A decision-making method, equipment, and medium for tertiary oil recovery (EOR) schemes.

Abstract

This application discloses a decision-making method, equipment, and medium for tertiary oil recovery (EOR) schemes, relating to the field of oil and gas field development technology. The method includes: fusing multidimensional heterogeneous data of a target reservoir to obtain fused data; constructing a digital twin model corresponding to the target reservoir based on the fused data; using the digital twin model to rapidly extrapolate the short-term production dynamics of the target reservoir under certain triggering conditions, obtaining extrapolation results, and generating risk warning information based on the extrapolation results; responding to the risk warning information, using an optimization engine and the digital twin model as a virtual environment, executing an iterative optimization algorithm to generate candidate control strategies; inputting the candidate control strategies into the digital twin model for coupled simulation, obtaining reservoir dynamic prediction results output by the digital twin model; verifying the candidate control strategies based on the reservoir dynamic prediction results; and determining the candidate control strategy as the target control strategy and issuing it for execution if the candidate control strategy passes verification. This approach can improve the timeliness and accuracy of EOR decisions for continental sandstone reservoirs with high water cut and strong reservoir heterogeneity in the later stages.

Description

Technical Field

[0001] This application relates to the field of oil and gas field development technology, and in particular to a decision-making method, equipment and medium for a tertiary oil recovery (EOR) scheme. Background Technology

[0002] In oilfield development, high-water-cut, late-stage, and highly heterogeneous continental sandstone reservoirs are a key area of ​​concern. These reservoirs typically have undergone decades of water injection development, resulting in generally high overall water cuts and highly dispersed and complex distribution of remaining underground oil. The core challenge stems from the inherent strong heterogeneity of continental sediments, manifested in significant permeability differences (often tens to hundreds of times). This causes injected fluids to easily flow along high-permeability "dominant channels," creating ineffective circulation, while large amounts of crude oil in low-permeability areas remain untapped.

[0003] Existing tertiary oil recovery (EOR) control technologies for the aforementioned reservoirs involve a cycle that can take weeks or even months, from identifying problems (such as soaring water cut), analyzing causes, validating operational models, to developing and implementing new solutions. This decision-making model is incapable of addressing the rapid crossflow of underground fluids in high water-cut oilfields, ultimately leading to significant ineffective water circulation and continuous loss of production benefits. Furthermore, decision-making relies heavily on experience, resulting in poor targeting and low success rates. This leads to high investment costs, significant risks, and unstable efficiency and oil recovery effects, leaving a large amount of remaining oil unrecoverable economically and effectively. Summary of the Invention

[0004] This application provides a decision-making method, equipment, and medium for tertiary oil recovery (EOR) schemes to solve the following technical problem: how to solve the technical problem of untimely and inaccurate control decisions for continental sandstone reservoirs with high water cut and strong reservoir heterogeneity in the late stage of high water cut.

[0005] In a first aspect, embodiments of this application provide a decision-making method for a tertiary oil recovery (EOR) scheme. The method includes: fusing multidimensional heterogeneous data of a target reservoir to obtain fused data, wherein the target reservoir is a continental sandstone reservoir, and the overall water cut of the target reservoir is higher than a preset water cut threshold, and the reservoir permeability range is greater than a preset range threshold; constructing a digital twin model corresponding to the target reservoir based on the fused data, wherein the digital twin model is a multiphysics coupling model driven by real-time data to maintain synchronous evolution with the physical reservoir; and, under certain triggering conditions, using the digital twin model to... The short-term production dynamics of the target reservoir are rapidly extrapolated to obtain the extrapolation results, and risk warning information is generated based on the extrapolation results. In response to the risk warning information, an optimization engine is used to execute an iterative optimization algorithm in the digital twin model as a virtual environment to generate candidate control strategies. The candidate control strategies are input into the digital twin model for coupled simulation to obtain the reservoir dynamic prediction results output by the digital twin model. Based on the reservoir dynamic prediction results, the candidate control strategies are verified. If the candidate control strategy passes the verification, it is determined as the target control strategy and executed.

[0006] Secondly, embodiments of this application also provide a decision-making device for a tertiary oil recovery (EOR) scheme. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute a decision-making method for a tertiary oil recovery (EOR) scheme as described in the first aspect above.

[0007] Thirdly, embodiments of this application also provide a computer storage medium storing computer-executable instructions, which, when executed, implement a decision-making method for a tertiary oil recovery (EOR) scheme as described in the first aspect above.

[0008] The decision-making method, equipment, and medium for tertiary oil recovery (EOR) schemes provided in this application have the following beneficial effects: Data fusion can be performed on multidimensional heterogeneous data of the target reservoir to obtain fused data. Based on this fused data, a digital twin model corresponding to the target reservoir can be constructed. Then, under certain triggering conditions, the digital twin model can be used for rapid simulation, and risk warning information can be generated based on the simulation results. Subsequently, based on the risk warning information, an optimization engine is used, with the digital twin model as the virtual environment, to execute an iterative optimization algorithm to generate candidate control strategies. This allows for the rapid acquisition of relatively excellent control schemes. The candidate control strategies are input into the digital twin model for coupled simulation, obtaining the reservoir dynamic prediction results output by the digital twin model; this not only ensures the speed of prediction but also maintains the physical reliability of the prediction. Then, the candidate control strategies can be verified based on the reservoir dynamic prediction results. If the candidate control strategy passes verification, it is determined as the target control strategy and implemented. This improves the timeliness and accuracy of control decisions for continental sandstone reservoirs with high water cut and strong reservoir heterogeneity, ensuring the efficiency and effectiveness of EOR management. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a decision-making method for a tertiary oil recovery (EOR) scheme, provided as an embodiment of this application; Figure 2 A schematic diagram of a pressure field provided for an embodiment of this application; Figure 3 A schematic diagram of a residual oil saturation field provided for an embodiment of this application; Figure 4 This is a schematic diagram of the internal structure of a decision-making device for a tertiary oil recovery (EOR) scheme provided in an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] This application provides a decision-making scheme for tertiary oil recovery (EOR) programs. The technical solution proposed in this application will be described in detail below with reference to the accompanying drawings.

