Saline water layer non-pure CO2 plume inversion method based on artificial intelligence expert system
By combining an artificial intelligence expert system with a global optimization algorithm and using bottomhole pressure data to invert the distribution of impure CO2 plumes, the problems of high cost and low resolution in CO2 geological storage safety monitoring were solved, achieving low-cost and efficient monitoring results.
Patent Information
- Application Number
- CN202410434602.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-11
- Publication Date
- 2025-10-21
AI Technical Summary
The existing CO2 geological storage safety monitoring equipment is expensive and has low signal resolution, making it difficult to widely use for long-term monitoring. Traditional history matching methods have high computational overhead, making it difficult to describe the spatial evolution of plumes after the injection of impure CO2 into saline aquifers.
An artificial intelligence expert system-based inversion method for impure CO2 plumes in saline aquifers was adopted. The bottomhole pressure data was predicted by the first artificial intelligence expert system, and the optimal reservoir description parameters were solved by combining the global optimization algorithm. The underground distribution of the impure CO2 plume was predicted using the second artificial intelligence expert system.
It reduces monitoring costs, improves the accuracy of inversion results, is suitable for large-scale promotion and application, and achieves efficient identification of monitoring signals and assessment of cover safety.
Smart Images
Figure CN120823889A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of carbon dioxide geological storage, and in particular to an inversion method for impure CO2 plumes in saline layers based on an artificial intelligence expert system. Background Art
[0002] Studying the long-term migration and distribution of CO2 in saline aquifers is crucial for ensuring the safety of geological storage. The plume formed by CO2 injection into saline aquifers evolves over time and can potentially leak into shallower layers through cracks, faults, and abandoned, poorly sealed wellbores. Long-term physical and chemical interactions with the caprock can exacerbate CO2 leakage and potentially compromise its integrity, leading to fault activation, caprock failure, and crack expansion.
[0003] By installing different sensing devices and collecting data such as pressure, strain, microseismic signals, and carbon isotope concentration, the safety of CO2 geological storage can be monitored. Park et al. have established a technology for evaluating the risk of carbon dioxide leakage by monitoring the bottom hole flow pressure (see "A pressure-monitoring method to warn CO2 leakage in geological storage sites", Park YC et al., Environmental Earth Sciences, 2012, 67 (2): 425-433). Xue et al. have proposed a technology for evaluating the integrity of the caprock by measuring the strain of the caprock by installing an optical fiber cable at the bottom of the well (see "Geomechanical Monitoring of Caprock and Wellbore Integrity Using Fiber Optic Cable:Strain Measurement from the Fluid Injection and Extraction Field Tests", Xue Z et al., Energy Procedia, 2017, 114: 3305-3311). Mitchell and Green proposed a method for monitoring caprock tightness by collecting microseismic signals induced by CO2 fluid migrating through caprock to shallower layers (see “Some induced seismicity considerations in geo-energy resource development,” Mitchell JK et al., Geomechanics for Energy and the Environment, 2017, 10:3-11). Barth et al. developed a CO2 leakage risk monitoring method based on isotope tracing and applied it in the Ketzin pilot project in Germany (see “Monitoring of caprock integrity during CCS from field data at the Ketzin pilot site (Germany): Evidence from gas composition and stable carbon isotopes,” Barth JAC et al., International Journal of Greenhouse Gas Control, 2015, 43:133-140).In addition, CO2 leakage can also be remotely monitored by installing surface equipment near the injection site (see "CO2 capture and storage monitoring based on remote sensing techniques: Areview", Zhang T et al., Journal of Cleaner Production, 2021, 281: 124409). The above monitoring methods generally have the problems of expensive equipment investment and labor costs, and low signal resolution, and cannot be widely used in the long-term monitoring of CO2 storage safety. In addition to the above technical methods, Bruno et al. proposed a comprehensive evaluation method for CO2 storage safety that integrates formation description, numerical simulation, data analysis and other means, and applied it in multiple carbon storage regional projects funded by the U.S. Department of Energy (see "Development of Improved caprock Integrity Analysis and Risk Assessment Techniques", Bruno MS et al., Energy Procedia, 2014, 63: 4708-4744).
