Multi-objective optimization method for balancing CO2 storage amount and geological risk

Abstract

The invention relates to the technical field of CO2 capture and storage, in particular to a multi-objective optimization method for balancing CO2 storage amount and geological risk, a multi-feature fusion neural network is used in objective function calculation of comprehensive optimization of CO2 geological storage efficiency and safety, a new method is introduced, and the method aims to maximize CO2 storage volume and improve CO2 storage efficiency. According to the method, the CO2 injection rate and the geological risk can be balanced, the geological risk related to pressure accumulation is reduced, the efficiency and safety of CO2 storage operation are ensured through the comprehensive optimization method, the agent model and the NSGA-II algorithm are combined, the effective method is shown, the CO2 injection rate and the geological risk can be balanced, and the CO2 storage efficiency is improved. According to the method, well control scheduling and the opening time of the production well are optimized under geological uncertainty, the CO2 injection volume is maximized, and meanwhile, uniform pressure distribution is ensured, so that the technical problem that in an optimization method in the prior art, when a complex multi-objective optimization task is faced, an agent model is limited in the aspects of flexibility and calculation efficiency is solved.

Description

Technical Field

[0001] The present invention relates to CO 2 The field of capture and storage technology, particularly a method for balancing CO 2 A multi-objective optimization approach for storage capacity and geological risk. Background Art

[0002] Global warming caused by massive global CO2 emissions is one of the most significant challenges of this century. Promoting carbon capture and storage (CCS) on a global scale to reduce CO2 emissions is an important goal in the fields of climate change, sustainable development and energy. Among various storage options, saline aquifers show great geological storage potential. In order to apply carbon storage technology on a large scale in saline aquifers and depleted oil and gas reservoirs, CO 2 The injection method must be technically feasible, safe, economically sound, and optimized for storage efficiency. 2 Storage technology faces three main challenges: assessing the storage capacity of saline aquifers to maximize CO injection 2 , optimizing injection rates and well locations under geological uncertainty, and evaluating CO 2 Geological risks during injection, CO 2 The technical and geological challenges of injecting into these saline aquifers require advanced optimization techniques.

[0003] In optimizing CO 2 It is crucial to consider geological uncertainties when evaluating storage efficiency, as these uncertainties can affect CO 2 storage due to the heterogeneous distribution of geological parameters such as porosity and permeability. 2 Injection and leakage risks.

[0004] In the pursuit of effective CO2 storage, it is crucial to achieve a balance between CO2 injection rate and associated geological risks. During CO2 injection, excessive pressure accumulation will bring geological risks. Therefore, studying the balance between CO2 injection rate and associated geological risks, especially exploring the timing of well placement, well control and production well activation, while considering geological uncertainties, is crucial to maximizing CO2 storage efficiency and minimizing geological risks. In addition, during the optimization process, the number of well placement and well control sequences depends on the changes in the timing of production well activation, which increases the complexity of the solution space and makes the search for the best solution more complicated. This complex multi-objective optimization problem requires a large number of forward simulations, which are computationally expensive.

[0005] With the rapid development of artificial intelligence, machine learning and deep learning have promoted the application of high-precision proxy models in the fields of CCS and environmental science. Some multi-feature neural network surrogate models can take complex geological parameters and well control sequences as inputs and generate outputs such as production, pressure distribution, and saturation distribution, providing strong support for multi-objective optimization.

[0006] However, in existing optimization methods, when faced with complex multi-objective optimization tasks, the proxy model is limited in flexibility and computational efficiency. Summary of the invention

[0007] The object of the present invention is to provide a balanced CO 2 The multi-objective optimization method of storage volume and geological risk aims to solve the technical problem that the proxy model is limited in flexibility and computational efficiency when facing complex multi-objective optimization tasks in the optimization methods in the existing technology.

[0008] To achieve the above purpose, the present invention adopts a balanced CO 2 The multi-objective optimization method of storage volume and geological risk includes the following steps:

[0009] Process multi-feature input and output based on multi-feature fusion neural network;

[0010] Prediction of cumulative CO based on multi-feature fusion neural network 2 injection rate, reservoir pressure and saturation;

[0011] A non-dominated sorting genetic algorithm II is combined with a multi-feature fusion neural network to form a multi-objective optimization framework to optimize well control of injection and production, as well as the timing of production well opening under geological uncertainties.

