A method and apparatus for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs
By combining experiments, numerical simulations, and machine learning, the problem of evaluating the non-pure CO2 storage potential in tight sandstone gas reservoirs was solved, achieving accurate evaluation of CO2 storage potential and analyzing storage patterns and influencing factors.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA NAT PETROLEUM CORP
- Filing Date
- 2024-12-25
- Publication Date
- 2026-06-26
AI Technical Summary
Existing technologies are insufficient to effectively evaluate the storage potential of non-pure CO2 in tight sandstone gas reservoirs. In particular, the complex phase changes of the non-pure CO2-CH4 thermodynamic system and the coupling effect of CO2-water-rock chemical reactions make it difficult to predict changes in reservoir permeability characteristics.
By employing a combination of numerical simulation and experimental methods, including experiments on the non-pure CO2-CH4 phase, the non-pure CO2-water-rock reaction, the CO2 burial experiment in long core injection, and the caprock breakthrough pressure test, a numerical simulation mechanism model and a mine model were constructed. The surrogate model was trained using machine learning, and the model with the highest accuracy was selected through multiple algorithms to achieve qualitative and quantitative evaluation.
This study provides an accurate evaluation method for the non-pure CO2 storage potential of tight sandstone gas reservoirs, analyzes the influence of different parameters on the storage effect, and realizes qualitative and quantitative assessment of CO2 storage potential.
Smart Images
Figure CN122287281A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of CO2 geological storage technology, specifically to a method and apparatus for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs. Background Technology
[0002] Carbon dioxide capture, utilization, and storage (CCUS) is a crucial technological approach to reducing carbon emissions and currently the only technological option for achieving large-scale zero-emission utilization of fossil fuels. Due to the long-term accumulation of hydrocarbons in tight sandstone gas reservoirs, ensuring the integrity of the trap, CO2 can be stored in the reservoir for extended periods, similar to natural gas. Tight sandstone gas reservoirs have proven to be a low-cost, high-capacity suitable location for CO2 storage, and CO2 injection into tight sandstone gas reservoirs holds promise as a feasible technology for large-scale CO2 storage. However, the lack of key theoretical research related to the storage of non-pure CO2 gas reservoirs hinders the need for pilot field tests.
[0003] In the process of injecting impure CO2 into tight sandstone gas reservoirs, on the one hand, the phase characteristics of impure CO2 become complex due to drastic temperature and pressure changes after contact with reservoir CH4; on the other hand, the mineral composition of tight sandstone gas reservoirs is mainly quartz and clay minerals, and the CO2-water-rock chemical reaction during impure CO2 injection and storage has a significant impact on reservoir porosity and permeability. To address the problems of incomplete thermodynamic experimental data and wide but uneven temperature distribution of the CO2-CH4 thermodynamic system, various phase equilibrium models of the CO2-hydrocarbon-formation water system have been established, mainly divided into: activity coefficient-fugacity coefficient models and state equation models based on molecular thermodynamics theory. However, due to the dependence of these models on experimental data, the complexity of theoretical formulas, and the large computational burden, they are difficult to apply to numerical simulation software. Furthermore, the impure CO2-water-rock chemical reaction simultaneously causes primary mineral dissolution and secondary mineral precipitation, altering the pore structure and thus affecting reservoir permeability. The coupling of these two phenomena makes the results of the impure CO2-water-rock reaction difficult to predict.
[0004] Therefore, the difficulty in evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs lies in the coupling effect between the phase changes of the non-pure CO2-CH4 thermodynamic system and the non-pure CO2-water-rock chemical reaction, which makes it difficult to evaluate the non-pure CO2 storage potential of tight sandstone gas reservoirs.
[0005] To address the above issues, this invention provides a solution: the collaborative application of numerical simulation and experimental methods to evaluate the non-pure CO2 storage potential of tight sandstone gas reservoirs. The experimental support database is used for machine learning training and numerical modeling. Summary of the Invention
[0006] The purpose of this invention is to address at least one of the aforementioned deficiencies in the prior art. For example, one objective of this invention is to provide a method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs; another objective of this invention is to provide a device for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs.
[0007] To achieve the above objectives, the present invention provides a method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs, the evaluation method comprising:
[0008] Based on the reservoir conditions of the target gas reservoir, experiments were conducted on the non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, CO2 injection and storage in long cores, and caprock breakthrough pressure testing.
[0009] Based on the geological model of the target gas reservoir, a numerical simulation mechanism model and a mining field model are constructed to explore the non-pure CO2 sequestration law and influencing factors of the target gas reservoir and to clarify the CO2 sequestration amount of the target gas reservoir.
[0010] Process experimental and simulation data to construct machine learning datasets;
[0011] A proxy model for predicting the non-pure CO2 geological sequestration of tight sandstone gas reservoirs was constructed by using multiple algorithms for machine learning training.
[0012] The accuracy of the surrogate model is evaluated using regression algorithm evaluation metrics, and the surrogate model with the highest accuracy is selected.
[0013] By combining experimental, numerical simulation and machine learning prediction results, the non-pure CO2 storage potential of the target tight sandstone gas reservoir is evaluated.
[0014] Alternatively, the reservoir conditions may include the original formation pressure, formation temperature, temperature coefficient, pressure coefficient, and formation fluid parameters of the target gas reservoir stratum.
[0015] Optionally, the core drill plunger samples and rocks required for the experiment are taken from the main strata and caprock, ensuring that the porosity and permeability parameters, mineral composition, and mechanical properties are close to the reservoir average.