[0012] Figure 1 A flowchart illustrating a decision-making method for a tertiary oil recovery (EOR) scheme provided in this application embodiment. Figure 1 As shown in the embodiment of this application, a decision-making method for a tertiary oil recovery (EOR) scheme specifically includes the following steps: Step 101: Obtain fused data by performing data fusion on the multidimensional heterogeneous data of the target reservoir.

[0013] The target reservoir is a continental sandstone reservoir, and the overall water cut of the target reservoir is higher than the preset water cut threshold and the reservoir permeability difference is greater than the preset difference threshold.

[0014] In this embodiment, the target reservoir is a high-water-cut late-stage (high-water-cut late-stage refers to a development stage where the overall water cut is very high, mainly focused on tapping remaining oil, with the overall water cut exceeding a preset water cut threshold, typically greater than 90%), highly heterogeneous continental sandstone reservoir. Heterogeneity refers to the spatially uneven distribution of the physical properties (such as permeability and porosity) of the reservoir rocks. Strong heterogeneity emphasizes the degree of heterogeneity. Due to rapid changes in sedimentary environments (such as rivers and deltas), continental sandstone often exhibits very significant differences in rhythmic layers, interlayers, and permeability gradients, resulting in extremely strong heterogeneity. In practical applications, the reservoir permeability range of the target reservoir is greater than a preset range threshold, typically greater than 50.

[0015] In this embodiment, multidimensional heterogeneous data of the target reservoir can be fused to obtain fused data. In practical applications, reservoir data comes from a wide range of sources, such as geological exploration data, well logging data, and engineering dynamic data. These data can describe reservoir characteristics from different perspectives, but a single data source may have limitations or errors.

[0016] By fusing multidimensional heterogeneous data, we can combine the advantages of various data sources, eliminate inconsistencies between data, and thus improve data reliability, laying a solid data foundation for subsequent steps.

[0017] Step 102: Based on the fused data, construct a digital twin model corresponding to the target reservoir.

[0018] The digital twin model is a multi-physics coupling model driven by real-time data to maintain synchronous evolution with the physical reservoir.

[0019] In practical applications, a corresponding digital twin model can be constructed for the target reservoir based on fused data. Moreover, this digital twin model, driven by real-time data, can make its computational state infinitely close to the actual state of the physical reservoir, thereby solving the problem of lagging updates in traditional models and ensuring that the digital twin model always maintains dynamic consistency with the physical reservoir.

[0020] Step 103: When the triggering conditions are met, the digital twin model is used to quickly extrapolate the short-term production dynamics of the target reservoir, obtain the extrapolation results, and generate risk warning information based on the extrapolation results.

[0021] In practical applications, triggering conditions can be periodic, event-driven, or include early warnings, without specific limitations. Under normal circumstances, rapid simulations can be automatically triggered at preset, relatively long intervals (e.g., every 4 hours or per shift). However, rapid simulations can also be triggered when specific events occur, such as the opening of a new well, the opening of an old well after operation, the switching of the injection system (water to polymer), the receipt of new production or water injection targets from superiors, the start-up or shutdown of key pumps or compressors, or the abnormally rapid decrease in the flowing pressure of a certain well. Furthermore, if the results of this simulation are serious, the next rapid simulation can be conducted in advance.

[0022] The simulation results are generally divided into two types. One is when no risks are identified (economic benefit risks, production safety risks, formation resource risks, etc.), such as "no clear breakthrough risks were detected, and all indicators are within the expected range." In this case, the generated risk warning information is generally a system health status report. However, if risks are identified, for example, model predictions show that the injected fluid velocity is abnormally accelerated in a specific high-permeability band, with highly concentrated streamlines. The water / polymer breakthrough time of the production well is predicted significantly earlier (e.g., the polymer breakthrough is predicted to occur in 10 days, much shorter than the expected 60 days). The risk warning information can include: the spatial location of the identified high-permeability channel, the predicted location and time of the breakthrough water channeling, and the expected cumulative water production and cost losses due to ineffective circulation. For example, in the above example, the risk warning information could be: Risk type: dominant channel water channeling risk (ineffective water circulation); Spatial path: Injection well I-107 -> high-permeability band (main channel sand body, permeability > 2000 mD) -> Production well P-209; Geological layer: Sha-2 section, 3 sub-layers. Risk Mechanism: Long-term water injection causes the injected water to "finger" along historically formed high-permeability channels, resulting in uneven displacement at the leading edge and the formation of a water flow "highway." Current Status: Real-time water cut of well P-209: 92.1%; Flow rate through the target channel: 38%; Predicted breakthrough time: 20:00 (±4 hours); Predicted peak water cut: 96.5% (expected to be reached within 24 hours after breakthrough); Predicted ineffective water production: After breakthrough, the average daily ineffective water production is expected to increase by 180 cubic meters / day. Expected daily profit loss: Approximately 10,000 yuan / day. Risk Duration: Without intervention, this high-risk state is expected to last for 5-7 days.

[0023] Step 104: In response to the risk warning information, an optimization engine is used to execute an iterative optimization algorithm in the form of the digital twin model as a virtual environment to generate candidate control strategies.

[0024] In practical applications, a multi-objective optimization engine can be built-in to maximize recovery, minimize operating costs (such as injector costs), or control gas and water channeling time. This optimization engine can utilize iterative optimization algorithms, such as reinforcement learning or evolutionary algorithms, to generate multiple possible control schemes. For example, adjusting the gas / water injection rate, injection-production ratio, flowing pressure of production wells, or even changing the slug combination, will ultimately yield candidate control strategies. Generally, the time period corresponding to the candidate control strategies is shorter than the time period for rapid deduction mentioned above.

[0025] If the risk warning information indicates no urgent risk, the goal of optimizing the engine can be to find opportunities to improve efficiency. Based on the current stable operation, it's possible to explore whether fine-tuning can further increase oil production, reduce water cut, or save energy. If the risk warning information indicates an urgent risk, the aforementioned risk warning information needs to be considered. For example, optimization goals could include suppressing or avoiding the specific risks identified in the risk warning information, thereby proactively mitigating the risk.

