surface facilities, and solve for the target production indicators of each well level to achieve outer layer optimization of intelligent well completion. S5: On the near-well scale proxy model, with well section profile stability and water inrush delay as the objective function, constrain the total production and bottom hole pressure difference, optimize the well section ICV opening vector to achieve inner layer optimization for intelligent well completion; S6: Feed the equivalent productivity index or skin coefficient obtained by near-well scale optimization back to the reservoir scale proxy model to update the reservoir distribution and prediction target; S7: During field operation, based on real-time / near real-time data, S4-S5 are periodically re-executed to perform outer and inner layer optimization, and a sliding window method is used to perform light correction on the surrogate model. When moisture content, pressure or equipment limiting conditions are triggered, a rollback condition and conservative opening strategy are executed.

[0009] Furthermore, in S2, the reservoir-scale surrogate model is established based on a neural differential equation structure to ensure the continuous expression and extrapolation capability of the prediction results to the dynamic laws.

[0010] Furthermore, in S3, the near-well-scale surrogate model learns a differentiable mapping between the aperture vector and the equivalent skin coefficient to support gradient-based optimization solutions.

[0011] Furthermore, in S4, the outer layer optimization is solved using the particle swarm optimization algorithm, and a dynamic inertia weight adjustment mechanism is adopted. Through multiple iterations of optimization and result comparison, it avoids getting trapped in local optima.

[0012] Furthermore, in step S5, the objective function for the inner layer optimization is: In the formula, The water cut distribution in the well section is dimensionless. The variance of the water content of the section is represented by a smaller value, which indicates a more uniform fluid distribution in the profile. It is dimensionless. The maximum gradient of water cut along the well section reflects the risk of water coning / channeling; the smaller the value, the more stable the condition. The unit is 1 / m. This refers to the actual well section length, in meters (m). To account for the difference between total production and target production, and to ensure that well-level production meets the requirements of the large closed-loop allocation, Total output per segment, in meters (m). 3 / d, The target output is expressed in meters (m). 3 / d; α, β, γ are weighting factors, which are dimensionless.

[0013] Furthermore, in S6, if the improvement of the optimization target is less than the threshold or reaches the set time window, the control strategy is output; otherwise, return to S4 to perform outer layer optimization.

[0014] Furthermore, in S7, when the water cut of the well section exceeds the preset upper limit, the bottom pressure is lower than the preset lower limit, or the surface facility capacity is limited, a retreat condition is triggered, and the opening degree is adjusted to a conservative step value.

[0015] Furthermore, this invention provides an intelligent well completion dual-closed-loop control system based on a dual-scale surrogate model, which, when running the aforementioned intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model, includes: Data acquisition and management module: used to acquire and manage reservoir, wellbore, and field operation data; Reservoir-scale proxy module: used to predict reservoir dynamics and production based on multi-well control parameters and static parameters; Near-well scale proxy module: used to predict the production / water distribution of the well section based on wellbore control parameters, and generate an equivalent production index or skin coefficient; The dual-layer optimization module consists of an outer multi-well optimization submodule and an inner ICV optimization submodule. Online calibration module: used to perform sliding window calibration of reservoir agent and near-well agent during real-time operation; Safety control module: used to execute a fallback strategy and conservative opening adjustment when threshold conditions are triggered; Execution and monitoring interface module: Used to interact with downhole intelligent completion tools to issue and transmit control commands.

[0016] Furthermore, the dual-layer optimization module supports hierarchical timescale separation and alternating iteration, with the outer layer optimization based on a daily / weekly cycle and the inner layer optimization based on an hourly / dayly cycle.

[0017] Furthermore, the present invention provides a computer-readable storage medium storing a computer algorithm, which, when executed by a processor, performs the aforementioned data processing.

[0018] The advantages and positive effects of this invention are: 1. This invention adopts a dual-scale surrogate model, which takes into account the dynamic characteristics of the reservoir far field and the wellbore near field, and realizes joint optimization of multiple wells and single wells.

[0019] 2. This invention introduces a neural differential equation surrogate model, which can capture the continuous dynamic process driven by oil and water, and has stronger generalization and extrapolation capabilities.

[0020] 3. This invention proposes a dual-layer control strategy of large closed loop and small closed loop, which takes into account both maximizing recovery rate and profile stability, and significantly delays water inrush.