[0012] The multi-objective optimization method for balancing CO2 storage and geological risks also includes:

[0013] Mask optimization is used to flexibly adjust the number of optimization variables during the optimization process to handle changes in the activation time of production wells.

[0014] Among them, the multi-feature fusion neural network adopts the combination of U-FNO and Transformer Encoders.

[0015] Among them, in the multi-objective optimization framework:

[0016] Two optimization objective functions are defined. The first objective function focuses on maximizing the CO that can be safely stored in the saline aquifer reservoir. 2 volume, and the second objective function focuses on optimizing the geological risk associated with the pressure distribution.

[0017] Among them, in the multi-objective optimization framework:

[0018] The crowding distance is introduced to measure the distance between an individual and its surrounding neighbors in the target space. After sorting and crowding distance calculation, the parent individuals are selected from the initial population.

[0019] Among them, in the multi-objective optimization framework:

[0020] Multi-objective optimization aims to balance the conflicts among different objectives and find the best set of non-dominated solutions. This Pareto optimal solution set forms a set usually called Pareto frontier or Pareto curve;

[0021] The Pareto optimal solution set is defined as:

[0022]

[0023] The non-dominated sorting genetic algorithm II is used to generate offspring through a series of steps involving selection, crossover, and mutation until the convergence criterion is met. During the iteration of the algorithm, individuals are sorted based on dominance relations and crowding distances to determine the Pareto front.

[0024] A balanced CO 2 A multi-objective optimization method for storage capacity and geological risk uses a multi-feature fusion neural network to take into account CO 2 The present invention successfully develops and applies a new multi-objective optimization framework for CO2 storage in saline aquifers. 2 Archive.

[0025] This invention introduces a new approach to optimize well opening timing: This invention introduces a new approach to maximize not only CO 2 This comprehensive optimization approach ensures that CO 2 Efficiency and safety of containment operations.

[0026] This paper combines the agent model with the NSGA-II algorithm: it demonstrates an effective method to balance CO 2 This method optimizes well control scheduling and production well opening timing under geological uncertainty, maximizing CO 2 Inject volume while ensuring even pressure distribution.

[0027] In this way, the technical problem of the limited flexibility and computational efficiency of the proxy model in the optimization method in the prior art when facing complex multi-objective optimization tasks is solved.

[0028] This paper constructs an innovative alternative model in which CO is enhanced by effectively modeling, reducing data requirements and improving accuracy.2 Prediction of storage state distribution, taking into account geological uncertainties;

[0029] A deep learning-based agent model was constructed specifically for CO 2 Confinement problem, approximating the relationship between permeability realization, well control sequence and dynamic response characteristics (such as pressure, saturation and production sequence); using this approach, the reservoir simulation process is modeled as:

[0030] X=[P,S,q i ,q p ]=g(m,w i ,w p );

[0031] The surrogate model takes as input a realization of geological uncertainty and a set of well control sequences that vary over time. Outputs include subsurface flow state variables (such as pressure and saturation) and a production sequence that varies over time;

[0032] After acquiring the pressure and saturation state images, convolutional layers are used for feature extraction to convert the state images into low-dimensional latent variables. These latent variables are then processed through a surrogate model to reconstruct the production rate series;

[0033] Use 2D convolution to build surrogate models for each layer of the 3D reservoir model, thereby reducing computational cost while maintaining acceptable accuracy. The surrogate model can be trained at each layer through parallel computing with graphics processing unit (GPU) acceleration;

[0034] After data preprocessing, the surrogate model is ready for training. In the present invention, the PyTorch deep learning framework is used to implement the training and evaluation process. The adaptive moment estimation (ADAM) algorithm is used for optimization. The training scheme is carefully designed to balance computational efficiency and prediction accuracy. The dataset is divided into training, validation, and test sets in a ratio of 8:1:1;

[0035] After the training process, various evaluation criteria are used to test the generalization performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 It is a well configuration diagram of the 3D actual saline water layer reservoir model of the present invention.

[0038] Figure 2 is the actual CO of the present invention 2 Transition diagram of injection strategy under well control constraints in storage scenario.

[0039] Figure 3 The CO after breakthrough of the production wells with different opening times according to the present invention 2 Productivity graph.

[0040] Figure 4 It is a diagram showing the solution distribution and Pareto front evolution in the iterative process of the present invention.

[0041] Figure 5 It is a schematic diagram of the Pareto preamble verification using 5,000 random experiments in the present invention.