[0016] Alternatively, the non-pure CO2-CH4 phase experiment aims to adjust the composition of the CO2-CH4 mixture and determine the phase change parameters under different mixture conditions.
[0017] Optionally, the non-pure CO2-water-rock reaction experiment includes CO2-water-rock interaction experiments with different compositions under different temperature and pressure conditions, aiming to clarify the reservoir damage and reservoir sensitivity after CO2 injection in tight sandstone gas reservoirs and the dissolution and precipitation of reservoir minerals by non-pure CO2 injection, wherein formation water and rock powder are referenced or taken from the actual reservoir.
[0018] Alternatively, the CO2 burial experiment of the long core injection can preliminarily clarify the factors and laws that are conducive to CO2 burial by measuring the CO2 burial amount under different non-pure CO2 injection pressure, injection rate and injection gas composition conditions.
[0019] Alternatively, the caprock breakthrough pressure test experiment uses a direct method, including a continuous method, a stepwise method, a displacement method, or a pulse method, to determine the non-pure CO2 breakthrough pressure of the caprock, thereby clarifying the CO2 injection threshold.
[0020] Optionally, the construction of the numerical simulation mechanism model and the mining field model requires the acquisition of geological data of the target tight sandstone gas reservoir, including structural contour maps, sand body thickness distribution contour maps, effective thickness distribution contour maps, porosity distribution contour maps, permeability distribution contour maps, interlayer distribution maps, original formation pressure, temperature, pressure coefficient, temperature gradient, original gas-water / distribution, original gas-water interface, geological reserve report, fault and edge / bottom water distribution.
[0021] Optionally, the construction of the numerical simulation mechanism model and the mining field model requires obtaining production data of the target tight sandstone gas reservoir, including the gas production, liquid production, well fluid composition, bottom hole flowing pressure, casing pressure, and dynamic fluid level of the well; and the daily gas production, daily liquid production, and recovery rate of the target block.
[0022] Alternatively, the investigation into the non-pure CO2 sequestration patterns and influencing factors of the target gas reservoir is conducted on a small scale, with the sequestration volume as the target, by combining mechanistic models and experimental data, adjusting the injection pressure, injection rate, and injection gas composition, to explore the non-pure CO2 sequestration patterns and influencing factors of tight sandstone gas reservoirs.
[0023] Alternatively, the specific target CO2 storage capacity of the gas reservoir is determined by conducting numerical simulations based on production data and field models with CO2 storage capacity as the target. The maximum CO2 storage capacity is taken as the target, and the CO2 injection parameters are adjusted to the peak CO2 storage capacity by integrating the laws obtained from the mechanism model, so as to roughly explore the CO2 storage potential of the target tight sandstone gas reservoir.
[0024] Optionally, the processing of experimental and simulation data includes preprocessing existing data, cleaning and removing outliers, and normalizing the data.
[0025] Alternatively, the construction of the machine learning dataset can be based on experimental data, using a bootstrap method to provide multiple different training sets under the condition of a small training sample size.
[0026] Specifically, this includes: sampling a dataset D containing m samples to generate a dataset D'; randomly selecting samples from D and copying them into D'; then putting the samples back into the original dataset D; repeating this process m times until a dataset D' containing m training samples is obtained. The limiting probability that a sample in D will not be sampled in any of the m samplings is... Approximately 36.8% of the samples in D did not appear in the sampled dataset D'. Therefore, D' was selected as the training set, and D-D' was selected as the test set.
[0027] A dataset is built based on the simulated data. Cross-validation is used to obtain a training set with high stability and fidelity under the condition of a large training sample size. The method includes: dividing the dataset D into 10 mutually exclusive subsets of similar size through stratified sampling; each time, the union of 9 subsets is used as the training set, and the remaining subset is used as the test set, thus obtaining 10 training / test sets. Finally, the mean of these 10 test results is returned. To reduce the difference caused by different sample partitions, the 10-fold cross-validation is repeated 10 times with different random partitions, i.e., 10 times of 10-fold cross-validation, to build the dataset.
[0028] Alternatively, the various algorithms include linear regression, support vector machine, k-nearest neighbor, decision tree, random forest, and gradient boosting algorithms.
[0029] Alternatively, the evaluation of the surrogate model's accuracy using regression algorithm evaluation indicators, and the selection of the surrogate model with the highest accuracy, is based on three major indicators: model fit evaluation, difference evaluation between model predictions and actual values, and the estimation criterion of the maximum likelihood method.
[0030] Among them, the improved coefficient of determination R is selected for evaluating the goodness of fit of the model. 2 The difference between the model's predicted values and the actual values is evaluated using the squared absolute error (MAE), and the maximum likelihood estimation criterion is the Akaike Information Criterion (AIC), in order to optimize the surrogate model.
[0031] Optionally, the evaluation of the non-pure CO2 storage potential of the target tight sandstone gas reservoir by combining experimental, numerical simulation, and machine learning prediction results includes: qualitatively evaluating the non-pure CO2 storage potential of the target tight sandstone gas reservoir based on the synergistic mechanism model simulation of non-pure CO2-CH4 phase experiments, non-pure CO2-water-rock reaction experiments, and long core CO2 burial experiments; and quantitatively evaluating the non-pure CO2 storage potential of the target gas reservoir based on the caprock breakthrough pressure test, field model numerical simulation, and machine learning surrogate model prediction results.
[0032] The present invention provides an evaluation device for the non-pure CO2 storage potential of tight sandstone gas reservoirs, which can be used to implement the above-mentioned evaluation method for the non-pure CO2 storage potential of tight sandstone gas reservoirs.