[0021] 4. This invention supports online sliding window correction and a safety rollback mechanism to ensure the safety and real-time performance of control under model drift and parameter uncertainty conditions. Attached Figure Description

[0022] Figure 1 This is an overall flowchart of an embodiment of the present invention.

[0023] Figure 2 This is a target oilfield well location map according to an embodiment of the present invention.

[0024] Figure 3 This is a comparative diagram of the reservoir pressure profile of the reservoir model in an embodiment of the present invention.

[0025] Figure 4 This is a comparison diagram of the water saturation profile of the reservoir proxy model in an embodiment of the present invention.

[0026] Figure 5 This is a diagram illustrating the water control effect of the oil well particle swarm optimization according to an embodiment of the present invention.

[0027] Figure 6 This is a diagram showing the optimized ICV opening result under the optimal production rate in an embodiment of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, the intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model includes the following steps: S1: Obtain the geological and engineering parameters of the target reservoir and block. If production data is available, use the production data for model training and calibration. If there is no production data, generate sample data based on reservoir numerical simulation. Use a local fine model or coupled model to generate wellbore segment water control samples, and clean, normalize and time-align the above data.

[0030] S2: A reservoir-scale surrogate model is established by modeling multi-well control parameters and static geological parameters using a neural differential equation structure to predict formation pressure, saturation distribution, well production, and recovery rate. Specifically, the reservoir-scale surrogate model is built based on a neural differential equation structure to ensure the continuous expression and extrapolation capability of the prediction results to dynamic laws.

[0031] S3: Using bottomhole flowing pressure, ICV segment opening vector, and wellbore friction parameters as inputs, a near-wellbore-scale surrogate model is established based on neural differential equations, outputting well section production distribution, section water cut, and pressure drop. Specifically, the near-wellbore-scale surrogate model learns the differentiable mapping between the opening vector and the equivalent skin coefficient to support gradient-based optimization solutions.

[0032] S4: Based on the reservoir-scale surrogate model, with cumulative oil production, water cut, and net present value as optimization objectives, and constrained by upper and lower limits of formation pressure, upper limit of water cut, and surface facility capacity, the target production indicators for each well level are obtained, achieving outer-layer optimization for intelligent well completion. Specifically, the outer-layer optimization is solved using the particle swarm optimization algorithm, employing a dynamic inertia weight adjustment mechanism. Through multiple iterations of optimization and result comparison, it avoids getting trapped in local optima.

[0033] S5: Based on the near-wellbore scale proxy model, with well section profile stability and water inrush delay as the objective functions, and constraining total production and bottom hole pressure differential, the well section ICV aperture vector is optimized to achieve inner-layer optimization for intelligent well completion. Specifically, the objective function for inner-layer optimization is: In the formula, The water cut distribution in the well section is dimensionless. The variance of the water content of the section is represented by a smaller value, which indicates a more uniform fluid distribution in the profile. It is dimensionless. The maximum gradient of water cut along the well section reflects the risk of water coning / channeling; the smaller the value, the more stable the condition. The unit is 1 / m. This refers to the actual well section length, in meters (m). To account for the difference between total production and target production, and to ensure that well-level production meets the requirements of the large closed-loop allocation, Total output per segment, in meters (m). 3 / d, The target output is expressed in meters (m). 3 / d; α, β, γ are weighting factors, which are dimensionless.

[0034] S6: Feedback the equivalent productivity index or skin coefficient obtained from near-well-scale optimization to the reservoir-scale surrogate model to update the reservoir distribution and prediction targets. Specifically, if the improvement of the optimization target is less than the threshold or reaches the set time window, the control strategy is output; otherwise, return to S4 for outer-layer optimization.

[0035] S7: During field operation, based on real-time / near real-time data, S4-S5 are periodically re-executed to optimize the outer and inner layers. A sliding window method is used to perform minor corrections on the surrogate model. When water cut, pressure, or equipment limitations are triggered, a rollback condition and conservative well opening strategy are implemented. Specifically, when the water cut in the well section exceeds the preset upper limit, the bottom hole pressure is lower than the preset lower limit, or the surface facility capacity is limited, a rollback condition is triggered, and the well opening is adjusted to a conservative step value.