[0004] With the increasing application of artificial intelligence technology in related fields such as seepage mechanics and rock mechanics, initial progress has been made in the research of AI-based CO2 storage safety monitoring methods. Chen et al. proposed a machine learning-assisted CO2 geological storage monitoring method, characterizing the uncertainty of CO2 leakage by integrating expert systems and statistical analysis methods (see “Geologic CO2 sequestration monitoring design: A machine learning and uncertainty quantification-based approach,” Chen B et al., Applied Energy, 2018, 225:332-345). Zhou et al. developed a method for detecting CO2 leakage by identifying microseismic signals using convolutional neural networks (CNNs). Leveraging the advantages of CNNs in pattern recognition accuracy and data processing speed, they efficiently extracted microseismic signals that represent CO2 leakage (see “A data-driven CO2 leakage detection using seismic data and spatial–temporal densely connected convolutional neural networks,” Zhou Z et al., International Journal of Greenhouse Gas Control, 2019, 90:102790).
[0005] According to the current research status at home and abroad, the equipment for CO2 geological storage safety monitoring has become mature, but the monitoring signal data obtained is large in volume, noisy, and has low resolution.
[0006] Therefore, establishing an artificial intelligence method that can efficiently identify monitoring signals and accurately assess cap rock safety and CO2 leakage risks is of great significance to ensuring the safe geological storage of CO2. Summary of the Invention
[0007] In view of the problems existing in the prior art, the present invention provides an inversion method for impure CO2 plumes in saline aquifers based on an artificial intelligence expert system. By coupling a first artificial intelligence expert system with a global optimization algorithm, the optimal reservoir description parameters are solved by minimizing the history fitting error. Then, a second artificial intelligence expert system is used to predict the underground distribution of the impure CO2 plume to obtain the impure CO2 plume inversion result. This solves the problem in the prior art of high inversion cost and low result resolution of relying on geological exploration methods, and high computational overhead of traditional history fitting methods, which leads to difficulty in describing the spatiotemporal evolution law of the plume after impure CO2 is injected into the saline aquifer.
[0008] To achieve this object, the present invention adopts the following technical solutions:
[0009] The present invention provides a method for inverting impure CO2 plumes in saline water layers based on an artificial intelligence expert system. The method comprises the following steps:
[0010] (1) Using the actual pressure data collected in the monitoring wells as a reference, the spatial heterogeneity distribution of the reservoir permeability of the saline layer is initialized, and the bottom hole pressure data of the monitoring wells is predicted using the first artificial intelligence expert system;
[0011] (2) inputting the bottom hole pressure data of the monitoring well into the coupling system of the first artificial intelligence expert system and the global optimization algorithm, and solving the optimal reservoir description parameters by minimizing the historical fitting error;
[0012] (3) Using the optimal reservoir description parameters as input data, a second artificial intelligence expert system is used to predict the underground distribution of impure CO2 plumes, thereby obtaining impure CO2 plume inversion results.
[0013] The artificial intelligence expert system-based inversion method for impure CO2 plumes in saline aquifers described in the present invention first initializes the spatially heterogeneous distribution of the reservoir permeability of the saline aquifer using actual pressure data collected in monitoring wells as a reference, and uses a first artificial intelligence expert system to predict the bottomhole pressure data of the monitoring wells. The bottomhole pressure data of the monitoring wells is then input into a coupled system of the first artificial intelligence expert system and a global optimization algorithm, where the optimal reservoir description parameters are solved by minimizing the historical fitting error. Finally, the optimal reservoir description parameters are used as input data, and a second artificial intelligence expert system is used to predict the underground distribution of the impure CO2 plume, obtaining the impure CO2 plume inversion results. The inversion method for impure CO2 plumes in saline aquifers described in the present invention is rationally designed, has low inversion costs, and has high accuracy. Compared with traditional geological exploration methods such as seismic, well logging, and electromagnetic waves, it can significantly reduce project monitoring costs and is suitable for large-scale promotion and application.
[0014] Preferably, in step (1), the first artificial intelligence expert system is a deep neural network model.
[0015] Preferably, in step (1), the input parameter of the first artificial intelligence expert system is the spatially heterogeneous distribution of reservoir permeability, and the prediction process of the deep neural network model is: the first input layer is transmitted to the first convolutional layer-1, to the first ReLU layer-1 to the first maximum pooling layer-1 to the first convolutional layer-2 to the first ReLU layer-2 to the first maximum pooling layer-2 to the first Dropout layer to the first convolutional layer-3; the first convolutional layer-3 is connected to the first fully connected layer.