[0033] The device includes: a gas reservoir carbon sequestration experimental unit, a gas reservoir numerical model establishment unit, a gas reservoir numerical simulation unit, a machine learning training unit, and a model evaluation unit;
[0034] Among them, the gas reservoir carbon sequestration experimental unit is used to carry out experiments on non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, long core CO2 injection and burial experiments, and caprock breakthrough pressure test experiments according to reservoir conditions.
[0035] The gas reservoir numerical model building unit constructs a geological model based on the target tight sandstone gas reservoir data, and then transforms it to construct a gas reservoir numerical simulation mechanism model and a mining field model.
[0036] Based on comprehensive experimental data and mechanistic models, the gas reservoir numerical simulation unit explores the non-pure CO2 sequestration law and influencing factors of tight sandstone gas reservoirs; based on production data and mineral models, the injection parameters are adjusted according to the law obtained from the mechanistic model to predict the maximum CO2 sequestration of the target gas reservoir.
[0037] The machine learning training unit uses bootstrapping and ten-fold cross-validation to process experimental and simulated data respectively, establishes a machine learning training set, performs supervised paradigm learning, and obtains multiple surrogate models.
[0038] The model evaluation unit evaluates the accuracy of the surrogate model based on three major indicators, obtains the best surrogate model to verify the accuracy of the laws obtained from the experiment, and evaluates the non-pure CO2 storage potential of the actual target gas reservoir.
[0039] Compared with the prior art, the beneficial effects of the present invention include at least one of the following:
[0040] (1) The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs provided by the present invention conducts non-pure CO2-CH4 phase state experiments, non-pure CO2-water-rock reaction experiments, long core CO2 injection and burial experiments, and caprock breakthrough pressure test experiments according to reservoir conditions. By adjusting experimental parameters such as temperature, pressure, rock mineral composition, and non-pure CO2 composition, the storage amount, phase transformation and mineral composition change results under different conditions are obtained. The influence of each parameter on the CO2 storage effect is analyzed, providing experimental verification for subsequent numerical simulation and enriching the sample library for subsequent machine learning training.
[0041] (2) The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs provided by the present invention combines experimental, numerical simulation and machine learning prediction results to achieve qualitative and quantitative evaluation of the non-pure CO2 storage potential of tight sandstone gas reservoirs. Attached Figure Description
[0042] The above and other objects and / or features of the present invention will become clearer from the following description taken in conjunction with the accompanying drawings, in which:
[0043] Figure 1 A flowchart illustrating the method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to the present invention is shown.
[0044] Figure 2 A flowchart illustrating the method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to the present invention is shown. Detailed Implementation
[0045] In the following sections, an exemplary embodiment of the present invention will be used to describe in detail a method and apparatus for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs.
[0046] This invention provides a method and apparatus for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs. Based on reservoir conditions, this invention conducts experiments on the non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, long core CO2 injection and burial, and caprock breakthrough pressure testing. Experimental parameters such as temperature, pressure, rock mineral composition, and non-pure CO2 composition are adjusted to obtain the storage capacity, phase transition, and mineral composition changes under different conditions. The influence of each parameter on the CO2 storage effect is analyzed. Numerical simulation using a synergistic mechanism model verifies the non-pure CO2 storage patterns and influencing factors in tight sandstone gas reservoirs. Simultaneously, a subsequent machine learning training database is established. Based on actual conditions... Geological parameters were used to construct a geological model of the target area, followed by a numerical simulation mechanism model and a mining field model. Using experimental parameters as input, the CO2 sequestration law was verified based on the mechanism model. Using CO2 sequestration amount as an indicator, CO2 sequestration simulation was conducted based on the mining field model to predict the non-pure CO2 geological sequestration amount in the tight sandstone gas reservoir of the target area. Based on the data obtained from experiments and numerical simulations, a machine learning sample library was constructed using bootstrapping and cross-validation methods. Supervised learning was conducted using linear regression, support vector machine, k-nearest neighbor, decision tree, random forest, and gradient enhancement algorithms to construct a proxy model for the non-pure CO2 geological sequestration amount in the tight sandstone gas reservoir. A collaboratively improved version of the coefficient of determination R... 2 The model squared absolute error (MAE) and the AIC (Akaike Information Criterion) were used to select the most accurate surrogate model for predicting CO2 geological reserves. Finally, by combining experimental, numerical simulation, and machine learning prediction results, a qualitative and quantitative evaluation of the non-pure CO2 storage potential of tight sandstone gas reservoirs was achieved.
[0047] Exemplary Example 1
[0048] This exemplary embodiment provides a method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs, such as... Figure 1 As shown, the evaluation method includes the following steps.
[0049] S1. Conduct experiments on the non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, CO2 burial in long core, and caprock breakthrough pressure test based on the reservoir conditions of the target gas reservoir.
[0050] In this embodiment, the reservoir conditions include the original formation pressure, formation temperature, temperature coefficient, pressure coefficient, and formation fluid parameters of the target gas reservoir stratum.
[0051] In this embodiment, the core drill plunger samples and rocks required for the experiment are taken from the main strata and caprock, ensuring that the porosity and permeability parameters, mineral composition, and mechanical properties are close to the reservoir average.
[0052] In this embodiment, the non-pure CO2-CH4 phase experiment aims to adjust the composition of the CO2-CH4 mixture and determine the phase change parameters under different mixture conditions.