[0042] Figure 6 is the error plot between the surrogate model of the Pareto front of the present invention and the full physics-based simulation.

[0043] Figure 7 The present invention maximizes carbon dioxide storage (F 1 )’s well control sequence diagram.

[0044] Figure 8 The present invention makes the Pareto front pressure uniform (F 2 ) maximized well control sequence diagram.

[0045] Fig. 9 is the balance CO of the present invention 2 Schematic flow diagram of the multi-objective optimization approach for storage capacity and geological risk. DETAILED DESCRIPTION

[0046] Embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0047] See also Figures 1 to 9 ,in Figure 1 is a well configuration diagram of the 3D actual saline water reservoir model of the present invention, Figure 2 is the actual CO of the present invention 2 Transition diagram of injection strategy under well control constraints in storage scenario. Figure 3 The CO after breakthrough of the production wells with different opening times according to the present invention 2 Productivity graph, Figure 4 is the solution distribution and Pareto front evolution diagram in the iterative process of the present invention, Figure 5 This is a schematic diagram of the present invention using 5,000 random experiments to verify the Pareto predicate. Figure 6is the error plot between the surrogate model of the Pareto front of the present invention and the full physics-based simulation, Figure 7 is a well control sequence diagram of the present invention for maximizing carbon dioxide storage (F1) on the Pareto front, Figure 8 is the well control sequence diagram of the present invention that maximizes the Pareto front pressure uniformity (F2), Fig. 9 is the balance CO of the present invention 2 Schematic flow diagram of the multi-objective optimization approach for storage capacity and geological risk.

[0048] The present invention provides a balanced CO 2 The model used in this invention represents a three-dimensional (3D) actual saline aquifer reservoir for CO 2 The optimization and analysis of storage provides a real-world example. The reservoir domain is discretized into a corner grid with 101×135×6 grid cells along the x, y, and z axes, respectively.

[0049] The top two layers of the model are cap layers, which provide CO 2 closure of the storage process, while the lower four layers are designated as saline reservoirs.

[0050] The reservoir model contains a total of 81,810 grid cells and has four injection wells and one production well, each of which vertically penetrates four layers below. Figure 1 shown.

[0051] In order to mitigate the continuous CO 2 Due to the risk of excessive pressure accumulation during injection, the control strategy of the injection well is switched from injection rate-based control to maximum injection pressure control to prevent the pressure from exceeding the fracture threshold of the formation. Figure 2 Shows a real CO 2 Injection scenario, where the solid and dashed lines represent four injection wells injecting CO at a rate of 1 million cubic meters and 500,000 cubic meters per day, respectively. 2 The figure shows that as CO 2 As the injection proceeds, the bottom hole pressure gradually approaches the formation fracture pressure, and the injection strategy changes to a mode constrained by the maximum pressure.

[0052] In addition, production wells are strategically timed to mitigate pressure buildup and thereby increase CO 2 The timing of starting a production well is critical. Starting it too early may lead to CO 2 A premature breakthrough and a delayed start may lead to localized concentration of pressure. Figure 3 It shows the CO 2 Production rate after breakthrough.

[0053] Data preparation: SGeMS software was used to randomly generate 10 permeability samples to reflect the uncertainty of the geological model.

[0054] 500 sets of well control sequences were randomly generated, where the control range of injection wells was between 200,000 and 1,000,000 m3 / day, the opening time of production wells was between 0 and 16 control time steps, and the well control range was between 40 and 45 MPa.

[0055] 500 simulations were performed for each model implementation, using the 500 sets of well control sequences generated, resulting in a total of 5,000 sample sets as training datasets.

[0056] The following steps are involved:

[0057] S100, processing multi-feature input and output based on multi-feature fusion neural network;

[0058] For this specific implementation, the multi-feature fusion neural network adopts a combination of U-FNO and TransformerEncoders.

[0059] S200, Prediction of cumulative CO based on multi-feature fusion neural network 2 injection rate, reservoir pressure and saturation;

[0060] S300, combines the non-dominated sorting genetic algorithm II (NSGA-II) with a multi-feature fusion neural network to form a multi-objective optimization framework to optimize well control of injection and production, as well as the timing of production well opening under geological uncertainty.

[0061] For this specific implementation, in the multi-objective optimization framework, the following optimization steps are included:

[0062] Define the multi-objective optimization function: the first objective function F 1 (·)(Equation 1) focuses on maximizing the CO that can be safely stored in saline aquifer reservoirs 2 Volume, the second objective function F 2 (·)(Formula 2) focuses on optimizing the geological risks associated with pressure distribution.