[0036] The present invention will now be described in detail with reference to specific embodiments: The target reservoir is a continental sandstone reservoir with edge and bottom water energy, and an oil-bearing area of ​​approximately 12 km². 2 The average effective thickness is 12m, porosity is 0.18–0.24, permeability is 50–800D, and the original formation pressure is 28MPa. Fifteen horizontal wells, each with a horizontal section length of 1200m, were installed in this block, all using the ICV intelligent well completion method.

[0037] Specifically, it includes the following steps: S1: Data and Sample Construction.

[0038] Obtain the geological and engineering parameters of the target reservoir and block. If production data is available, use it for model training and calibration. If no production data is available, generate sample data based on reservoir numerical simulation. Further, use a local fine model or coupled model to generate wellbore segment water control samples. Clean, normalize and time-align the above data.

[0039] Specifically, this reservoir is a newly developed block. Training samples were generated through numerical simulation to produce training data at both the reservoir scale and near-wellbore scale. The target reservoir well location distribution map is shown below. Figure 2 As shown.

[0040] S2: Training of reservoir-scale surrogate model.

[0041] By using neural differential equations to model multi-well control parameters and static geological parameters, a reservoir-scale surrogate model is established to predict formation pressure, saturation distribution, well production, and recovery rate.

[0042] Specifically, reservoir-scale modeling is performed using neural differential equations (NDEs). The inputs are multi-well control parameters (such as bottomhole flowing pressure). ,Yield Injection-production ratio (VRR, etc.) and static parameters (permeability k, porosity, etc.) (Relative permeability curve, initial saturation field). The output consists of field variables evolving over time, including pressure distribution. Water saturation Well production forecast and cumulative recovery rate.

[0043] Its basic dynamic equation is: (1) In the formula: For the field state variable vector (pressure field, saturation field, well production); For control vectors (bottom hole pressure, injection rate, etc.); Represents static parameters (pore permeability characteristics, relative permeability, etc.); The differential equation is parameterized by a neural network.

[0044] Training utilizes numerical simulation results, minimizing the loss function between predicted and true values: (2) For the proxy model at time For the Production forecast for each well, in m³ 3 / d; For a moment No. Numerical simulation production of wells, in m³. 3 / d; For reference production, the maximum value of the numerical simulation production of this well is taken, in m³. 3 / d; The regularization coefficient is dimensionless. These are the parameters of the neural network, dimensionless. This is the loss function used during model training.

[0045] The training results of the reservoir surrogate model are as follows Figure 3 and Figure 4 As shown.

[0046] S3: Near-well scale surrogate model training.

[0047] Using bottom hole flowing pressure, ICV segment opening vector, and wellbore friction parameters as inputs, a near-wellbore scale surrogate model is established based on neural differential equations to output well section production distribution, section water cut, and pressure drop.

[0048] Specifically, the near-wellbore scale surrogate model models the segmented ICV control of a single horizontal well, employing a neural differential equation + convolutional network approach. Inputs include: bottomhole flowing pressure and ICV opening vectors for each segment. Parameters such as tubing friction. The output is a wellbore profile distribution, including the production rate of each section. Section moisture content and pressure drop distribution .

[0049] The surrogate model also learned the mapping relationship between the aperture vector and the equivalent wellbore parameters (segmented productivity index PI, skin coefficient S): (3) (4) in, For the first The equivalent productivity index of each completed section, in m³. 3 / (d·MPa); For the first The equivalent skin coefficient of each completed well section, dimensionless; is the well completion section opening vector, dimensionless (0-1); For the first The equivalent permeability of the formation corresponding to each well completion section, in mD; For the first The effective pressure difference of each well completion section is expressed in MPa. , This is a neural network function.

[0050] S4: Outer layer optimization, i.e., large closed-loop optimization.

[0051] Based on the reservoir-scale proxy model, the cumulative oil production, recovery rate and net present value are used as optimization objectives. The upper and lower limits of formation pressure, the upper limit of water cut and the capacity of surface facilities are constrained to obtain the target production indicators for each well level, thereby realizing the outer layer optimization of intelligent well completion.

[0052] Specifically, the first step is to optimize the comprehensive oil production and water cut of each well in the target oilfield as the optimization objectives. The specific optimization process is as follows: Figure 5 As shown in the figure. Subsequently, the economic benefit parameter NPV is introduced, and the production allocation requirements of each oil well in the oilfield are obtained by calculating NPV.