[0016] The size N of the first fully connected layer of the present invention is determined by the amount of data available in the mine.
[0017] Preferably, the global optimization algorithm in step (2) takes the heterogeneous spatial distribution of formation permeability as a variable, stores it in the form of a two-dimensional matrix, and uses the history fitting error HME as the objective function:
[0018]
[0019] In formula (1), n t is the number of bottom hole pressure observation points, P is the predicted value of bottom hole pressure by artificial intelligence expert system A, and O is the actual pressure value monitored in the mine.
[0020] Preferably, the steps of history matching in step (2) are as follows:
[0021] ① Compare the bottomhole pressure of the monitoring well predicted by the first artificial intelligence expert system with the field monitoring data, and calculate HME using formula (1); if HME < ε, the spatial heterogeneous distribution of reservoir permeability is output, and history fitting is completed;
[0022] ② If HME ≥ ε, the global optimizer is used to adjust the permeability parameters;
[0023] ③ According to the reasonable range of reservoir permeability, randomly generate n groups of permeability samples, each group of samples contains m permeability sampling values;
[0024] ④ For n groups of samples and m permeability values, initialize the velocity term v, and all initialized v values are 0;
[0025] ⑤ Using the first artificial intelligence expert system, calculate the HME of n groups of permeability samples, and obtain n groups of HMEs. Among these n groups of samples, the permeability corresponding to the lowest HME is g. During the history fitting process, the permeability corresponding to the lowest HME value that each sample can achieve is p. g and p are two groups of permeability heterogeneous distributions with m values each.
[0026] ⑥ Update the velocity term of history fitting cycle k+1 using formula (2), where k=0 corresponds to the initial value;
[0027] v i (k+1)=wv i (k)+c1[p(k)-x i (k)]+c2[g(k)-x i (k)] Formula (2)
[0028] x i (k) represents the heterogeneous spatial distribution of permeability corresponding to the i-th sample in the k-th fitting cycle, containing m permeability values; vi (k) represents the speed of the i-th sample in the k-th fitting cycle, w, c1, and c2 are all random numbers between 0 and 1;
[0029] ⑦ Update the permeability distribution using formula (3);
[0030] x i (k+1)=x i (k)+v i (k+1) Formula (3)
[0031] ⑧ Use formula (1) to calculate the HME corresponding to g. If HME < ε, stop history fitting.
[0032] And output g; otherwise return to step ⑤.
[0033] Preferably, the optimal reservoir description parameter in step (2) is the heterogeneous spatial distribution of formation permeability obtained when HME is less than the error tolerance ε.
[0034] In the present invention, the error tolerance ε can be set according to actual conditions, for example, it can be set to 5%.
[0035] Preferably, in step (3), the second artificial intelligence expert system is a deep neural network model.
[0036] Preferably, in step (3), the signal transmission process of the second artificial intelligence expert system is: the second input layer is transmitted to the second convolutional layer-1 to the second ReLU layer-1 to the second maximum pooling layer-1 to the second convolutional layer-2 to the second ReLU layer-2 to the second maximum pooling layer-2 to the second Dropout layer to the second convolutional layer-3; the second convolutional layer-3 is connected to the second fully connected layer.
[0037] The size N×M of the second fully connected layer of the present invention is determined by the temporal-spatial resolution of the impure CO2 plume fluid profile.
[0038] Preferably, the inversion results of the impure CO2 plume in step (3) include gas saturation and a profile of typical mineral chemical reaction products.
[0039] Preferably, the typical minerals include calcite, anorthite and kaolinite.