[0026] For example, the final candidate control strategy could be: T+0h: Reduce the daily water injection rate of injection well I-107 from 200 m³ / d to 170 m³ / d in steps. T+2h: Lower the bottomhole flowing pressure setpoint of production well P-209 by 0.3 MPa, temporarily increasing its daily fluid production from 50 m³ / d to 55 m³ / d to increase the production pressure differential and "pull" out crude oil from the near-wellbore area. T+6h: Initiate pulse water injection (on for 2 hours, off for 1 hour) to the adjacent injection well I-106 to attempt to disturb the flow field.

[0027] Step 105: Input the candidate control strategy into the digital twin model for coupled simulation to obtain the reservoir dynamic prediction results output by the digital twin model.

[0028] In practical applications, candidate control strategies can be input into the aforementioned digital twin model for coupled simulation, thereby obtaining the reservoir dynamic prediction results for the target reservoir within a preset future time period. The reservoir dynamic prediction results are not specifically limited and can include key production indicators, such as pressure field, saturation field, displacement front location, as well as recovery rate, water cut, and gas-oil ratio. Specifically, it can be a pressure field evolution sequence, a schematic diagram of the pressure field at future times T1, T2, ..., Tn. The saturation field (remaining oil distribution) evolution sequence includes schematic diagrams of oil saturation at future times T1, T2, ..., Tn, i.e., a dynamic remaining oil distribution map. It can also be a daily oil / liquid production curve, a cumulative oil production curve, for example, predicting that the daily oil production of the block will gradually increase from 50 tons to 65 tons in the next 7 days; it can also be a comprehensive water cut curve, for example, predicting that the water cut will first slightly increase from 92.1% to 92.5%; or a water injection / fluid injection pressure curve, for example, the predicted water injection wellhead pressure of I-107 will increase from 15.2 MPa to 15.8 MPa. Single well index curves: water cut, oil production, and flowing pressure curves of a specified well. Example values: the predicted water cut curve data for the P-209 production well are 0h: 92.1%, 12h: 92.8%, 48h: 93.2%, 72h: 93.5%.

[0029] In practical applications, reservoir dynamic prediction results can also include technical effect assessments. For example, the magnitude of improved oil recovery (EOR) can be estimated, such as predicting that this strategy can increase EOR by 0.05 percentage points within 7 days; the oil exchange rate / water retention rate can also be assessed, such as predicting a polymer oil exchange rate of 120 tons of oil per ton of polymer. Simultaneously, rapid economic assessments can be included, such as expected operating costs, such as an estimated increase in reagent costs of 50,000 yuan over 7 days, resulting in an increased crude oil production value of 350,000 yuan. In practical applications, reservoir dynamic prediction results can also include risk data, such as pressure risk, such as predicting no overpressure risk, with all node pressures below 85% of the fracture pressure; and water channeling risk, for example, predicting that the water cut surge trend in well P-209 is effectively contained, with the peak value not exceeding 93.5%.

[0030] For example, the reservoir dynamics prediction results corresponding to the candidate control strategies for target reservoir B are as follows: First, a summary of key indicator predictions can be included. Cumulative oil production increase in the block: +150 tons; Water cut of well P-209: increased from 92.1% to 93.5% (preventing the risk of a jump to 96.5%); Injection pressure of well I-107: gradually increased from 15.2 MPa to 15.8 MPa; Enhanced oil recovery: +0.05%.

[0031] Secondly, it can include keyframes for the evolution of the core field map. For example... Figure 2 As shown, at T=24h: the pressure field indicates that the rate of expansion of the high-pressure front of well I-107 towards well P-209 has slowed. For example... Figure 3As shown, at T=72h: the remaining oil field indicates that the remaining oil in the main flow direction of well P-209 is effectively driven, but the improvement in lateral spillover is limited. Finally, the risk assessment conclusions can also be included: Water channeling risk in the dominant channel: effectively suppressed. The predicted channel flow rate share decreased from 38% to 28%. Pressure safety risk: none. Economic efficiency: the input-output ratio is approximately 1:6.

[0032] The aforementioned candidate scheduling strategies are the boundary conditions that the digital twin model needs to apply at each computation time step. Based on this instruction, the model can calculate how the fluid moves underground.

[0033] In this embodiment, the aforementioned digital twin model is a computational model that can be driven and calibrated using data synchronized with a physical entity (target reservoir), and can simulate the future state of that physical entity. Using this digital twin model for simulation ensures the reliability of the predictions.

[0034] In practical applications, there may be more than one candidate control strategy. In this case, a copy of the digital twin model can be established based on the number of candidate control strategies. Then, each candidate scheduling strategy can be input into the corresponding copy to obtain the reservoir dynamic prediction result corresponding to each candidate control strategy. This can save time and improve efficiency.

[0035] Step 106: Based on the reservoir dynamic prediction results, verify the candidate control strategies.

[0036] In practical applications, in order to ensure the reliability of decision-making, candidate control strategies cannot be used directly. Instead, they can be input into a digital twin model to obtain reservoir dynamic prediction results, and then the candidate control strategies can be verified. This ensures that the decision-making is both safe and effective.

[0037] In practical applications, the simulation and verification steps described above can prevent the "optimal" strategy from becoming the "most dangerous" instruction. Some strategies score highly in the simplified evaluation of the optimization engine, but high-fidelity simulation may reveal fatal risks such as pressure over-limit, accelerated gas flow, and equipment overload. Meanwhile, the complete and authoritative reservoir dynamic prediction results generated (e.g., detailed pressure fields, remaining oil distribution maps, and economic indicators) can serve as a quantitative basis for using the aforementioned candidate control strategies. Furthermore, when the optimization engine is internally seeking optimization, it may use simplified models or short-sighted rewards in pursuit of efficiency. The final verification uses a more accurate model with a longer horizon to review and identify and correct strategies that offer high short-term gains but significant long-term damage, ensuring that the recommended solution is a truly sustainable global optimum.

[0038] Step 1076: If the candidate control strategy passes verification, determine the candidate control strategy as the target control strategy and issue it for execution.

[0039] In this embodiment, once a candidate control strategy passes verification, it can be determined as the target control strategy and executed. This means the optimized control command can be sent to the automated actuators in the field (such as smart wellheads or digital valves). The results of the execution are then fed back into the digital twin model using new real-time data, forming an optimized control process. This approach achieves a shift from manual experience-based analysis to AI-driven autonomous optimization, enabling the rapid acquisition and automatic execution of superior control solutions, significantly improving the efficiency and effectiveness of EOR management.