[0053] Specifically, the formula for calculating Net Present Value (NPV) is as follows: (5) In the formula, The discount rate is a dimensionless value that is typically determined based on factors such as market interest rates, project risks, and the company's cost of capital. It is a moment The revenue, expressed in yuan, is calculated using the following formula: (6) In the formula, For a moment Production, in m³ 3 , For a moment The price of oil is in yuan / m³ 3 .

[0054] It is a moment The cost is calculated using the following formula: (7) In the formula, Fixed costs, such as initial investment costs for drilling and equipment purchase, are typically in the range of... It may be included in a lump sum at the time of evaluation, or amortized in equal installments over the evaluation period, with the unit being yuan; In order to be in The operating costs of each moment include daily operations, maintenance, and labor costs, in yuan; In order to be in Variable costs at any given time are typically related to output, such as material consumption and energy usage, and are expressed in yuan / m. 3 .

[0055] Assuming oil prices =4000 yuan / ton (equivalent to approximately 3400 yuan / m³) 3 ), Operating variable costs =200 yuan / m 3 Fixed costs =20 million yuan, with annual operating costs of =500,000 yuan, discount rate d=10%. The optimized oil production of a certain well in years 1 to 5 are 100, 95, 90, 80, and 70 m³, respectively. 3 / d, assuming continuous production for 365 days a year, then: Year 1 output Q1 = 36,500 m³ 3 The corresponding revenue is R1≈124 million yuan; The output in the second year was Q2 = 34,675 m³. 3 The corresponding revenue is R2≈118 million yuan.

[0056] Calculate the net cash flow for each year in turn, discount and sum them, and after deducting fixed costs, the NPV for the well over 5 years is approximately 371 million yuan. Therefore, the method of this invention can achieve optimal economic benefits in reservoir development while ensuring recovery rate. Based on the above NPV optimization and constraints, the initial production allocation and water cut schemes for six representative wells are shown in Table 1.

[0057] Table 1. Production and water cut of 6 wells S5: Inner layer optimization, i.e. small closed-loop optimization.

[0058] In the near-wellbore scale proxy model, with well section profile stability and water inrush delay as the objective functions, and constraining total production and bottom hole pressure difference, the well section ICV opening vector is optimized to achieve inner-layer optimization for intelligent well completion.

[0059] Specifically, given the well production or bottom hole pressure targets in the large closed-loop system, the ICV (Internal Volume Vault) vector for each well is optimized. The objective function is: In the formula, The water cut distribution in the well section is dimensionless. The variance of the water content of the section is represented by a smaller value, which indicates a more uniform fluid distribution in the profile. It is dimensionless. The maximum gradient of water cut along the well section reflects the risk of water coning / channeling; the smaller the value, the more stable the condition. The unit is 1 / m. This refers to the actual well section length, in meters (m). To account for the difference between total production and target production, and to ensure that well-level production meets the requirements of the large closed-loop allocation, Total output per segment, in meters (m). 3 / d, The target output is expressed in meters (m). 3 / d; α, β, γ are weighting factors, which are dimensionless.

[0060] Specifically, particle swarm optimization (PSO) is used to obtain the intelligent well completion method for each oil well with optimal production. The PSO algorithm introduces dynamic inertia weights to enhance convergence. The iterative optimization process of the PSO algorithm is as follows: Figure 6 As shown.

[0061] In the particle swarm optimization algorithm, the position of the particles... and speed Updates are performed according to the following rules: Speed ​​updates: (9) Location update: (10) In the formula, For particles in The speed of time; For particles in The position at that moment; It is a particle The best location found so far It is the optimal position found by all particles in the entire swarm; parameters It is the inertial weight that controls the search range of the particles; and It is the acceleration constant, which determines the magnitude of the influence of individual particles and social experience on velocity; and It is a random number, used to maintain the randomness and diversity of the search.

[0062] To prevent iterations from getting trapped in local optima, the particle swarm optimization algorithm uses production improvement as its optimization objective. Through multiple optimizations, it obtains the optimal production rate and number of segments at different water cuts for each oil well within different production cycles, as well as the ICV opening combination. The specific optimization process for ICV opening is as follows: Figure 6 As shown.

[0063] S6: Cooperative iteration and convergence determination.