[0053] In this embodiment, the non-pure CO2-water-rock reaction experiment includes CO2-water-rock interaction experiments with different compositions under different temperature and pressure conditions. The aim is to clarify the reservoir damage and reservoir sensitivity after CO2 injection into tight sandstone gas reservoirs and the dissolution and precipitation of reservoir minerals by non-pure CO2 injection. The formation water and rock powder are referenced or taken from the actual reservoir.
[0054] In this embodiment, the CO2 burial experiment of the long rock core injection was conducted to determine the amount of CO2 burial under different non-pure CO2 injection pressure, injection rate and injection gas composition conditions, so as to preliminarily clarify the factors and laws that are conducive to CO2 burial.
[0055] In this embodiment, the caprock breakthrough pressure test experiment uses a direct method, including the continuous method, the stepwise method, the displacement method, or the pulse method, to determine the non-pure CO2 breakthrough pressure of the caprock, thereby clarifying the CO2 injection threshold.
[0056] S2. Based on the geological model of the target gas reservoir, construct a numerical simulation mechanism model and a mining field model to explore the non-pure CO2 sequestration law and influencing factors of the target gas reservoir and clarify the CO2 sequestration amount of the target gas reservoir.
[0057] In this embodiment, the construction of the numerical simulation mechanism model and the mining field model requires the acquisition of geological data of the target tight sandstone gas reservoir, including the structural contour map, sand body thickness distribution contour map, effective thickness distribution contour map, porosity distribution contour map, permeability distribution contour map, interlayer distribution map, original formation pressure, temperature, pressure coefficient, temperature gradient, original gas-water / distribution, original gas-water interface, geological reserve report, fault and edge / bottom water distribution.
[0058] In this embodiment, the construction of the numerical simulation mechanism model and the mining field model requires obtaining production data of the target tight sandstone gas reservoir, including the gas production, liquid production, well fluid composition, bottom hole flowing pressure, casing pressure, and dynamic fluid level of the well; and the daily gas production, daily liquid production, and recovery rate of the target block.
[0059] In this embodiment, the investigation of the non-pure CO2 sequestration law and influencing factors of the target gas reservoir is carried out on a small scale with the sequestration amount as the target, by combining the mechanism model and experimental data, adjusting the injection pressure, injection rate and injection gas composition, to explore the non-pure CO2 sequestration law and influencing factors of tight sandstone gas reservoirs.
[0060] In this embodiment, the CO2 storage capacity of the target gas reservoir is determined by conducting numerical simulations based on production data and a mining model with CO2 storage capacity as the target. The maximum CO2 storage capacity is taken as the target, and the CO2 injection parameters are adjusted to the peak value of CO2 storage capacity by integrating the laws obtained from the mechanism model, so as to roughly explore the CO2 storage potential of the target tight sandstone gas reservoir.
[0061] S3. Process experimental and simulation data to build machine learning datasets.
[0062] In this embodiment, the processing of experimental and simulation data includes preprocessing existing data, cleaning and removing outliers, and normalizing the data.
[0063] In this embodiment, the construction of the machine learning dataset is based on the experimental data, and the bootstrap method is used to provide multiple different training sets under the condition that the training sample size is small.
[0064] Specifically, this includes: sampling a dataset D containing m samples to generate a dataset D'; randomly selecting samples from D and copying them into D'; then putting the samples back into the original dataset D; repeating this process m times until a dataset D' containing m training samples is obtained. The limiting probability that a sample in D will not be sampled in any of the m samplings is... Approximately 36.8% of the samples in D did not appear in the sampled dataset D'. Therefore, D' was selected as the training set, and D-D' was selected as the test set.
[0065] A dataset is built based on the simulated data. Cross-validation is used to obtain a training set with high stability and fidelity under the condition of a large training sample size. The method includes: dividing the dataset D into 10 mutually exclusive subsets of similar size through stratified sampling; each time, the union of 9 subsets is used as the training set, and the remaining subset is used as the test set, thus obtaining 10 training / test sets. Finally, the mean of these 10 test results is returned. To reduce the difference caused by different sample partitions, the 10-fold cross-validation is repeated 10 times with different random partitions, i.e., 10 times of 10-fold cross-validation, to build the dataset.
[0066] S4. Multiple algorithms are used for machine learning training to construct a proxy model for predicting the geological sequestration of non-pure CO2 in tight sandstone gas reservoirs.
[0067] In this embodiment, the various algorithms include linear regression, support vector machine, k-nearest neighbor, decision tree, random forest, and gradient boosting.
[0068] S5. Evaluate the accuracy of the surrogate model using regression algorithm evaluation indicators, and select the surrogate model with the highest accuracy.
[0069] In this embodiment, the evaluation of the surrogate model's accuracy using regression algorithm evaluation indicators and the selection of the surrogate model with the highest accuracy are based on three major indicators: model fit evaluation, difference evaluation between model predictions and actual values, and the estimation criterion of the maximum likelihood method.
[0070] Among them, the improved coefficient of determination R is selected for evaluating the goodness of fit of the model. 2 The difference between the model's predicted values and the actual values is evaluated using the squared absolute error (MAE), and the maximum likelihood estimation criterion is the Akaike Information Criterion (AIC), in order to optimize the surrogate model.
[0071] S6. Combining experimental, numerical simulation, and machine learning prediction results, evaluate the non-pure CO2 storage potential of the target tight sandstone gas reservoir.