[0063]

[0064] Among them, P SD is the standard deviation of pressure at each time step. P i,j,k (u) represents the pressure at each grid point at each time step, represents the average pressure at each time, a, b, c represent the number of grids in the x, y, z directions respectively, m grid It represents the total number of corner grids of the reservoir model.

[0065]

[0066] Where u represents N u The optimization vector of all injection and production wells in the dimensional optimization space, N c Indicates the total number of control time steps. represents the CO of the jth well at the kth time step 2 Injection rate. represents the well control of the mth production well at the lth time step, u t It indicates the start-up time of the production well.

[0067] Multi-objective optimization: The trained surrogate model is combined with the NSGA-II algorithm to perform multi-objective optimization to determine the Pareto optimal solution set.

[0068] Define the concepts of Pareto frontier and non-dominated, where the Pareto optimal solution set is defined as:

[0069]

[0070] Among them, P * represents the Pareto optimal solution set, Ω is the vector set containing all feasible solutions, and x' is the corresponding arbitrary feasible solution vector in Ω.

[0071] Result analysis: Analyze the optimization results and determine the maximum CO 2 Well control strategies for storage volume and pressure uniformity.

[0072] The accuracy and reliability of the proxy model are verified by comparing the results of the proxy model with those of the full physics simulation.

[0073] Using a balanced CO 2 A multi-objective optimization method for storage capacity and geological risk uses a multi-feature fusion neural network to take into account CO 2 The present invention successfully develops and applies a new multi-objective optimization framework for CO2 storage in saline aquifers. 2 Archive.

[0074] This invention introduces a new approach to optimize well opening timing: This invention introduces a new approach to maximize not only CO 2 This comprehensive optimization approach ensures that CO 2 Efficiency and safety of containment operations.

[0075] This paper combines the agent model with the NSGA-II algorithm: it demonstrates an effective method to balance CO 2Injection rate and geological risk. This method optimizes well control and production well opening timing under geological uncertainty, maximizing CO 2 Fill volume while ensuring even pressure distribution.

[0076] In this way, the technical problem of the limited flexibility and computational efficiency of the proxy model in the optimization method in the prior art when facing complex multi-objective optimization tasks is solved.

[0077] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A multi-objective optimization method to balance CO2 storage and geological risks. It is characterized in that The following steps are involved: Process multi-feature input and output based on multi-feature fusion neural network; Predict cumulative CO2 injection, reservoir pressure and saturation based on multi-feature fusion neural network; A non-dominated sorting genetic algorithm II is combined with a multi-feature fusion neural network to form a multi-objective optimization framework to optimize well control of injection and production, as well as the timing of production well opening under geological uncertainties.

2. The multi-objective optimization method for balancing CO2 storage and geological risks according to claim 1, characterized in that: The multi-objective optimization method for balancing CO2 storage and geological risks also includes: Used to mask the number of optimization variables during optimization to handle variations in production well activation times.

3. The multi-objective optimization method for balancing CO2 storage and geological risks according to claim 2, characterized in that: The multi-feature fusion neural network combines U-FNO with Transformer Encoders.

4. The multi-objective optimization method for balancing CO2 storage and geological risk according to claim 3, characterized in that: In the multi-objective optimization framework: Two optimization objectives are defined, the first one focusing on maximizing the volume of CO2 that can be safely stored in the saline aquifer reservoir, and the second one focusing on optimizing the geological risks associated with the pressure distribution.

5. The multi-objective optimization method for balancing CO2 storage and geological risk according to claim 4, characterized in that: In the multi-objective optimization framework: The crowding distance is introduced to measure the distance between an individual and its surrounding neighbors in the target space. After sorting and crowding distance calculation, the parent individuals are selected from the initial population.

6. The multi-objective optimization method for balancing CO2 storage and geological risks according to claim 5, characterized in that: In the multi-objective optimization framework: Multi-objective optimization aims to balance the conflicts among different objectives and find the best set of non-dominated solutions. This Pareto optimal solution set forms a set usually called Pareto frontier or Pareto curve; The non-dominated sorting genetic algorithm II is used to generate offspring through a series of steps involving selection, crossover, and mutation until the convergence criterion is met. During the iteration of the algorithm, individuals are sorted based on dominance relations and crowding distances to determine the Pareto front.

Images