[0040] In this embodiment, multidimensional heterogeneous data of the target reservoir can be fused to obtain fused data. Based on this fused data, a digital twin model corresponding to the target reservoir is constructed. Then, under certain triggering conditions, the digital twin model is used for rapid simulation, and risk warning information is generated based on the simulation results. Subsequently, based on the risk warning information, an optimization engine is used, with the digital twin model as the virtual environment, to execute an iterative optimization algorithm to generate candidate control strategies. This allows for the rapid acquisition of relatively excellent control schemes. The candidate control strategies are input into the digital twin model for coupled simulation to obtain the reservoir dynamic prediction results output by the digital twin model; this not only ensures the speed of prediction but also maintains the physical reliability of the prediction. Then, the candidate control strategies can be verified based on the reservoir dynamic prediction results. If the candidate control strategy passes verification, it is determined as the target control strategy and executed. This improves the timeliness and accuracy of control decisions for continental sandstone reservoirs with high water cut and strong reservoir heterogeneity, ensuring the efficiency and effectiveness of EOR management.

[0041] In one possible implementation, the fused data obtained by fusing multidimensional heterogeneous data of the target reservoir includes: Acquire multidimensional heterogeneous data of the target reservoir, wherein the multidimensional heterogeneous data includes: geological static data, engineering dynamic data, and scheme design data; The multidimensional heterogeneous data are fused in the same spatiotemporal coordinate system using a spatiotemporal alignment algorithm to obtain fused data.

[0042] In the above embodiments, geological static data (the original state data of the target reservoir before it is exploited, which determines the storage and flow space of oil and gas, such as seismic interpretation, well logging interpretation, core analysis data, etc.) can be collected in real time, and engineering dynamic data (data generated by the target reservoir during the production process and changing over time, which can reflect its dynamic response, such as wellhead / downhole pressure, temperature, flow rate, injection concentration, production profile data, etc.) can be received simultaneously. Simultaneously, scheme design data (data on various engineering schemes formulated and implemented to improve development effectiveness, such as injection-production schemes, slug sizes, injection well trajectories, etc.) can be received.

[0043] For example, the multidimensional heterogeneous data of target reservoir A may include a complete set of digital logging curves (such as natural gamma ray GR, deep, medium and shallow resistivity, sonic transit time, neutron porosity, density, etc.) of 500-meter core samples obtained from 10 key wells, as well as experimental analysis data (porosity, permeability, particle size analysis, mercury injection capillary pressure curve, relative permeability curve, etc.) of 500-meter core samples from 10 key wells; oil production, water production, water injection volume, injection pressure, etc. of each well; and injection and production plans (the daily water injection volume of well I-101 will be maintained at 200 cubic meters next month, and the daily fluid production of well P-205 will be controlled at 50 cubic meters).

[0044] Subsequently, spatiotemporal alignment algorithms can be used to fuse the aforementioned multidimensional heterogeneous data from different sources, frequencies, and dimensions within the same spatiotemporal coordinate system. Specifically, this includes spatial alignment, such as mapping well point data to 3D geological grid nodes based on well trajectory coordinates and seismic grids, and generating attribute fields through kriging or inverse distance weighted interpolation; and temporal alignment, such as using Kalman filtering or interpolation methods to synchronize high-frequency and low-frequency data to a unified timestamp, and using time-series prediction models to fill in missing data. This lays a solid data foundation for the subsequent construction of digital twin models.

[0045] In practical applications, a digital twin model refers to a computational model capable of synchronously mapping and dynamically predicting the physical state of an oil reservoir. Its core technological essence lies in its data-driven capability for synchronization and prediction; in practical applications, it can be a gridded numerical model.

[0046] In one possible implementation, constructing a digital twin model of the target reservoir based on the fused data includes: Based on the geological static data in the fused data, a three-dimensional geological grid model corresponding to the target reservoir is established; Based on the fused data, an initial attribute field is assigned to the grid cells in the three-dimensional geological grid model, wherein the initial attribute field includes at least one of the following: porosity field, permeability field, fluid saturation field, pressure field, and temperature field. In the three-dimensional geological grid model, governing equations for describing the physicochemical processes in the target reservoir are coupled to form a first model, wherein the governing equations include equations describing seepage mechanics, geomechanics, thermodynamics and chemical reaction processes; Based on the fused data, the first model is assigned initial rock physical properties and fluid state to obtain the second model; The second model is configured to evolve synchronously with the physical reservoir through real-time data-driven processes to obtain a digital twin model.

[0047] In practical applications, during the construction of a digital twin model, a blank multiphysics coupling model (i.e., the first model) can be built first based on the fused data. The first model includes a precise three-dimensional mesh skeleton containing physical and chemical laws, but all mesh properties (such as porosity, pressure, and saturation) are still at default values ​​or zero.

[0048] Subsequently, based on the aforementioned fused data, the first model can be assigned initial rock physical properties and fluid states. This means extracting known geological static properties at the current moment (the starting point of the real-time simulation) from the fused data. For example, data such as porosity, permeability, net-to-gross ratio, and rock compressibility coefficient obtained from well logging interpretation and core analysis can be read and populated into the corresponding grids of the first model. Thus, the first model transforms from a geometrically empty shell into an entity with precise rock physical properties (the second model). Assigning initial rock physical properties can be understood as labeling each subsurface grid cell. For example, Label 1: Porosity = 0.25 (rock has 25% storage space). Label 2: Permeability = 100 millidarcy (fluid flows relatively easily). Label 3: Rock type = sandstone. Assigning initial fluid states indicates what is inside each grid cell at the current moment and what the pressure is. State 1: Oil saturation = 0.65 (65% of the space in this grid is oil). State 2: Water saturation = 0.35 (35% is water). State 3: Pressure = 20 MPa. After completing the above operations, a physically complete and state-accurate model can be obtained.