[0040] As a preferred technical solution of the present invention, the inversion method for impure CO2 plumes in saline layers includes the following steps:
[0041] (1) Using the actual pressure data collected in the monitoring wells as a reference, the spatial heterogeneity distribution of the reservoir permeability of the saline layer is initialized, and the bottom hole pressure data of the monitoring wells is predicted using the first artificial intelligence expert system;
[0042] The first artificial intelligence expert system is a deep neural network model; the input parameter of the first artificial intelligence expert system is the spatially heterogeneous distribution of reservoir permeability, and the prediction process of the deep neural network model is: the first input layer is transmitted to the first convolution layer-1, to the first ReLU layer-1, to the first maximum pooling layer-1, to the first convolution layer-2, to the first ReLU layer-2, to the first maximum pooling layer-2, to the first Dropout layer, to the first convolution layer-3; the first convolution layer-3 is connected to the first fully connected layer;
[0043] (2) inputting the bottom hole pressure data of the monitoring well into the coupling system of the first artificial intelligence expert system and the global optimization algorithm, and solving the optimal reservoir description parameters by minimizing the historical fitting error;
[0044] The global optimization algorithm uses the heterogeneous spatial distribution of formation permeability as a variable, stores it in the form of a two-dimensional matrix, and uses the history fitting error HME as the objective function:
[0045]
[0046] In formula (1), n t is the number of bottom hole pressure observation points, P is the predicted value of bottom hole pressure by artificial intelligence expert system A, and O is the actual pressure value monitored in the mine;
[0047] The steps of history matching are as follows:
[0048] ① Compare the bottomhole pressure of the monitoring well predicted by the first artificial intelligence expert system with the field monitoring data, and calculate HME using formula (1); if HME < ε, the spatial heterogeneous distribution of reservoir permeability is output, and history fitting is completed;
[0049] ② If HME ≥ ε, the global optimizer is used to adjust the permeability parameters;
[0050] ③ According to the reasonable range of reservoir permeability, randomly generate n groups of permeability samples, each group of samples contains m permeability sampling values;
[0051] ④ For n groups of samples and m permeability values, initialize the velocity term v, and all initialized v values are 0;
[0052] ⑤ Using the first artificial intelligence expert system, calculate the HME of n groups of permeability samples, and obtain n groups of HMEs. Among these n groups of samples, the permeability corresponding to the lowest HME is g. During the history fitting process, the permeability corresponding to the lowest HME value that each sample can achieve is p. g and p are two groups of permeability heterogeneous distributions with m values each.
[0053] ⑥ Update the velocity term of history fitting cycle k+1 using formula (2), where k=0 corresponds to the initial value;
[0054] v i (k+1)=wv i (k)+c1[p(k)-x i (k)]+c2[g(k)-x i (k)] Formula (2)
[0055] x i (k) represents the heterogeneous spatial distribution of permeability corresponding to the i-th sample in the k-th fitting cycle, containing m permeability values; v i (k) represents the speed of the i-th sample in the k-th fitting cycle, w, c1, and c2 are all random numbers between 0 and 1;
[0056] ⑦ Update the permeability distribution using formula (3);
[0057] x i (k+1)=x i (k)+v i (k+1) Formula (3)
[0058] ⑧ Use formula (1) to calculate the HME corresponding to g. If HME < ε, stop history fitting.
[0059] And output g; otherwise return to step ⑤.
[0060] The optimal reservoir description parameter is the heterogeneous spatial distribution of formation permeability obtained when HME is less than the error tolerance ε;
[0061] (3) using the optimal reservoir description parameters as input data, using a second artificial intelligence expert system to predict the underground distribution of impure CO2 plumes, and obtaining impure CO2 plume inversion results;
[0062] The second artificial intelligence expert system is a deep neural network model; the signal transmission process of the second artificial intelligence expert system is: the second input layer is transmitted to the second convolution layer-1 to the second ReLU layer-1 to the second maximum pooling layer-1 to the second convolution layer-2 to the second ReLU layer-2 to the second maximum pooling layer-2 to the second Dropout layer to the second convolution layer-3; the second convolution layer-3 is connected to the second fully connected layer;
[0063] The inversion results of the impure CO2 plume include gas saturation and a profile of typical mineral chemical reaction products; the typical minerals include calcite, anorthite and kaolinite.
[0064] Compared with the prior art, the present invention has at least the following beneficial effects:
[0065] The artificial intelligence expert system-based inversion method for detecting impure CO2 plumes in saline aquifers, provided by this invention, leverages the pattern recognition capabilities of artificial intelligence expert systems. By analyzing only bottomhole pressure data from monitoring wells, it can rapidly invert the spatial and temporal distribution of impure CO2 plumes, as well as the profiles of typical mineralization products, with high accuracy. Compared with traditional geological exploration methods such as seismic, well logging, and electromagnetic waves, this method offers lower monitoring costs and has the potential for widespread application. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] Figure 1 It is a flow chart of the method for inverting impure CO2 plumes in saline layers based on an artificial intelligence expert system in a specific embodiment of the present invention.
[0067] Figure 2 It is a schematic diagram of reservoir-well positions in a specific embodiment of the present invention.
