Fractured bed buried hill oil reservoir gas injection parameter intelligent optimization method and system

Abstract

The present application relates to a kind of fractured bedrock buried hill oil reservoir gas injection parameter intelligent optimization method and system, belong to oil and gas field development field.The method includes establishing buried hill oil reservoir double medium model and buried hill oil reservoir embedded model;Multiple double medium model sample set and multiple embedded model sample set are obtained;The prediction set of double medium model and the prediction set of embedded model are obtained;Final prediction model is obtained;According to the prediction of each set of steam injection parameter in gas injection parameter combination according to final prediction model, obtain the predicted oil production set;Based on the predicted oil production set, obtain the optimized oil reservoir gas injection parameter.In the construction process of high-precision prediction model, the present application fuses the prediction set of double medium model and embedded fracture model as the sample set of prediction model, so that the result of prediction model is more in line with reality, solves the long time consumption of buried hill oil reservoir using different models to optimize gas injection parameter.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas field development, and specifically relates to a method and system for intelligent optimization of gas injection parameters in fractured bedrock buried hill reservoirs. Background Technology

[0002] Buried hill fractured reservoirs are highly heterogeneous, resulting in rapid production increases, short stable production periods, and rapid declines. Water channeling is severe during water injection development, leading to low recovery rates. To ensure stable oil production, gas injection development in buried hill reservoirs is imperative. However, the distinct vertical zonation and severe fracture effects of buried hill reservoirs significantly impact gas injection development. Furthermore, the impact of gas injection at different times is even more pronounced, making the selection of appropriate injection parameters crucial. In addition, while dual-medium models often outperform embedded models in the early stages of buried hill reservoir simulation, their performance declines slightly in the later stages. Therefore, appropriately integrating dual-medium and embedded models to optimize injection parameters is equally important.

[0003] Existing research on reservoir gas injection parameter optimization often uses a single physical model (such as a dual-medium model or an embedded model) to simulate gas injection schemes. However, when optimizing gas injection parameters for buried hill reservoirs, the accuracy of the simulated gas injection parameters is reduced because the single physical model cannot fully reflect the characteristics of the buried hill fractured reservoir. Furthermore, there is no mention of integrating dual-medium models and embedded models to optimize gas injection parameters. Summary of the Invention

[0004] To address the above problems, this invention provides a method and system for intelligent optimization of gas injection parameters in fractured bedrock buried hill reservoirs.

[0005] The first objective of this invention is to provide an intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs, comprising:

[0006] Establish a dual-medium model and an embedded model of a buried hill reservoir;

[0007] Based on the dual-medium model and the embedded model of the buried hill reservoir, sample sets of multiple dual-medium models and sample sets of multiple embedded models were obtained.

[0008] Based on sample sets of multiple dual-media models and multiple embedded models, as well as a predictive agent model of decision trees, the prediction sets of dual-media models and embedded models are obtained.

[0009] The final prediction model is obtained based on the prediction set of the dual-media model, the prediction set of the embedded model, and the random forest prediction model.

[0010] Based on the prediction of each set of steam injection parameters in the final prediction model, the predicted oil production set is obtained.

[0011] Based on the predicted oil production set, the optimized reservoir gas injection parameters are obtained.

[0012] In a specific embodiment of the present invention, the buried hill reservoir dual-medium model includes a weathering zone, a fracture-vuggy zone, and a semi-filled fracture zone;

[0013] The embedded model of the buried hill reservoir includes weathering zone, fracture-vuggy zone and semi-filled fracture zone;

[0014] The weathering zone in the embedded model of the buried hill reservoir has no cracks, but cracks exist in both the fracture-cavity development zone and the semi-filled fracture development zone.

[0015] In a specific embodiment of the present invention, the process of obtaining multiple sample sets of dual-medium models and multiple sample sets of embedded models based on the buried hill reservoir dual-medium model and the buried hill reservoir embedded model includes:

[0016] Based on the dual-medium model or embedded model of buried hill reservoir, multiple different injection timings are determined;

[0017] Based on multiple different injection timings and gas injection parameters for buried hill reservoirs, multiple reservoir gas injection schemes were constructed.

[0018] Simulations were performed on multiple reservoir gas injection schemes using both dual-medium and embedded models, resulting in sample sets for both dual-medium and embedded models.

[0019] In a specific embodiment of the present invention, determining multiple different injection timings based on a buried hill reservoir dual-medium model or a buried hill reservoir embedded model includes:

[0020] Simulations of natural energy extraction are performed on dual-medium models or embedded models of buried hill reservoirs. During the simulation, the simulation is stopped when the reservoir pressure drops to a preset value of the initial pressure.