[0064] The equivalent productivity index or skin coefficient obtained from near-well-scale optimization is fed back to the reservoir-scale surrogate model to update the reservoir distribution and prediction targets. If the improvement of the optimization target is less than the threshold or reaches the set time window, the control strategy is output; otherwise, the process returns to the outer optimization step.

[0065] Specifically, the near-well scale proxy model outputs the production rate, water cut, and pressure drop distribution of each well section, aggregating them into the equivalent production index or skin coefficient of the whole well; at the same time, the reservoir scale proxy model maps the updated wellbore equivalent parameters to boundary conditions (such as bottom hole pressure, phase supply rate, etc.) to re-predict the pressure field, saturation distribution, and recovery rate of the entire field.

[0066] The optimizer then calculates the updated objective function value. If the improvement in the objective function compared to the previous iteration is less than a preset threshold... (For example If the convergence rate is 0.5%, or if the set maximum number of iterations or optimization window (e.g., 30 days, 60 days) is reached, the system is considered to have converged, and the current optimal control strategy is output. If the convergence condition is not met, the system returns to the reservoir-scale outer layer optimization step, readjusts the multi-well control variables, and continues iterating.

[0067] S7: Online operation and calibration.

[0068] During field operation, based on real-time / near real-time data, the outer and inner optimization steps are periodically re-executed, and a sliding window method is used to perform light correction on the surrogate model. When moisture content, pressure or equipment limiting conditions are triggered, a rollback condition and conservative opening strategy are executed.

[0069] Specifically, the system re-executes the dual-loop optimization steps of the outer and inner layers within a set period (e.g., daily or weekly). During this period, a sliding window method is used to perform lightweight corrections on the surrogate model. Sliding window correction refers to using only production data from the most recent period (e.g., the past 15-30 days) to locally retrain or update the surrogate model parameters during each optimization, thereby enabling the model to capture the latest production dynamics and avoiding prediction bias caused by long-term model drift.

[0070] Meanwhile, to ensure the security of system operation, multi-level constraints and rollback mechanisms were set up: When the water cut of a well exceeds a preset threshold (e.g., 50%) or the bottom pressure is lower than the formation safety limit, the system immediately triggers the safety control module and automatically switches to conservative operating conditions, such as reducing the well opening, limiting production, or switching to standby mode.

[0071] If the equipment capacity (such as pumping capacity, tubing differential pressure) exceeds the upper limit, a rollback strategy is executed to restore the ICV opening to the most recent stable state.

[0072] The system will record abnormal events and narrow down the search space in the next round of optimization to avoid re-entering high-risk operating conditions.

[0073] This combined strategy of "periodic re-optimization + sliding window lightweight correction + safety backoff protection" ensures that the water control method can maintain a high recovery rate while guaranteeing the robustness and safety of the system during long-term dynamic operation.

[0074] The advantages and positive effects of this invention are: 1. This invention adopts a dual-scale surrogate model, which takes into account the dynamic characteristics of the reservoir far field and the wellbore near field, and realizes joint optimization of multiple wells and single wells.

[0075] 2. This invention introduces a neural differential equation surrogate model, which can capture the continuous dynamic process driven by oil and water, and has stronger generalization and extrapolation capabilities.

[0076] 3. This invention proposes a dual-layer control strategy of large closed loop and small closed loop, which takes into account both maximizing recovery rate and profile stability, and significantly delays water inrush.

[0077] 4. This invention supports online sliding window correction and a safety rollback mechanism to ensure the safety and real-time performance of control under model drift and parameter uncertainty conditions.