[0049] Then, the second model can be configured. Specifically, a real-time data pipeline can be configured, creating a data receiver for each real well (e.g., I-1, P-1, P-2). For example, the wellhead pressure sensor and injection flow meter of the real well I-1 can be identified, and their real-time readings (e.g., once per second) can be continuously sent to the corresponding "data receiver" in the digital twin model via an IoT network. The same operation can be performed on wells P-1 and P-2, connecting their bottomhole pressure gauges and production / water cut meters. In this way, every pressure fluctuation and fluid flow in the target reservoir can be sensed by the digital twin model within 1 second. Simultaneously, a data assimilation self-calibration engine can be configured. This is the core intelligent configuration, enabling the model to perform automatic calibration. For example, a Kalman filter algorithm package (a standard and powerful data assimilation tool) can be selected and installed on the backend of the aforementioned digital twin model. Then, the data assimilation self-calibration engine can be configured, for example, by configuring the trigger timing: setting it to run automatically every 6 hours. Then, the comparison content can be set: each time it runs, a set of the latest data can be automatically captured. For example, the actual measured bottom-hole flowing pressure of the real P-1 well at this moment is 15.2 MPa. However, the model calculates the P-1 well pressure as 14.8 MPa based on its current parameters, resulting in an error of 0.4 MPa. The model can reflect on this and, through complex mathematical inversion, determine the most likely cause: for example, the permeability of a grid about 50 meters southwest of the P-1 well is overestimated by 20%. Therefore, the permeability value of that grid can be automatically reduced by 20%. In this way, the model reflects on and fine-tunes itself every 6 hours. After several days or weeks of continuous calibration, the parameters within the model (especially the permeability field) can become increasingly closer to the actual underground conditions. Simultaneously, the digital twin model possesses learning and calibration capabilities. Finally, state synchronization and prediction loops can be configured. State synchronization configuration means that before each new prediction is made, state synchronization is forced. For example, the latest pressure data received from all wells is automatically used to generate a current pressure field covering the entire reservoir through an interpolation algorithm, and then completely overlaid on the pressure field of the digital twin model. This ensures that the model always predicts the future from the most realistic present, rather than continuing from its previous prediction, thus avoiding error accumulation. Simultaneously, prediction-driven configurations can be implemented, such as building a command receiver that triggers the model to start a new round of simulation when a production control command is received.

[0050] The digital twin model constructed in this way not only includes static attributes but also indicates a complete starting point representing the actual fluid distribution and energy state of the target reservoir at the current moment. Thus, the aforementioned digital twin model is not merely a static three-dimensional geological grid but also a dynamically evolving digital copy. Driven by real-time data, the digital twin model ensures a high degree of synchronization between the digital model and the physical reservoir's state. The digital twin model obtained through this method is essentially tailor-made for the target reservoir, more accurately reflecting its physical state.

[0051] In practical applications, before using a digital twin model for coupled simulation, a very short simulation (e.g., 0 days) can be performed on the digital twin model to check whether the model is in physical equilibrium under the set initial conditions (e.g., whether the pressure distribution is stable, and whether the fluid undergoes unreasonable instantaneous flow). This ensures that the input initial state is self-consistent and stable, and will not lead to numerical explosions or physically impossible phenomena at the beginning of the simulation.

[0052] In practical applications, digital twin models can also be models trained based on deep learning.

[0053] In another possible implementation, constructing a digital twin model of the target reservoir based on the fused data includes: Obtain the multiphysics coupled basic model corresponding to the target reservoir; The multiphysics coupling basic model is fine-tuned using the fused data of the target reservoir to obtain a digital twin model corresponding to the target reservoir.

[0054] In practical applications, various virtual reservoir geological models can be constructed, differing in structure, lithology, physical properties, and fluid distribution. For these models, the dynamic evolution of their multiphysics processes under different exploitation strategies is simulated, generating massive amounts of simulation data pairs. Each data pair includes input conditions and the corresponding multiphysics state response. Using these massive simulation data pairs, an initial neural network model is trained, enabling it to learn and internalize the dynamic response laws of multiphysics coupling in the reservoir, thus obtaining a basic multiphysics coupling model.

[0055] In practical applications, the initial neural network model mentioned above can be a physical information neural network, whose training loss function includes the residual terms of the physical control equations and the initial and boundary condition matching terms; the multiphysics dynamic evolution process can include seepage mechanics processes and chemical reaction processes.

[0056] In the above embodiments, fine-tuning of the basic model can be achieved through constraint fine-tuning. This involves using the geological static data, determined initial fluid state data, and historical production dynamic data from the fused data of the target reservoir as hard constraints or strong monitoring signals to drive the parameters of the multiphysics coupled basic model to update in a direction that conforms to the actual situation of the target reservoir. Alternatively, an efficient parameter fine-tuning method can be employed, updating only the parameters of some layers in the multiphysics coupled basic model or injecting a small number of adapter parameters, achieving rapid adaptation while maintaining the model's general knowledge. Although training a high-quality general basic model requires significant computing power and data, the resulting digital twin model can save considerable time in subsequent fine-tuning. Furthermore, it can reduce costs and improve efficiency when analyzing multiple reservoirs.

[0057] In practical applications, the fine-tuning process can adopt a strategy that combines transfer learning and meta-learning. The fused data of the target reservoir is divided into multiple time stages or geological region subsets. In a multi-task learning manner, the above-mentioned multi-physics coupled basic model can simultaneously learn and adapt to the features of different stages or subsets, thereby improving its ability to capture and generalize the strong heterogeneity and high water content of the target reservoir.

[0058] Similarly, the digital twin model obtained through the above method can be verified. For example, the simulation results of the digital twin model on historical production data can be compared with the actual historical observation data. If the fitting error of the key indicators is within the preset threshold, the verification is successful. Otherwise, the fine-tuning strategy can be adjusted or specific types of training data can be added for fine-tuning.

[0059] In practical applications, digital twin simulation is a process of synchronizing the state, setting conditions, iteratively calculating the future based on physical laws, and then compiling and outputting the results. Essentially, it's a high-fidelity simulation of possible future scenarios in the real world, strictly following physical laws within the digital world. The simulation process of the aforementioned digital twin model can be described as driving the solver within the model to numerically solve for future physicochemical changes in the reservoir based on the boundary and initial conditions defined by production control commands.