[0021] Based on the simulation results of natural energy extraction, a formation pressure-cumulative oil production curve is plotted.

[0022] Different injection timings are determined based on the slope changes in the formation pressure-cumulative oil production curve.

[0023] In a specific embodiment of the present invention, the sample set based on multiple dual-media models and multiple embedded models, and the prediction proxy model of the decision tree, to obtain the prediction set of the dual-media model and the prediction set of the embedded model, includes:

[0024] Using sample sets from multiple dual-medium models as test and training sets, a predictive surrogate model based on decision trees is used for iterative training of reservoir gas injection.

[0025] The prediction set of the dual-media model is obtained from the iterative training results of the prediction agent model based on the decision tree.

[0026] Using sample sets of multiple embedded models as test and training sets, the predictive agent model of decision tree is used for iterative training of reservoir gas injection.

[0027] The prediction set of the embedded model is obtained from the iterative training results of the prediction agent model based on the decision tree.

[0028] In a specific embodiment of the present invention, the process of using a sample set of multiple dual-media models as the test set and training set is as follows:

[0029] The sample sets of multiple dual-media models are divided into K-fold data using hierarchical K-fold cross-validation, with the sample sets of multiple dual-media models evenly distributed in each fold.

[0030] Use 1 fold of the K-fold data as the test set and the remaining K-1 fold data as the training set.

[0031] In a specific embodiment of the present invention, during the iterative training, the injected parameters are used as input values ​​and the recovery rate is used as the output value.

[0032] In a specific embodiment of the present invention, the prediction set based on the dual-media model and the prediction set based on the embedded model, along with the random forest prediction model, yields the final prediction model, which includes:

[0033] The prediction sets of the dual-media model and the embedded model were used as the test set and training set, respectively, and the random forest prediction model was trained in one iteration.

[0034] The accuracy of the dual-media model and the embedded model were obtained from the dataset of training results of a single iteration of the random forest prediction model.

[0035] Based on the comparison and processing of the accuracy of the dual-media model and the embedded model, a sample set for the random forest prediction model is obtained.

[0036] Using the sample set of the random forest prediction model as the test set and training set, the random forest prediction model is trained in two iterations.

[0037] Based on the results of the second iteration training of the random forest prediction model, the prediction model with the highest accuracy is selected as the final prediction model.

[0038] In a specific embodiment of the present invention, the comparison and processing of the accuracy of the dual-media model and the embedded model to obtain the sample set of the random forest prediction model includes:

[0039] If the accuracy of the dual-media model is greater than or equal to that of the embedded model, then the predicted values ​​obtained by training the random forest prediction model with the sample data in the prediction set of the dual-media model, along with the corresponding sample data in the prediction set, are used as the sample set of the random forest prediction model.

[0040] If the accuracy of the dual-media model is lower than that of the embedded model, then the sample set of the random forest prediction model is obtained by training the random forest prediction model with the sample data in the prediction set of the embedded model, the actual value in the prediction set of the embedded model, and the corresponding actual value in the prediction set of the dual-media model. The sum of these three values ​​and the arithmetic square root are then taken together with the sample data in the prediction set of the embedded model.

[0041] In a specific embodiment of the present invention, the construction of the gas injection parameter combination includes:

[0042] Determine the range of injection parameters;

[0043] Based on the range of injection parameters, a combination of gas injection parameters is constructed.

[0044] The second objective of this invention is to provide an intelligent optimization system for gas injection parameters in fractured bedrock buried hill reservoirs, comprising:

[0045] The physical model building module is used to build dual-medium models and embedded models of buried hill reservoirs;

[0046] The sample set creation module is used to obtain sample sets of multiple dual-medium models and multiple embedded models based on buried hill reservoir dual-medium models and buried hill reservoir embedded models;

[0047] The prediction set building module is used for sample sets based on multiple dual-media models and multiple embedded models, and the prediction proxy model of decision tree, to obtain the prediction set of dual-media model and the prediction set of embedded model.

[0048] The prediction model building module is used to train the random forest prediction model for reservoir gas injection iterative training using the prediction sets of the dual-medium model and the embedded model as training sets; and to obtain the final prediction model through iterative training based on the random forest prediction model.

[0049] The prediction module is used to predict each set of steam injection parameters in the combination of gas injection parameters based on the final prediction model, so as to obtain the predicted oil production set.