[0068] Figure 3 It is a neural network structure diagram of the first artificial intelligence expert system provided in a specific embodiment of the present invention.
[0069] Figure 4 1 is a graph of bottom hole pressure data of monitoring wells in the first 10 years in a specific embodiment of the present invention.
[0070] Figure 5 This is a bottom hole pressure data diagram after history matching in a specific embodiment of the present invention.
[0071] Figure 6 It is a neural network structure diagram of the second artificial intelligence expert system in a specific embodiment of the present invention.
[0072] Figure 7 This is a graph of the gas plume inversion results for the first 10 years in a specific embodiment of the present invention.
[0073] Figure 8 It is a cross-sectional view of the chemical reaction products of calcite in the first 10 years in a specific embodiment of the present invention.
[0074] Figure 9 This is a cross-sectional view of the chemical reaction products of anorthite in the first 10 years in a specific embodiment of the present invention.
[0075] Figure 10 It is a cross-sectional view of the chemical reaction products of kaolinite in the first 10 years in a specific embodiment of the present invention.
[0076] Figure 11 This is the inversion result diagram of the gas plume in the first 10 years in the reference group of the present invention.
[0077] Figure 12It is a cross-sectional diagram of the chemical reaction products of calcite in the first 10 years of the reference group of the present invention.
[0078] Figure 13 This is a cross-sectional view of the chemical reaction products of anorthite in the first 10 years of the reference group of the present invention.
[0079] Figure 14 This is a cross-sectional view of the chemical reaction products of kaolinite in the first 10 years of the reference group of the present invention.
[0080] Figure 15 This is a diagram of the gas plume inversion result after 100 years in a specific embodiment of the present invention.
[0081] Figure 16 It is a cross-sectional view of the chemical reaction product of calcite after 100 years in a specific embodiment of the present invention.
[0082] Figure 17 This is a cross-sectional view of the chemical reaction product of anorthite after 100 years in a specific embodiment of the present invention.
[0083] Figure 18 It is a cross-sectional view of the chemical reaction product of kaolinite after 100 years in a specific embodiment of the present invention.
[0084] Figure 19 This is the inversion result of the gas plume in the reference group of the present invention after 100 years.
[0085] Figure 20 It is a cross-sectional view of the chemical reaction products of calcite in the reference group of the present invention after 100 years.
[0086] Figure 21 This is a cross-sectional view of the chemical reaction products of anorthite in the reference group of the present invention after 100 years.
[0087] Figure 22 This is a cross-sectional view of the chemical reaction products of kaolinite in the reference group of the present invention after 100 years. DETAILED DESCRIPTION
[0088] The technical solution of the present invention will be further described below with reference to the accompanying drawings and through specific implementation methods.
[0089] The present invention is further described in detail below. However, the following examples are merely simplified examples of the present invention and do not represent or limit the scope of protection of the present invention. The scope of protection of the present invention shall be subject to the claims.
[0090] As a specific embodiment of the present invention, a method for inverting impure CO2 plumes in saline layers based on an artificial intelligence expert system is provided, and its flow chart is as follows: Figure 1 shown.
[0091] Acid waste gas, consisting of 50% carbon dioxide and 50% hydrogen sulfide, is injected into an ideal saline aquifer, 1000 meters long, 50 meters thick, and 100 meters wide, for wastewater treatment. The saline aquifer has a porosity of 25% and a permeability of 10 to 100 m / s. The project's injection well injects acid waste gas at one end of the saline aquifer, and a monitoring well is set up at x = 500 m to collect pressure data. The injection volume is 1000 m 3 / day, a total of 10 years of gas injection, and 100 years of monitoring. The reservoir-well position diagram in this specific embodiment is as follows Figure 2 shown.
[0092] The inversion method for impure CO2 plumes in saline layers comprises the following steps:
[0093] (1) Using the actual pressure data collected in the monitoring wells as a reference, the spatial heterogeneity distribution of the reservoir permeability of the saline layer is initialized, and the bottom hole pressure data of the monitoring wells is predicted using the first artificial intelligence expert system;
[0094] The first artificial intelligence expert system is a deep neural network model, and its neural network structure is as follows Figure 3 As shown in the figure, the prediction error of the first artificial intelligence expert system for bottom hole pressure is 0.28%;
[0095] The input parameter of the first artificial intelligence expert system is the spatial heterogeneous distribution of reservoir permeability. The heterogeneous spatial distribution of reservoir permeability is regularized into a two-dimensional matrix of 50×50. The prediction process of the deep neural network model is: the first input layer is transmitted to the first convolutional layer-1 (50×50×8), to the first ReLU layer-1 (50×50×8) to the first maximum pooling layer-1 (24×24×8) to the first convolutional layer-2 (24×24×16) to the first ReLU layer-2 (24×24×16) to the first maximum pooling layer-2 (12×12×16) to the first Dropout layer (12×12×16) to the first convolutional layer-3 (12×12×2), where the integer in the brackets represents the spatial size of the convolutional layer; the first convolutional layer-3 is connected to the first fully connected layer; the mine can provide pressure data for 120 time recording points, so N=120.