[0060] In one possible implementation, the use of an optimization engine, with the digital twin model as a virtual environment, to execute an iterative optimization algorithm to generate candidate control strategies includes: An optimization engine is used in the digital twin model to generate at least one tentative operation command based on the fused data; For each of the aforementioned exploratory operation commands, the digital twin model is invoked for rapid simulation to predict the reservoir state changes caused by the exploratory operation command and to calculate the reward value corresponding to the exploratory operation command. Based on the exploratory operation instructions, reward values, and changes in reservoir state, the decision model within the optimization engine is updated; Repeat the above steps until the preset exploration termination condition is met; Based on the updated decision model, an operation sequence starting from the current state and covering a preset future time period is obtained, and the operation sequence is encapsulated as a candidate control strategy.

[0061] In situations where risks exist, the aforementioned optimization process primarily aims to increase production while controlling or resolving a identified risk. In practical applications, the optimization objective of the iterative optimization algorithm can include the suppression or avoidance of specific risks identified by risk warning information; specifically, the generation and evaluation of tentative operation instructions prioritize operations that can mitigate the specific risk. When risk warning information indicates the presence of a dominant seepage channel leading to ineffective water circulation, the optimization objective includes a penalty term for the flow rate through that channel and / or a reward term for the fluid affecting the inefficient area connected by that channel. Simultaneously, when generating tentative operation instructions, their action space or sampling range can be constrained or weighted based on the well location, stratigraphy, or region involved in the risk warning information to focus the optimization direction.

[0062] In practical applications, the aforementioned iterative optimization algorithm can be a reinforcement learning algorithm. In this case, the decision model within the optimization engine includes a policy network and a value network. The policy network outputs a probability distribution of tentative operation commands based on the input fused data (which can indicate the current state of the target reservoir). The value network evaluates the long-term value of executing specific operation commands under specific reservoir conditions. The reinforcement learning algorithm can employ proximal policy optimization or deep deterministic policy gradient algorithms. Furthermore, the reward function can be set to provide positive rewards for oil production and negative rewards for water production and high-permeability channel flow, driving the optimization direction towards both increasing oil recovery and controlling water volume.

[0063] In practical applications, the above iterative optimization algorithm can also be an evolutionary algorithm; in the exploration steps, the multiple tentative operation instructions generated can constitute a population; in the virtual evaluation step using a digital twin model, the fitness of each individual (instruction) in the population is evaluated, and the fitness is the reward value; the step of updating the decision model inside the optimization engine can be selection, crossover, and mutation operations to generate a new generation of population; after the evolution terminates, the operation sequence represented by the individual with the highest fitness in each generation of population can be used as a candidate regulatory strategy.

[0064] In practical applications, rapid simulation can invoke a proxy model based on a physical information neural network within the digital twin model to achieve millisecond-level response, meeting the high-frequency interaction requirements of the aforementioned iterative optimization algorithm. Furthermore, when generating tentative operation commands, Gaussian noise or random noise following an Ornstein-Uhlenbeck process can be introduced to balance the algorithm's exploration and utilization.

[0065] In one possible implementation, the digital twin model is coupled for simulation in real time in the following manner: For the conventional physical field changes of the target reservoir, a fast solver based on physical information neural network is used for near real-time simulation; To address the complex physicochemical processes of the target reservoir, a high-precision numerical simulator is periodically triggered to calibrate the fast solver.

[0066] In the above embodiments, to balance computational speed and accuracy, the digital twin model can employ a hybrid approach of mechanistic and AI models for simulation. For common physical field changes (such as pressure diffusion and saturation advection), a fast solver based on Physics-Informed Neural Networks (PINNs) can be trained. PINNs incorporate the governing equations, initial conditions, and boundary conditions as part of the loss function, approximating the solution function through a deep neural network. The trained network can complete field prediction for a given time step in milliseconds, hundreds of times faster than traditional numerical simulations. This allows for a several-order-of-magnitude improvement in computational speed while maintaining accuracy, achieving near real-time simulation.

[0067] Simultaneously, for regions or times involving complex physicochemical processes (such as polymer shear degradation, emulsion formation, chemical mass transfer, and phase changes), a high-precision numerical simulator is periodically (e.g., every 10 time steps) triggered to perform full-physics field calculations. The calculation results are then used to correct the output of the aforementioned fast solver, preventing error accumulation. The calculation process of the high-precision numerical simulator recreates the evolution of the target reservoir over a future period in the digital world using the most rigorous methods, the finest scale, and the most complete physical laws, pursuing absolute physical accuracy and rigorous numerical convergence.

[0068] The specific process is as follows: Load an ultra-fine three-dimensional mesh with tens of millions of pixels and activate all physicochemical equations.

[0069] Crawling: Starting from the current moment, using extremely short time steps (such as a few minutes), it moves step by step towards the future.

[0070] Each step of the solution: At each step, a giant set of equations containing tens of millions of unknowns needs to be established and solved to describe the pressure, flow rate, and composition changes in every corner of the target reservoir at this moment.

[0071] The process of iterating and solving the system of equations is repeated hundreds of thousands of times until the entire prediction period (e.g., the next 7 days) is covered. The result is a dynamic dataset of the future with extremely high temporal resolution and complete spatial detail, considered the most reliable physical answer.

[0072] Subsequently, the above results can be used to correct errors in the fast surrogate model. Finally, the reservoir dynamic prediction results for the preset future time period can be output. In this way, the digital twin model integrates a high-precision mechanistic model and a fast solver, which not only ensures the speed of calculation but also maintains the physical reliability of the prediction. It can more accurately capture the complex dynamics of the EOR process and obtain more accurate reservoir dynamic prediction results.

[0073] In one possible implementation, the verification of the candidate control strategy based on the dynamic prediction results includes: Based on the dynamic prediction results, it is determined whether the key parameters of the target reservoir exceed the safety threshold corresponding to the key parameters after the candidate control strategy is implemented. Based on the dynamic prediction results, it is determined whether the comprehensive score of the target reservoir in terms of recovery rate, economic cost and environmental indicators exceeds the scoring threshold after the implementation of the candidate control strategy.