[0050] The optimization module is used to obtain optimized reservoir gas injection parameters based on the predicted oil production set.

[0051] The beneficial effects of this invention are:

[0052] This invention discloses an intelligent optimization method and system for gas injection parameters in fractured bedrock buried hill reservoirs. It simulates gas injection schemes using two physical models: a dual-medium model and an embedded discrete fracture model. This results in a sample set encompassing both models, ensuring that the data largely reflects the unique stratigraphic and fracture development characteristics of the buried hill reservoir. Furthermore, based on this sample set, a higher-precision surrogate model is constructed using machine learning algorithms, leading to a high-precision prediction set and improving the subsequent construction of high-accuracy prediction models. The gas injection scheme addresses the lack of optimization methods for injection parameters at different injection times and pressure levels, thus achieving optimization of injection parameters for various injection times and pressure levels.

[0053] Significantly, in the process of constructing a high-precision prediction model, this invention utilizes an algorithm to integrate the prediction sets of the dual-medium model and the embedded fracture model as the sample set of the prediction model, making the prediction model results more consistent with reality. It also solves the problem of the long time required to optimize gas injection parameters in buried hill reservoirs using different models, and effectively guides the optimization of gas injection parameters in buried hill reservoirs.

[0054] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart of an intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to an embodiment of the present invention is shown;

[0057] Figure 2 A framework diagram of an intelligent optimization system for gas injection parameters in fractured bedrock buried hill reservoirs according to an embodiment of the present invention is shown.

[0058] In the diagram: Physical model building module 1; Sample set building module 2; Prediction set building module 3; Prediction model building module 4; Prediction module 5; Optimization module 6. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] like Figure 1 As shown, an intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to an embodiment of the present invention includes:

[0061] Step S1: Establish a dual-medium model and an embedded model of a buried hill reservoir;

[0062] Step S2: Based on the dual-medium model and the embedded model of the buried hill reservoir, obtain sample sets of multiple dual-medium models and sample sets of multiple embedded models;

[0063] Step S3: Based on the sample sets of multiple dual-media models and multiple embedded models, as well as the predictive agent model of the decision tree, the prediction set of the dual-media model and the prediction set of the embedded model are obtained.

[0064] Step S4: Based on the prediction set of the dual-media model and the prediction set of the embedded model, as well as the random forest prediction model, the final prediction model is obtained.

[0065] Step S5: Based on the prediction of each group of steam injection parameters in the final prediction model, obtain the predicted oil production set.

[0066] Step S6: Based on the predicted oil production set, obtain the optimized reservoir gas injection parameters.

[0067] In this embodiment of the invention, two buried hill reservoir models are established based on the publicly disclosed geological features of Xinggu 7: a buried hill reservoir dual-medium model and a buried hill reservoir embedded model. The buried hill reservoir dual-medium model includes a weathering zone, a fracture-vuggy development zone, and a semi-filled fracture development zone.

[0068] The embedded model of the buried hill reservoir includes weathering zone, fracture-vuggy zone and semi-filled fracture zone;

[0069] The weathering zone in the embedded model of the buried hill reservoir has no cracks, but cracks exist in both the fracture-cavity development zone and the semi-filled fracture development zone.

[0070] The models are 800m long and 200m wide, divided into three layers from top to bottom, with a porosity ratio of 5:3:1 and a permeability ratio of 100:20:1. Each model contains three wells: a production well and two injection wells. All production wells are set to be operational, and the injection wells are designed according to the production plan.

[0071] In step S2, the process of obtaining sample sets of multiple dual-medium models and multiple embedded models based on the buried hill reservoir dual-medium model and the buried hill reservoir embedded model includes:

[0072] Step A1: Based on the dual-medium model or embedded model of buried hill reservoir, determine multiple different injection timings;

[0073] Step A2: Based on multiple different injection timings and gas injection parameters for buried hill reservoirs, construct multiple reservoir gas injection schemes, wherein the reservoir gas injection parameters include gas injection rate and gas injection stage;

[0074] Step A3: Simulate multiple reservoir gas injection schemes for the dual-medium model and the embedded model respectively to obtain sample sets of multiple dual-medium models and multiple embedded models.

[0075] In step A1, the determination of multiple different injection timings based on the buried hill reservoir dual-medium model or the buried hill reservoir embedded model includes:

[0076] Step B1: Simulate the extraction of natural energy from a buried hill reservoir using a dual-medium model or an embedded model. During the simulation, the simulation is stopped when the reservoir pressure drops to a preset value of the initial pressure.