[0096] (2) inputting the bottom hole pressure data of the monitoring well into the coupling system of the first artificial intelligence expert system and the global optimization algorithm, and solving the optimal reservoir description parameters by minimizing the historical fitting error;
[0097] The global optimization algorithm uses the heterogeneous spatial distribution of formation permeability as a variable, stores it in the form of a two-dimensional matrix, and uses the history fitting error HME as the objective function:
[0098]
[0099] In formula (1), n t is the number of bottom hole pressure observation points, P is the predicted value of bottom hole pressure by artificial intelligence expert system A, and O is the actual pressure value monitored in the mine;
[0100] The steps of history matching are as follows:
[0101] ① Compare the bottomhole pressure of the monitoring well predicted by the first artificial intelligence expert system with the field monitoring data, and calculate HME using formula (1); if HME < ε, output the spatial heterogeneous distribution of reservoir permeability, and the history fitting is completed; in this specific implementation, the error tolerance ε is 5%;
[0102] ② If HME ≥ ε, the global optimizer is used to adjust the permeability parameters. The input parameter of the first artificial intelligence expert system is a 50×50 two-dimensional matrix, so 2500 parameters need to be adjusted.
[0103] ③ Based on the reasonable range of reservoir permeability, randomly generate n groups of permeability samples, each group of samples contains 2500 permeability sampling values;
[0104] ④ For n groups of samples, 2500 permeability values, initialize the velocity term v, v is a 200 × 2500 matrix, and all v values are initialized to 0;
[0105] ⑤ Using the first artificial intelligence expert system, calculate the HME of n groups of permeability samples, and you will get n groups of HMEs. Among these n groups of samples, the permeability corresponding to the lowest HME is g. During the history fitting process, the permeability corresponding to the lowest HME value that each sample can achieve is p. g and p are two groups of permeability heterogeneous distributions containing 2500 values each.
[0106] ⑥ Update the velocity term of history fitting cycle k+1 using formula (2), where k=0 corresponds to the initial value;
[0107] v i (k+1)=wv i (k)+c1[p(k)-x i (k)]+c2[g(k)-x i (k)] Formula (2)
[0108] x i (k) represents the heterogeneous spatial distribution of permeability corresponding to the i-th sample in the k-th fitting cycle, containing 50×50 permeability values; v i (k) represents the speed of the i-th sample in the k-th fitting cycle, w, c1, and c2 are all random numbers between 0 and 1;
[0109] ⑦ Update the permeability distribution using formula (3);
[0110] x i (k+1)=x i (k)+v i (k+1) Formula (3)
[0111] ⑧ Use formula (1) to calculate the HME corresponding to g. If HME < ε, stop history fitting.
[0112] And output g; otherwise return to step ⑤.
[0113] The optimal reservoir description parameter is the heterogeneous spatial distribution of formation permeability obtained when HME is less than the error tolerance ε;
[0114] The bottom hole pressure data of the monitoring wells in the first 10 years in this specific embodiment are shown in FIG. Figure 4 As shown in the figure, the bottom hole pressure data after history fitting is as follows Figure 5 As shown, the obtained history fitting error is 1.05%.