[0072] In this embodiment, the evaluation of the non-pure CO2 storage potential of the target tight sandstone gas reservoir by combining experimental, numerical simulation, and machine learning prediction results includes: qualitatively evaluating the non-pure CO2 storage potential of the target tight sandstone gas reservoir based on the synergistic mechanism model simulation of non-pure CO2-CH4 phase state experiments, non-pure CO2-water-rock reaction experiments, and long core CO2 burial experiments; and quantitatively evaluating the non-pure CO2 storage potential of the target gas reservoir based on the caprock breakthrough pressure test, field model numerical simulation, and machine learning surrogate model prediction results.
[0073] Exemplary Example 2
[0074] This exemplary embodiment provides a method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs, the evaluation method comprising the following steps.
[0075] Step a. Collect representative stratigraphic parameter data and actual core drill plunger samples of the target tight sandstone gas reservoir for use in conducting experiments on non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, long core CO2 injection and burial experiments, and caprock breakthrough pressure tests under reservoir conditions.
[0076] In the embodiments described in this invention, the formation parameter data of the target tight sandstone gas reservoir include the original formation pressure, formation temperature, pressure coefficient, temperature gradient, formation water ion composition, formation fluid pH value, and bound water saturation of the target strata of the gas reservoir under study.
[0077] In the embodiments described in this invention, the actual core drill plunger samples of the target tight sandstone gas reservoir are taken from the main CO2 burial layer and the caprock of the target gas reservoir, respectively. It is necessary to ensure that the porosity, permeability, mineral composition and mechanical properties of the drill plunger samples are representative of the reservoir.
[0078] The non-pure CO2-CH4 phase experiment determined the phase change parameters for different gas mixture compositions by adjusting the proportion of CO2 in the CO2-CH4 mixture and the types of impurities in the mixture, including O2, N2, and H2S.
[0079] The non-pure CO2-water-rock reaction experiment includes CO2-water-rock interaction experiments with different compositions under different temperature and pressure conditions. Formation water and rock powder are referenced or taken from actual reservoirs. This experiment is used to clarify the reservoir damage after CO2 injection into tight sandstone gas reservoirs and to clarify the reservoir sensitivity and the impact of non-pure CO2 injection on the reservoir.
[0080] The CO2 burial experiment conducted on long core samples aims to determine the amount of CO2 burial under different non-pure CO2 injection pressures, injection rates, and injection gas compositions, and to preliminarily clarify the factors and patterns that are conducive to CO2 burial.
[0081] The caprock breakthrough pressure test experiment uses direct methods, including continuous method, stepwise method, displacement method and pulse method, to determine the non-pure CO2 breakthrough pressure of the caprock, thereby clarifying the CO2 injection threshold.
[0082] Furthermore, based on the above experimental studies, the factors affecting CO2 storage and their influence patterns were determined.
[0083] Step b. Establish a geological model based on the target block, import the model into CMG numerical simulation software and coarseen the grid to reduce the number of grids, construct a numerical simulation mechanism model and a mining field model, combine the mechanism model and experimental data to explore the non-pure CO2 storage law of tight sandstone gas reservoirs, and carry out numerical simulation with CO2 storage as the target based on production data and mining field model.
[0084] In the embodiments described in this invention, establishing a geological model based on a target block refers to establishing an actual geological model of the gas reservoir based on the geological data, rock and fluid properties data, and production dynamic data of the target tight sandstone gas reservoir.
[0085] Furthermore, the geological data of the target tight sandstone gas reservoir includes structural contour maps, sand body thickness distribution contour maps, effective thickness distribution contour maps, porosity distribution contour maps, permeability distribution contour maps, interlayer distribution maps, original formation pressure, temperature, pressure coefficient, temperature gradient, original gas-water / distribution, original gas-water interface, geological reserve report, faults, edge / bottom water distribution, etc.
[0086] Furthermore, the rock and fluid properties data of the target tight sandstone gas reservoir include the lithofacies heat capacity, rock compressibility coefficient, and fluid analysis report of the block.
[0087] Furthermore, the production data includes the well's gas production, fluid production, well fluid composition, bottom hole flowing pressure, casing pressure, and dynamic fluid level; the target block's daily gas production, daily fluid production, and recovery rate, etc.
[0088] Furthermore, the study investigates the non-pure CO2 sequestration patterns and influencing factors in the target gas reservoir. By combining mechanistic models and experimental data, small-scale experiments are conducted with the sequestration volume as the target, adjusting injection parameters such as injection pressure, injection rate, and injected gas composition to clarify the non-pure CO2 sequestration patterns and influencing factors in tight sandstone gas reservoirs.
[0089] In the embodiments described in this invention, the numerical simulation based on production data and mine model with CO2 sequestration as the target includes adjusting the CO2 injection parameters to the peak CO2 sequestration value by taking the maximum CO2 sequestration value as the target and integrating the laws obtained from the mechanism model, so as to roughly explore the CO2 sequestration potential of the target tight sandstone gas reservoir.
[0090] Step c. Process the data obtained from experiments and simulations to build a high-quality machine learning dataset.
[0091] In the embodiments described in this invention, the processing of experimental and simulated data includes preprocessing existing data, cleaning and removing outliers, and normalizing the data; and establishing datasets for experimental and simulated data respectively.
[0092] The process involves establishing a dataset based on experimental data. Using a bootstrap method, multiple different training sets are provided given a small training sample size. This includes: sampling a dataset D containing m samples to generate a dataset D'; randomly selecting samples from D and copying them into D'; then returning these samples to the initial dataset D; repeating this process m times until a dataset D' containing m training samples is obtained. The limiting probability that a sample in D will not be sampled in any of the m sampling iterations is:
[0093]
[0094] Approximately 36.8% of the samples in D did not appear in the sampled dataset D'. Therefore, D' was selected as the training set, and D-D' was selected as the test set.