[0077] Step B2: Based on the simulation results of natural energy extraction, create a formation pressure-cumulative oil production curve.

[0078] Step B3: Determine different injection timings based on the slope changes in the formation pressure-cumulative oil production curve.

[0079] In step B1, the preset value is 10% for example, that is, when the reservoir pressure drops to 10% of the initial pressure during the simulation, the simulation is stopped; the preset value can be flexibly set during the actual simulation process, and the present invention does not limit this value.

[0080] In this embodiment of the invention, based on the simulation results, three slope segments are determined for formation pressure: a straight segment, a pressure drop segment, and a flat segment. This results in α(3) injection opportunities, thus completing step A1. Step A2 is then performed, based on the three determined injection opportunities, combined with the gas injection rate and injection segment. In this embodiment of the invention, the gas injection rate is 20*10... 3 m 3 / day, 30*10 3m 3 / day, 40*10 3 m 3 / day, 50*10 3 m 3 / day, 60*10 3 m 3 / day, 70*10 3 m 3 / day, injection section: 1 weathering zone, 2 weathering zone and crevice development zone, 3 fully developed layer;

[0081] The above reservoir gas injection parameters, injection timing, injection rate, and injection stage were obtained using the orthogonal experimental method, resulting in N(18) reservoir gas injection schemes, thus completing step A2. Then, step A3 was performed, which involved using numerical simulation software to simulate multiple reservoir gas injection schemes for the dual-medium model and the embedded model, respectively, to obtain multiple sample sets for the dual-medium model and multiple sample sets for the embedded model. The numerical simulation software could be existing simulation software, such as Petrel or CMG. The above two reservoir models (i.e., the dual-medium model and the embedded model) were simulated for N reservoir schemes at α injection timings. Based on the simulation results, the dual-medium model obtained a 3N1 dataset (i.e., a sample set of multiple dual-medium models), and the embedded model obtained a 3N2 dataset (i.e., a sample set of multiple embedded models). Thus, step A3 was completed.

[0082] Among them, 3N1 and 3N2 constitute the entire sample library, and 3N1 = 3N2;

[0083] 3N1 consists of 31N1, 32N1, 33N1, and so on. 31N1 refers to all sample sets at the first injection timing, 32N1 refers to all sample sets at the second injection timing, and so on. Each sample set includes a set of data on gas injection rate, gas injection stage, injection timing, and recovery rate.

[0084] In step S3, the sample sets based on multiple dual-media models and multiple embedded models, as well as the prediction surrogate model of the decision tree, are used to obtain the prediction sets of the dual-media models and the prediction sets of the embedded models, including:

[0085] Step C1: Using sample sets of multiple dual-medium models as test and training sets, perform iterative training of reservoir gas injection using a decision tree predictive surrogate model.

[0086] Specifically, based on the 3N1 dataset obtained in the above steps, a predictive agent model DT1 for decision trees is constructed and initialized;

[0087] The process of using sample sets from multiple dual-media models as the test and training sets is as follows:

[0088] The sample sets of multiple dual-media models are divided into K folds using hierarchical K-fold cross-validation, with the sample sets of multiple dual-media models evenly distributed in each fold. The 1-fold data in the K-fold data is used as the test set, and the remaining K-1 fold data is used as the training set.

[0089] Even if the hierarchical K-fold cross-validation method is used to divide 3A1 into K folds of data, each fold of data is 31N1, 32N1, 33N1... uniformly distributed, with 1 fold of data as the test set Test and the remaining K-1 folds of data as the training set Train;

[0090] The K-fold data corresponds to K DT1s;

[0091] In the dual-media model of this invention, there are 54 schemes in total. Three-fold cross-validation is used. Data from the beginning, middle and end of each fold are extracted as test data in 18 groups, and training data in 36 groups.

[0092] For the kth DT1 training, set the number of iterations n, select the kth fold of data as the test set Test, and the 1st, 2nd, 3rd...k-1st folds of data as the training set. Take the gas injection rate, gas injection stage, and gas injection timing as inputs and the recovery rate as outputs for training. When the maximum number of iterations is reached, the trained DT1 is obtained.

[0093] Repeat the above training steps continuously to obtain k DT1 values, thus completing step C1;

[0094] Step C2: The iterative training results of the decision tree-based predictive agent model yield the prediction set of the dual-media model;

[0095] In step C1, the iterative training results are k DT1s, and the gas injection rate, gas injection stage, gas injection timing, and recovery rate in the k DT1s are the prediction set of the dual-medium model.