[0115] (3) using the optimal reservoir description parameters as input data, using a second artificial intelligence expert system to predict the underground distribution of impure CO2 plumes, and obtaining impure CO2 plume inversion results;
[0116] The second artificial intelligence expert system is a deep neural network model, and its neural network structure is as follows Figure 6 As shown in the figure, the prediction error of the second artificial intelligence expert system for the spatial distribution of gas plumes and mineralization reaction products is 4.17%;
[0117] The signal transmission process of the second artificial intelligence expert system is as follows: the second input layer is transmitted to the second convolution layer-1 (50×50×8) to the second ReLU layer-1 (50×50×8) to the second maximum pooling layer-1 (25×25×8) to the second convolution layer-2 (25×25×16) to the second ReLU layer-2 (25×25×16) to the second maximum pooling layer-2 (13×13×16) to the second Dropout layer (13×13×16) to the second convolution layer-3 (13×13×32); the second convolution layer-3 is connected to the second fully connected layer; the resolution of the plume flow profile is 50×50×2, where 50×50 is the spatial resolution and 2 represents inversion at two time nodes, so N=2500 and M=2;
[0118] The inversion results of the impure CO2 plume include gas saturation and a profile of typical mineral chemical reaction products; the typical minerals include calcite, anorthite and kaolinite.
[0119] The reference group is generated by running a numerical simulator based on the finite volume method to generate gas saturation and typical mineralization reaction product profiles. The specific steps to generate the reference group data are:
[0120] 1. Figure 2 The saline water layer shown in the figure is discretized into 50×1×50 grids along the length (x-), width (y-), and height (z-) directions, and the size of each grid is 20 m×100 m×1 m.
[0121] 2. The porosity of each grid is set to 25%, and the reservoir description parameters (here, the spatial distribution of permeability) obtained in step (2) of this specific embodiment are imported into the grid system.
[0122] 3. At x=0, press 1000m 3 A gas containing 50% carbon dioxide and 50% hydrogen sulfide was injected into the saline aquifer at a constant flow rate of 10 / day for 10 years, and a numerical simulator was run.
[0123] After 10 years, shut down the gas injection well and continue to run the numerical simulator for another 100 years. Derive the gas plume and typical mineralization reaction product profiles after 10 years of gas injection and 100 years of well shut-in.
[0124] The gas plume inversion results for the first 10 years in this specific implementation are as follows: Figure 7 As shown in the figure, the cross section of the chemical reaction products of calcite, anorthite and kaolinite is as follows Figures 8-10 The gas plume inversion results for the first 10 years in the reference group are shown in Figure 11 As shown in the figure, the cross section of the chemical reaction products of calcite, anorthite and kaolinite is as follows Figures 12-14 In this embodiment, the gas plume inversion result after 100 years is as shown in Figure 15 As shown in the figure, the cross section of the chemical reaction products of calcite, anorthite and kaolinite is as follows Figures 16-18 The gas plume inversion results after 100 years in the reference group are shown in Figure 19 As shown in the figure, the cross section of the chemical reaction products of calcite, anorthite and kaolinite is as follows Figures 20-22 shown.
[0125] Figures 7 to 22 The gas saturation and mineralization reaction product profiles of the saline aquifer were described 10 years after gas injection and 100 years after well closure. The artificial intelligence expert system-based inversion method for impure CO2 plumes in saline aquifers provided in this specific embodiment achieved a relative error of only 6.77% compared to the reference group. Furthermore, a single inversion analysis took only 22 seconds to calculate, without the need for geological exploration tools.
[0126] In summary, the artificial intelligence expert system-based saline aquifer impure CO2 plume inversion method provided by the present invention utilizes the pattern recognition capability of the artificial intelligence expert system and only requires analysis of the bottom hole pressure data of the monitoring well. It can quickly invert the temporal and spatial distribution of the impure CO2 plume and the profile of typical mineralization reaction products. The inversion has high accuracy, low computational overhead, and low implementation cost, and is suitable for large-scale promotion and application in the field of description of impure CO2 plumes in saline aquifers.
[0127] The applicant declares that the above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention fall within the scope of protection and disclosure of the present invention.
Claims
1. A method for inverting impure CO2 plumes in saline aquifers based on an artificial intelligence expert system, characterized in that: The inversion method for impure CO2 plumes in saline layers comprises the following steps: (1) Using the actual pressure data collected in the monitoring wells as a reference, the spatial heterogeneity distribution of the reservoir permeability of the saline layer is initialized, and the bottom hole pressure data of the monitoring wells is predicted using the first artificial intelligence expert system; (2) inputting the bottom hole pressure data of the monitoring well into the coupling system of the first artificial intelligence expert system and the global optimization algorithm, and solving the optimal reservoir description parameters by minimizing the historical fitting error; (3) Using the optimal reservoir description parameters as input data, a second artificial intelligence expert system is used to predict the underground distribution of impure CO2 plumes, thereby obtaining impure CO2 plume inversion results.