[0095] The process of establishing a dataset based on simulated data and employing cross-validation to achieve a training set with high stability and fidelity under conditions of a large training sample size includes: dividing the dataset D into k mutually exclusive subsets of similar size through stratified sampling, where k is chosen to be 10; each time, the union of 9 subsets is used as the training set, and the remaining subset is used as the test set, resulting in 10 training / test sets; finally, the mean of these 10 test results is returned. To reduce the differences caused by different sample partitions, this 10-fold cross-validation is repeated 10 times using random different partitions, i.e., 10 times 10-fold cross-validation, to establish the dataset.
[0096] Step d. Based on the machine learning training sample library built from experiments and mathematical models, supervised learning is used to construct a proxy model for the non-pure CO2 geological sequestration of tight sandstone gas reservoirs.
[0097] In the embodiments described in this invention, the supervised paradigm learning algorithms include: linear regression algorithm, support vector machine (SVM) algorithm, k-nearest neighbor (K-NN) algorithm, decision tree algorithm, random forest algorithm, and gradient boosting algorithm.
[0098] Step e. Evaluate the accuracy of the surrogate model using multiple regression algorithm evaluation indicators, and select the surrogate model with the highest accuracy in predicting the non-pure CO2 storage capacity of the target tight sandstone gas reservoir.
[0099] In the embodiments described in this invention, the multiple regression algorithm evaluation metrics include: model fit evaluation, difference evaluation between model predictions and actual values, and maximum likelihood estimation criterion.
[0100] Furthermore, the model fit goodness evaluation method selected in this invention is a modified coefficient of determination R0. 2 The calculation formula is as follows:
[0101]
[0102] In the formula, n is the sample size, and k is the number of independent variables (features).
[0103] R 2 The larger the value, the better the model fit is considered to be, and the improved R value is also considered to be... 2 This avoids the influence of the number of independent variables.
[0104] R 2=1, predicted value = actual value, the model interprets the data well;
[0105] R 2 =0, predicted value = mean of actual value;
[0106] R 2 If the value is less than 0, the model is essentially a guess, and the data may not have a linear relationship.
[0107] Furthermore, the method used in this invention to evaluate the difference between the model's predicted values and the actual values is the squared absolute error (MAE), and its calculation formula is as follows:
[0108]
[0109] In the formula, y is the model's predicted value; i is the true value of the sample data; n is the number of samples.
[0110] MAE measures the absolute magnitude of the deviation between the actual value and the predicted value. It is relatively less affected by extreme values and ranges from [0, +∞). It equals 0 when the predicted value perfectly matches the actual value. The larger the error, the larger the value.
[0111] Furthermore, the maximum likelihood estimation criterion selected in this invention is the AIC (Akaike Information Criterion), and its calculation formula is as follows:
[0112] AIC = -2LL max +2k
[0113] In the formula, LL max is the log-likelihood estimate; k is the parameter quantity.
[0114] The smaller the AIC value of the model, the more accurate the model estimation.
[0115] Step f. Combine experimental, simulation, and machine learning prediction results to achieve qualitative and quantitative collaborative evaluation of the non-pure CO2 storage potential of tight sandstone gas reservoirs.
[0116] In the embodiments described in this invention, the qualitative evaluation of the non-pure CO2 storage potential of tight sandstone gas reservoirs includes:
[0117] By comparing the phase transition parameters obtained from the non-pure CO2-CH4 phase state experiment with the target reservoir conditions, the phase state changes of fluids in the reservoir are predicted. Changes in reservoir mineral composition, mineral dissolution, and precipitation are monitored through the non-pure CO2-water-rock reaction experiment to preliminarily evaluate the damage to the reservoir caused by the CO2-water-rock reaction. The changes in rock surface morphology before and after the reaction are compared using three-dimensional laser scanning technology to further evaluate the degree of mineral erosion. A preliminary laboratory-scale evaluation of the target block's storage potential is conducted through a long core CO2 burial experiment. Finally, by combining the long core CO2 burial experiment with mechanistic model numerical simulation, the patterns and influencing factors affecting CO2 burial volume are explored.
[0118] In the embodiments described in this invention, the quantitative evaluation of the non-pure CO2 storage potential of tight sandstone gas reservoirs includes:
[0119] Through caprock breakthrough pressure test experiments, the non-pure CO2 injection rate and injection pressure threshold were determined. The caprock mechanical parameters of CO2 geological storage layers were compared horizontally to evaluate the CO2 storage safety and storage potential of the target block. The non-pure CO2 storage volume of the target tight sandstone gas reservoir was preliminarily predicted by combining production data and mine model. The CO2 storage volume of the target gas reservoir was accurately predicted based on machine learning surrogate model.
[0120] This invention integrates qualitative and quantitative evaluation results such as CO2 damage to reservoirs, CO2 injection parameter thresholds, and CO2 storage prediction, and proposes a method for evaluating the non-pure CO2 storage potential of target tight sandstone gas reservoirs.
[0121] Exemplary Example 3
[0122] This exemplary embodiment provides an apparatus for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs. The apparatus can be used to implement the method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs described in Exemplary Embodiment 1 or Exemplary Embodiment 2.