[0074] In practical applications, the candidate control strategies can be validated for both safety and multi-objective effectiveness based on the aforementioned dynamic prediction results. Safety validation determines whether key parameters exceed safety thresholds after the strategy is implemented. Multi-objective effectiveness validation quantifies the comprehensive performance of the candidate control strategy in terms of recovery rate, economic cost, and environmental indicators. This ensures the safety and practicality of the candidate control strategies, guaranteeing not only the safety of the target reservoir but also the recovery effect.

[0075] In practical applications, when there are multiple candidate control strategies, all of which have been validated, a multi-objective optimization decision model can be used to comprehensively consider factors such as the recovery rate improvement, input costs, and environmental indicators (e.g., carbon footprint), and methods such as TOPSIS can be employed to recommend the target control strategy. This ensures the quality of the target control strategy.

[0076] In one possible implementation, before employing an optimization engine, using the digital twin model as a virtual environment, and executing an iterative optimization algorithm to generate candidate control strategies, the method further includes: Acquire health status prediction information and material supply chain status information of key equipment associated with the production of the target oil reservoir within the future preset time period; The equipment health status prediction information and the material supply chain status information are transformed into constraints or part of the objective function of the optimization engine.

[0077] In practical applications, when adjusting candidate control strategies, we can identify potential future equipment performance degradation or failure events and their impact based on equipment health status prediction information; we can also identify potential future material supply shortages or delays and their impact based on material supply chain status information; then, during the optimization process, when constructing the optimization engine structure, the equipment health status prediction information and material supply chain status information are transformed into constraints or part of the objective function of the optimization engine; and then, by solving this extended optimization engine, candidate control strategies with disturbance resistance capabilities can be generated.

[0078] In practical applications, converting equipment health status prediction information into constraints allows for the identification of potential performance degradation or unplanned downtime events that may occur in specific equipment within a predetermined optimization period. The optimization engine then imposes a capacity limit constraint on the corresponding equipment that matches the predicted performance degradation or downtime event. For example, if it is predicted that the efficiency of injection pump P-101 will decrease by 20% starting from the 5th day, the optimization model will lower the maximum allowable injection volume limit of the associated well by 20% starting from the 5th day.

[0079] In practical applications, transforming material supply chain status information into constraints allows for the identification of expected arrival times and quantities of key chemical agents based on this information. In the optimization engine, available agent inventory and replenishment plans are converted into constraints regarding the total amount of agent injected, injection timing, or injection concentration. For example, if supply chain information indicates that polymer agent inventory is only sufficient for the next 10 days, and the next replenishment will arrive in 15 days, then the optimization model imposes a constraint: the cumulative polymer injection amount within the next 15 days must not exceed the current inventory level, and an optimization strategy is generated that includes "gradually slowing down the agent injection rate until it is paused, and resuming after the replenishment arrives."

[0080] In practical applications, when equipment health status prediction information indicates that the risk of equipment performance degradation exceeds a predetermined threshold, or when material supply chain status information indicates that the risk of supply interruption exceeds a predetermined threshold, preventive maintenance work orders or emergency material purchase orders can be automatically generated and triggered.

[0081] In practical applications, in order to generate candidate control strategies with disturbance resistance capabilities, the injection volume distribution between different injection wells or well groups can be adjusted to bypass equipment that is about to experience performance degradation; the type or concentration of injected chemicals can be adjusted to match current and future material inventory; the implementation time of a certain development measure can be advanced or postponed to match the overhaul window of critical equipment or the delivery cycle of core materials; and backup equipment, pipelines or injection processes can be activated, etc.

[0082] In practical applications, each oil well has different specific conditions. In order to make more precise control, an optimization engine can be used to generate a target control strategy. Then, the edge intelligent controller deployed at the production well or injection well end can perform autonomous fine-tuning control of local parameters based on the above target control strategy and local real-time data.

[0083] In practical applications, based on the remaining oil saturation field of the target reservoir predicted by the aforementioned digital twin model, the system can automatically delineate target areas with remaining oil enrichment where the oil saturation exceeds a threshold. When the potential of the target area meets the conditions, the system can automatically generate and simulate evaluations of infill wells, sidetracking wells, or depth-controlled wells for that target area, and output recommended development suggestions. This can promote the precise development of target reservoirs, improve the success rate of new wells, and increase economic benefits.

[0084] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a decision-making device for tertiary oil recovery (EOR) schemes, the structure of which is as follows: Figure 4 As shown.

[0085] Figure 4 This is a schematic diagram of the internal structure of a decision-making device for a tertiary oil recovery (EOR) scheme provided in an embodiment of this application. Figure 4 As shown, the device includes: At least one processor 401; And a memory 402 that is communicatively connected to at least one processor; The memory 402 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 401 so that at least one processor 401 can: execute the decision-making method of the above-mentioned tertiary oil recovery (EOR) scheme.

[0086] In one possible implementation, the processor 401 can perform data fusion on multidimensional heterogeneous data of a target reservoir to obtain fused data, wherein the target reservoir is a continental sandstone reservoir, and the overall water cut of the target reservoir is higher than a preset water cut threshold, and the reservoir permeability range is greater than a preset range threshold; based on the fused data, a digital twin model corresponding to the target reservoir is constructed, wherein the digital twin model is a multiphysics coupling model driven by real-time data to maintain synchronous evolution with the physical reservoir; under the condition of meeting the triggering condition, the digital twin model is used to analyze the undefined features of the target reservoir. The system rapidly extrapolates short-term production dynamics to obtain extrapolation results and generates risk warning information based on these results. In response to the risk warning information, an optimization engine is used, employing the digital twin model as a virtual environment, to execute an iterative optimization algorithm to generate candidate control strategies. These candidate control strategies are then input into the digital twin model for coupled simulation to obtain reservoir dynamic prediction results output by the digital twin model. Based on the reservoir dynamic prediction results, the candidate control strategies are verified. If the candidate control strategy passes verification, it is determined as the target control strategy and executed.

[0087] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium stores computer-executable instructions, which are configured to execute the decision-making method for the aforementioned tertiary oil recovery (EOR) scheme.