2. The method for inverting impure CO2 plumes in saline layers according to claim 1, characterized in that: Step (1) The first artificial intelligence expert system is a deep neural network model.
3. The method for inverting impure CO2 plumes in saline layers according to claim 1 or 2, characterized in that: Step (1) The input parameter of the first artificial intelligence expert system is the spatially heterogeneous distribution of reservoir permeability, and the prediction process of the deep neural network model is: the first input layer is transmitted to the first convolution layer-1, to the first ReLU layer-1 to the first maximum pooling layer-1 to the first convolution layer-2 to the first ReLU layer-2 to the first maximum pooling layer-2 to the first Dropout layer to the first convolution layer-3; the first convolution layer-3 is connected to the first fully connected layer.
4. The method for inverting impure CO2 plumes in saline layers according to any one of claims 1 to 3, characterized in that: The global optimization algorithm in step (2) takes the heterogeneous spatial distribution of formation permeability as a variable, stores it in the form of a two-dimensional matrix, and uses the historical fitting error HME as the objective function: In formula (1), n t is the number of bottom hole pressure observation points, P is the predicted value of bottom hole pressure by artificial intelligence expert system A, and O is the actual pressure value monitored in the mine.
5. The method for inverting impure CO2 plumes in saline layers according to any one of claims 1 to 4, characterized in that: The steps of history matching in step (2) are as follows: ① Compare the bottomhole pressure of the monitoring well predicted by the first artificial intelligence expert system with the field monitoring data, and calculate HME using formula (1); if HME < ε, the spatial heterogeneous distribution of reservoir permeability is output, and history fitting is completed; ② If HME ≥ ε, the global optimizer is used to adjust the permeability parameters; ③ According to the reasonable range of reservoir permeability, randomly generate n groups of permeability samples, each group of samples contains m permeability sampling values; ④ For n groups of samples and m permeability values, initialize the velocity term v, and all initialized v values are 0; ⑤ Using the first artificial intelligence expert system, calculate the HME of n groups of permeability samples, and obtain n groups of HMEs. Among these n groups of samples, the permeability corresponding to the lowest HME is g. During the history fitting process, the permeability corresponding to the lowest HME value that each sample can achieve is p. g and p are two groups of permeability heterogeneous distributions with m values each. ⑥ Update the velocity term of history fitting cycle k+1 using formula (2), where k=0 corresponds to the initial value; v i (k+1)=wv i (k)+c1[p(k)-x i (k)]+c2[g(k)-x i (k)] Formula (2) x i (k) represents the heterogeneous spatial distribution of permeability corresponding to the i-th sample in the k-th fitting cycle, containing m permeability values; v i (k) represents the speed of the i-th sample in the k-th fitting cycle, w, c1, and c2 are all random numbers between 0 and 1; ⑦ Update the permeability distribution using formula (3); x i (k+1)=x i (k)+v i (k+1) Formula (3) ⑧ Use formula (1) to calculate the HME corresponding to g. If HME < ε, stop history fitting and output g; otherwise return to step ⑤.
6. The method for inverting impure CO2 plumes in saline layers according to any one of claims 1 to 5, characterized in that: The optimal reservoir description parameter in step (2) is the heterogeneous spatial distribution of formation permeability obtained when HME is less than the error tolerance ε.
7. The method for inverting impure CO2 plumes in saline layers according to any one of claims 1 to 6, characterized in that: Step (3) The second artificial intelligence expert system is a deep neural network model.
8. The method for inverting impure CO2 plumes in saline layers according to any one of claims 1 to 7, characterized in that: Step (3) The signal transmission process of the second artificial intelligence expert system is: the second input layer is transmitted to the second convolution layer-1 to the second ReLU layer-1 to the second maximum pooling layer-1 to the second convolution layer-2 to the second ReLU layer-2 to the second maximum pooling layer-2 to the second Dropout layer to the second convolution layer-3; the second convolution layer-3 is connected to the second fully connected layer.
9. The method for inverting impure CO2 plumes in saline layers according to any one of claims 1 to 8, characterized in that: The inversion results of the impure CO2 plume in step (3) include gas saturation and a profile of typical mineral chemical reaction products.
10. The method for inverting impure CO2 plumes in saline layers according to claim 9, characterized in that: The typical minerals include calcite, anorthite and kaolinite.