[0078] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A smart well completion dual-closed-loop control method based on a dual-scale surrogate model, characterized in that: Includes the following steps, S1: Obtain the geological and engineering parameters of the target reservoir and block. If production data is available, use the production data for model training and calibration. If no production data is available, generate sample data based on reservoir numerical simulation. Generate wellbore segment water control samples using a local fine model or coupled model. Clean, normalize and time-align the production data or the sample data. S2: Using neural differential equations, multi-well control parameters and static geological parameters are modeled to establish a reservoir-scale surrogate model for predicting formation pressure, saturation distribution, well production and recovery rate. S3: Using bottom hole flowing pressure, ICV segment opening vector and wellbore friction parameters as inputs, a near-well scale surrogate model is established based on neural differential equations, and the outputs well section production distribution, section water cut and pressure drop; S4: On the reservoir-scale proxy model, with the cumulative oil production, water cut and net present value as optimization objectives, constrain the upper and lower limits of formation pressure, the upper limit of water cut and the capacity of surface facilities, and solve for the target production indicators of each well level to achieve outer layer optimization of intelligent well completion. S5: On the near-well scale proxy model, with well section profile stability and water inrush delay as the objective function, constrain the total production and bottom hole pressure difference, optimize the well section ICV opening vector to achieve inner layer optimization for intelligent well completion; S6: Feed the equivalent productivity index or skin coefficient obtained by near-well scale optimization back to the reservoir scale proxy model to update the reservoir distribution and prediction target; S7: During field operation, based on real-time / near real-time data, S4-S5 are periodically re-executed to perform outer and inner layer optimization, and a sliding window method is used to perform light correction on the surrogate model. When moisture content, pressure or equipment limiting conditions are triggered, a rollback condition and conservative opening strategy are executed.

2. The intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model according to claim 1, characterized in that: In S2, the reservoir-scale surrogate model is established based on a neural differential equation structure to ensure the continuous expression and extrapolation ability of the prediction results to the dynamic laws.

3. The intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model according to claim 1 or 2, characterized in that: In S3, the near-well-scale surrogate model learns a differentiable mapping between the aperture vector and the equivalent skin coefficient to support gradient-based optimization solutions.

4. The intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model according to claim 1 or 2, characterized in that: In S4, the outer layer optimization is solved using the particle swarm optimization algorithm, and a dynamic inertia weight adjustment mechanism is adopted. Through multiple iterations of optimization and result comparison, it avoids getting trapped in local optima.

5. The intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model according to claim 1 or 2, characterized in that: In step S5, the objective function for the inner layer optimization is: In the formula, The water cut distribution in the well section is dimensionless. The variance of the water content of the section is represented by a smaller value, which indicates a more uniform fluid distribution in the profile. It is dimensionless. The maximum gradient of water cut along the well section reflects the risk of water coning / channeling; the smaller the value, the more stable the condition. The unit is 1 / m. This refers to the actual well section length, in meters (m). To account for the difference between total production and target production, and to ensure that well-level production meets the requirements of the large closed-loop allocation, Total output per segment, in meters (m). 3 / d, The target output is expressed in meters (m). 3 / d; α, β, γ are weighting factors, which are dimensionless.

6. The intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model according to claim 1 or 2, characterized in that: In step S6, if the improvement of the optimization target is less than the threshold or reaches the set time window, the control strategy is output; otherwise, return to step S4 to perform outer layer optimization.

7. The intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model according to claim 1 or 2, characterized in that: In S7, when the water cut of the well section exceeds the preset upper limit, the bottom pressure is lower than the preset lower limit, or the surface facility capacity is limited, the retreat condition is triggered, and the opening degree is adjusted to a conservative step value.

8. An intelligent well completion dual-closed-loop control system based on a dual-scale surrogate model, characterized in that: The intelligent well completion dual-closed-loop control method based on a dual-scale surrogate model as described in any one of claims 1 to 7 is characterized by comprising: Data acquisition and management module: used to acquire and manage reservoir, wellbore, and field operation data; Reservoir-scale proxy module: used to predict reservoir dynamics and production based on multi-well control parameters and static parameters; Near-well scale proxy module: used to predict the production / water distribution of the well section based on wellbore control parameters, and generate an equivalent production index or skin coefficient; The dual-layer optimization module consists of an outer multi-well optimization submodule and an inner ICV optimization submodule. Online calibration module: used to perform sliding window calibration of reservoir agent and near-well agent during real-time operation; Safety control module: used to execute a fallback strategy and conservative opening adjustment when threshold conditions are triggered; Execution and monitoring interface module: Used to interact with downhole intelligent completion tools to issue and transmit control commands.

9. The intelligent well completion dual-closed-loop control system based on a dual-scale surrogate model according to claim 8, characterized in that: The dual-layer optimization module supports hierarchical timescale separation and alternating iteration. The outer layer optimization is based on a daily / weekly cycle, while the inner layer optimization is based on an hourly / dayly cycle.

10. A computer-readable storage medium storing a computer algorithm, characterized in that, When the computer algorithm is executed by the processor, it performs the data processing as described in any one of claims 1 to 7.

Images