[0088] In one possible implementation, the aforementioned computer-executable instructions are configured to perform data fusion on multidimensional heterogeneous data of a target reservoir to obtain fused data, wherein the target reservoir is a continental sandstone reservoir, and the overall water cut of the target reservoir is higher than a preset water cut threshold, and the reservoir permeability range is greater than a preset range threshold; based on the fused data, a digital twin model corresponding to the target reservoir is constructed, wherein the digital twin model is a multiphysics coupling model driven by real-time data to maintain synchronous evolution with the physical reservoir; under the condition of meeting the triggering conditions, the digital twin model is used to analyze the target reservoir. The system rapidly extrapolates the short-term production dynamics of the reservoir, obtains the extrapolation results, and generates risk warning information based on the extrapolation results. In response to the risk warning information, an optimization engine is used, with the digital twin model as the virtual environment, to execute an iterative optimization algorithm to generate candidate control strategies. The candidate control strategies are input into the digital twin model for coupled simulation, obtaining the reservoir dynamic prediction results output by the digital twin model. Based on the reservoir dynamic prediction results, the candidate control strategies are verified. If the candidate control strategy passes verification, it is determined as the target control strategy and executed.

[0089] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0090] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0091] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0096] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0097] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0098] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0099] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A decision-making method for tertiary oil recovery (EOR) schemes, characterized in that, The method is used for high-water-cut, late-stage, highly heterogeneous continental sandstone reservoirs. The method includes: By fusing multidimensional heterogeneous data of the target reservoir, fused data is obtained. The target reservoir is a continental sandstone reservoir, and the comprehensive water cut of the target reservoir is higher than the preset water cut threshold and the reservoir permeability range is greater than the preset range threshold. Based on the fused data, a digital twin model corresponding to the target reservoir is constructed, wherein the digital twin model is a multiphysics coupling model driven by real-time data to maintain synchronous evolution with the physical reservoir. Under the condition that the triggering conditions are met, the digital twin model is used to quickly extrapolate the future short-term production dynamics of the target reservoir, obtain the extrapolation results, and generate risk warning information based on the extrapolation results; In response to the risk warning information, an optimization engine is used to execute an iterative optimization algorithm in the form of the digital twin model as a virtual environment to generate candidate control strategies. The candidate control strategy is input into the digital twin model for coupled simulation to obtain the reservoir dynamic prediction results output by the digital twin model; Based on the reservoir dynamic prediction results, the candidate control strategies are verified. If the candidate control strategy passes verification, the candidate control strategy is determined as the target control strategy and executed.

2. The method according to claim 1, characterized in that, The process involves fusing multidimensional heterogeneous data from the target reservoir to obtain fused data, including: Acquire multidimensional heterogeneous data of the target reservoir, wherein the multidimensional heterogeneous data includes: geological static data, engineering dynamic data, and scheme design data; The multidimensional heterogeneous data are fused in the same spatiotemporal coordinate system using a spatiotemporal alignment algorithm to obtain fused data.

3. The method according to claim 2, characterized in that, The construction of a digital twin model corresponding to the target reservoir based on the fused data includes: Based on the geological static data in the fused data, a three-dimensional geological grid model corresponding to the target reservoir is established; Based on the fused data, an initial attribute field is assigned to the grid cells in the three-dimensional geological grid model, wherein the initial attribute field includes at least one of the following: porosity field, permeability field, fluid saturation field, pressure field, and temperature field. In the three-dimensional geological grid model, governing equations for describing the physicochemical processes in the target reservoir are coupled to form a first model, wherein the governing equations include equations describing seepage mechanics, geomechanics, thermodynamics and chemical reaction processes; Based on the fused data, the first model is assigned initial rock physical properties and fluid state to obtain the second model; The second model is configured to evolve synchronously with the physical reservoir through real-time data-driven processes to obtain a digital twin model.

4. The method according to claim 1, characterized in that, The construction of a digital twin model corresponding to the target reservoir based on the fused data includes: Obtain the multiphysics coupled basic model corresponding to the target reservoir; The multiphysics coupling basic model is fine-tuned using the fused data of the target reservoir to obtain a digital twin model corresponding to the target reservoir.

5. The method according to claim 1, characterized in that, The step of employing an optimization engine, using the digital twin model as a virtual environment, to execute an iterative optimization algorithm to generate candidate control strategies includes: An optimization engine is used in the digital twin model to generate at least one tentative operation command based on the fused data; For each of the aforementioned exploratory operation commands, the digital twin model is invoked for rapid simulation to predict the reservoir state changes caused by the exploratory operation command and to calculate the reward value corresponding to the exploratory operation command. Based on the exploratory operation instructions, reward values, and changes in reservoir state, the decision model within the optimization engine is updated; Repeat the above steps until the preset exploration termination condition is met; Based on the updated decision model, an operation sequence starting from the current state and covering a preset future time period is obtained, and the operation sequence is encapsulated as a candidate control strategy.

6. The method according to claim 3, characterized in that, The digital twin model is coupled and simulated in real time through the following methods: For the conventional physical field changes of the target reservoir, a fast solver based on physical information neural network is used for near real-time simulation; To address the complex physicochemical processes of the target reservoir, a high-precision numerical simulator is periodically triggered to calibrate the fast solver.

7. The method according to claim 1, characterized in that, The verification of the candidate control strategy based on the dynamic prediction results includes: Based on the dynamic prediction results, it is determined whether the key parameters of the target reservoir exceed the safety threshold corresponding to the key parameters after the candidate control strategy is implemented. Based on the dynamic prediction results, it is determined whether the comprehensive score of the target reservoir in terms of recovery rate, economic cost and environmental indicators exceeds the scoring threshold after the implementation of the candidate control strategy.

8. The method according to claim 1, characterized in that, Before employing an optimization engine, using the digital twin model as a virtual environment, and executing an iterative optimization algorithm to generate candidate control strategies, the method further includes: Acquire health status prediction information and material supply chain status information of key equipment associated with the production of the target oil reservoir within the future preset time period; The equipment health status prediction information and the material supply chain status information are transformed into constraints or part of the objective function of the optimization engine.

9. A decision-making device for tertiary oil recovery (EOR) schemes, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a decision-making method for a tertiary oil recovery (EOR) scheme as described in any one of claims 1-8.

10. A computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, they implement a decision-making method for a tertiary oil recovery (EOR) scheme as described in any one of claims 1-8.

Images