[0123] The device includes: a gas reservoir carbon sequestration experimental unit, a gas reservoir numerical model establishment unit, a gas reservoir numerical simulation unit, a machine learning training unit, and a model evaluation unit;
[0124] Among them, the gas reservoir carbon sequestration experimental unit is used to carry out experiments on non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, long core CO2 injection and burial experiments, and caprock breakthrough pressure test experiments according to reservoir conditions.
[0125] The gas reservoir numerical model building unit constructs a geological model based on the target tight sandstone gas reservoir data, and then transforms it to construct a gas reservoir numerical simulation mechanism model and a mining field model.
[0126] Based on comprehensive experimental data and mechanistic models, the gas reservoir numerical simulation unit explores the non-pure CO2 sequestration law and influencing factors of tight sandstone gas reservoirs; based on production data and mineral models, the injection parameters are adjusted according to the law obtained from the mechanistic model to predict the maximum CO2 sequestration of the target gas reservoir.
[0127] The machine learning training unit uses bootstrapping and ten-fold cross-validation to process experimental and simulated data respectively, establishes a machine learning training set, performs supervised paradigm learning, and obtains multiple surrogate models.
[0128] The model evaluation unit evaluates the accuracy of the surrogate model based on three major indicators, obtains the best surrogate model to verify the accuracy of the laws obtained from the experiment, and evaluates the non-pure CO2 storage potential of the actual target gas reservoir.
[0129] Exemplary Example 4
[0130] This exemplary embodiment provides an apparatus for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs. The apparatus can be used to implement the method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs described in Exemplary Embodiment 1 or Exemplary Embodiment 2.
[0131] Since the principle behind this device's problem-solving approach is similar to the method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs, the implementation of this device can be found in the implementation of the method, and repetitions will not be repeated. The device described in the following embodiments employs a combination of software and hardware, where "unit" or "module" refers to a combination of software and / or hardware capable of achieving a predetermined function. Figure 2 As shown, the device includes: a gas reservoir carbon sequestration experimental unit C101, a gas reservoir numerical model establishment unit C102, a gas reservoir numerical simulation unit C103, a machine learning training unit C104, and a model evaluation unit C105.
[0132] The specific target functions of the above-mentioned units include:
[0133] Gas reservoir carbon sequestration experimental unit C101: Based on actual reservoir conditions, experiments were conducted on non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, long core CO2 injection and burial, and caprock breakthrough pressure test.
[0134] Unit C102 for establishing numerical models of gas reservoirs: Constructing geological models based on data of target tight sandstone gas reservoirs, and transforming them to construct numerical simulation mechanism models and mining field models of gas reservoirs;
[0135] Numerical simulation unit C103 for gas reservoirs: By integrating experimental data and mechanistic models, the non-pure CO2 sequestration law and influencing factors of tight sandstone gas reservoirs are explored; based on production data and mineral models, injection parameters are adjusted according to the laws obtained from the mechanistic model to predict the maximum CO2 sequestration of the target gas reservoir.
[0136] Machine learning training unit C104: The experimental and simulated data are processed using bootstrapping and ten-fold cross-validation methods respectively to establish a machine learning training set, perform supervised paradigm learning, and obtain multiple surrogate models.
[0137] Model Evaluation Unit C105: Evaluates the accuracy of the surrogate model based on three major indicators, obtains the best surrogate model to verify the accuracy of the laws obtained from the experiment, and evaluates the non-pure CO2 storage potential of the actual target gas reservoir.
[0138] It should be noted that the units and modules described in the above embodiments can be implemented using a computer and a physical experimental device. For ease of description, the devices are divided into units according to their functions in this specification. Of course, when implementing this invention, multiple devices can be placed in the same unit, or multiple units can be implemented in one software.
[0139] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the method provided by the present invention is not limited to the above embodiments. Any technical solution formed by utilizing the methods of the present invention through transformations and substitutions is within the protection scope of the present invention.
[0140] Although the present invention has been described above in conjunction with exemplary embodiments and accompanying drawings, those skilled in the art should understand that various modifications can be made to the above embodiments without departing from the spirit and scope of the claims.
Claims
1. A method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs, characterized in that, The evaluation methods include: Based on the reservoir conditions of the target gas reservoir, experiments were conducted on the non-pure CO2-CH4 phase state, non-pure CO2-water-rock reaction, CO2 injection and storage in long cores, and caprock breakthrough pressure testing. Based on the geological model of the target gas reservoir, a numerical simulation mechanism model and a mining field model are constructed to explore the non-pure CO2 sequestration law and influencing factors of the target gas reservoir and to clarify the CO2 sequestration amount of the target gas reservoir. Process experimental and simulation data to construct machine learning datasets; A proxy model for predicting the non-pure CO2 geological sequestration of tight sandstone gas reservoirs was constructed by using multiple algorithms for machine learning training. The accuracy of the surrogate model is evaluated using regression algorithm evaluation metrics, and the surrogate model with the highest accuracy is selected. By combining experimental, numerical simulation and machine learning prediction results, the non-pure CO2 storage potential of the target tight sandstone gas reservoir is evaluated.
2. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The reservoir conditions include the original formation pressure, formation temperature, temperature coefficient, pressure coefficient, and formation fluid parameters of the target gas reservoir stratum.
3. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The core drill plunger samples and rocks required for the experiment were taken from the main strata and caprock, ensuring that the porosity and permeability parameters, mineral composition, and mechanical properties were close to the reservoir average.
4. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The impure CO2-CH4 phase experiment aims to adjust the composition of the CO2-CH4 mixture and determine the phase change parameters under different mixture conditions.
5. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The non-pure CO2-water-rock reaction experiments include CO2-water-rock interaction experiments with different compositions under different temperature and pressure conditions. The aim is to clarify the reservoir damage and reservoir sensitivity after CO2 injection into tight sandstone gas reservoirs, as well as the dissolution and precipitation of reservoir minerals by non-pure CO2 injection. Formation water and rock powder are referenced or taken from actual reservoirs.
6. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The CO2 burial experiment conducted by injecting long rock cores determined the amount of CO2 burial under different non-pure CO2 injection pressures, injection rates, and injection gas compositions, thus preliminarily clarifying the factors and patterns that are conducive to CO2 burial.
7. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The caprock breakthrough pressure test experiment uses direct methods, including continuous method, stepwise method, displacement method or pulse method, to determine the non-pure CO2 breakthrough pressure of the caprock, thereby clarifying the CO2 injection threshold.
8. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The construction of the numerical simulation mechanism model and the mining field model requires the acquisition of geological data of the target tight sandstone gas reservoir, including the structural contour map of the block, the sand body thickness distribution contour map, the effective thickness distribution contour map, the porosity distribution contour map, the permeability distribution contour map, the interlayer distribution map, the original formation pressure, temperature, pressure coefficient, temperature gradient, original gas-water / distribution, original gas-water interface, geological reserve report, fault and edge / bottom water distribution.
9. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The construction of the numerical simulation mechanism model and the mining field model requires obtaining production data of the target tight sandstone gas reservoir, including the gas production, liquid production, well fluid composition, bottom hole flowing pressure, casing pressure, and dynamic fluid level of the well; and the daily gas production, daily liquid production, and recovery rate of the target block.
10. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The research aims to investigate the non-pure CO2 sequestration patterns and influencing factors in tight sandstone gas reservoirs by combining mechanistic models and experimental data to conduct small-scale investigations with sequestration volume as the target, adjusting injection pressure, injection rate, and injected gas composition to elucidate the non-pure CO2 sequestration patterns and influencing factors.
11. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The specific target CO2 storage capacity of the gas reservoir is determined by conducting numerical simulations based on production data and field models, with the maximum CO2 storage capacity as the target. The CO2 injection parameters are adjusted to the peak CO2 storage capacity by integrating the laws obtained from the mechanism model, so as to roughly explore the CO2 storage potential of the target tight sandstone gas reservoir.
12. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The processing of experimental and simulation data includes preprocessing existing data, cleaning and removing outliers, and normalizing the data.
13. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The construction of the machine learning dataset is based on the experimental data, and the bootstrap method is used to provide multiple different training sets under the condition of a small training sample size.
14. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The various algorithms include linear regression, support vector machine, k-nearest neighbor, decision tree, random forest, and gradient boosting.
15. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The evaluation of the surrogate model's accuracy using regression algorithms and the selection of the most accurate surrogate model are based on three main indicators: model goodness of fit, the difference between the model's predicted values and the actual values, and the estimation criterion of the maximum likelihood method. The improved coefficient of determination R was selected for evaluating the goodness of fit of the model. 2 The difference between the model's predicted values and the actual values is evaluated using the squared absolute error (MAE), and the maximum likelihood estimation criterion is the Akaike Information Criterion (AIC), in order to optimize the surrogate model.
16. The method for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs according to claim 1, characterized in that, The evaluation of the non-pure CO2 storage potential of the target tight sandstone gas reservoir, combining experimental, numerical simulation, and machine learning prediction results, includes qualitative evaluation of the non-pure CO2 storage potential of the target tight sandstone gas reservoir based on the synergistic mechanism model simulation of non-pure CO2-CH4 phase experiments, non-pure CO2-water-rock reaction experiments, and long core CO2 burial experiments; and quantitative evaluation of the non-pure CO2 storage potential of the target gas reservoir based on the caprock breakthrough pressure test, field model numerical simulation, and machine learning surrogate model prediction results.
17. A device for evaluating the non-pure CO2 storage potential of tight sandstone gas reservoirs, characterized in that, The apparatus is used to implement the evaluation method described in claims 1-16; the apparatus includes: a gas reservoir carbon sequestration experimental unit, a gas reservoir numerical model establishment unit, a gas reservoir numerical simulation unit, a machine learning training unit, and a model evaluation unit; wherein... The gas reservoir carbon sequestration experimental unit is used to carry out experiments on the non-pure CO2-CH4 phase state, experiments on the non-pure CO2-water-rock reaction, experiments on CO2 burial in long cores, and experiments on caprock breakthrough pressure testing. The gas reservoir numerical model building unit constructs a geological model based on the target tight sandstone gas reservoir data, and then transforms it to construct a gas reservoir numerical simulation mechanism model and a mining field model. Based on comprehensive experimental data and mechanistic models, the gas reservoir numerical simulation unit explores the non-pure CO2 sequestration law and influencing factors of tight sandstone gas reservoirs; based on production data and mineral models, the injection parameters are adjusted according to the law obtained from the mechanistic model to predict the maximum CO2 sequestration of the target gas reservoir. The machine learning training unit uses bootstrapping and ten-fold cross-validation to process experimental and simulated data respectively, establishes a machine learning training set, performs supervised paradigm learning, and obtains multiple surrogate models. The model evaluation unit evaluates the accuracy of the surrogate model based on three major indicators, obtains the best surrogate model to verify the accuracy of the laws obtained from the experiment, and evaluates the non-pure CO2 storage potential of the actual target gas reservoir.