[0096] Step C3: Using the sample sets of multiple embedded models as the test set and training set, use the prediction agent model of the decision tree to perform iterative training of reservoir gas injection.

[0097] Specifically, equivalent to step C1, based on the 3N2 dataset obtained in the above steps, a predictive agent model DT2 of the decision tree is constructed and initialized;

[0098] Similarly, the K-fold cross-validation method is used to divide the 3N2 data into K folds, which are then used as the test set Test and the training set Train.

[0099] The iterative training in step C1 continues until k DT2s are obtained, thus completing step C3.

[0100] Step C4: The prediction set of the embedded model is obtained from the iterative training results of the prediction agent model based on the decision tree.

[0101] In step C3, the iterative training results are k DT2s. The gas injection rate, gas injection stage, gas injection timing, and recovery rate in the k DT2s are the prediction set of the embedded model.

[0102] Based on steps S1-S3, recovery prediction data for two reservoir models (dual-medium model and embedded model) were completed, and then step S4 was performed.

[0103] In step S4, the prediction set based on the dual-media model and the prediction set based on the embedded model, along with the random forest prediction model, yields the final prediction model, including:

[0104] Step D1: Using the prediction sets of the dual-media model and the embedded model as the test set and training set respectively, perform one iteration of training using the random forest prediction model;

[0105] Specifically, the prediction set of the dual-media model is used as the test set and training set. The random forest prediction model Z is used for iterative training. The specific iterative training process is the same as step C1. The K-fold cross-validation method is used to divide k DT1 data into K-fold data, which are then divided into the test set Test and the training set Train for iterative training to obtain k DT1 predictions and obtain new recovery rate prediction values ​​(hereinafter referred to as prediction values, while the recovery rate values ​​predicted by the prediction surrogate model of the decision tree in the k DT1 set are referred to as actual values, and the gas injection rate, gas injection stage, and gas injection timing in the k DT1 set are referred to as sample data).

[0106] Using the prediction set of the embedded model as the test set and training set, the random forest prediction model Z is trained in one iteration, and the specific process is the same as above, to obtain k DT2 predictions and obtain new recovery rate prediction values ​​(hereinafter referred to as prediction values, while the recovery rate values ​​predicted by the prediction surrogate model of the decision tree in the k DT2 set are referred to as actual values, and the gas injection rate, gas injection stage, and gas injection timing in the k DT2 set are referred to as sample data).

[0107] Step D2: Based on the dataset of the first iteration training results of the random forest prediction model, obtain the accuracy of the dual-media model and the embedded model respectively;

[0108] The training result dataset of one iteration of training in step D1 is the above-mentioned predicted values, actual values ​​and their corresponding sample data (k predicted values, actual values ​​and their corresponding sample data of DT1, and k predicted values, actual values ​​and their corresponding sample data of DT2).

[0109] The accuracy rate is the ratio of the number of samples in which the model makes correct predictions (where the actual value and the predicted value are equal) to the total number of samples.

[0110] In the process of comparing the accuracy of DT1 and DT2, the input values ​​(injection rate values) corresponding to DT1 and DT2 should be the same. Only under this premise can the accuracy of DT1 and DT2 be compared.

[0111] In this embodiment of the invention, the accuracy rates of DT1 folding are 0.987, 0.979, and 0.98; and the accuracy rates of DT2 folding are 0.981, 0.976, and 0.982.

[0112] Step D3: Based on the comparison and processing of the accuracy of the dual-media model and the embedded model, the sample set of the random forest prediction model is obtained;

[0113] Specifically, it includes:

[0114] i) If the accuracy of the dual-media model is greater than or equal to the accuracy of the embedded model, then the predicted values ​​obtained by training the random forest prediction model with the sample data in the prediction set of the dual-media model and the corresponding sample data in the prediction set shall be used as the sample set of the random forest prediction model.

[0115] That is, when the accuracy of the kth DT1 is greater than or equal to that of the kth DT2, the predicted value of the kth DT1 in the kth fold data is matched one-to-one with the sample data and used as part of the sample of model Z (the predicted value is used as the actual value of the sample).

[0116] ii) If the accuracy of the dual-media model is less than that of the embedded model, then the sample set of the embedded model is obtained by training the random forest prediction model with the sample data in the prediction set, the actual value in the prediction set of the embedded model, and the actual value in the prediction set of the dual-media model. The sum of these three values ​​is then calculated as the arithmetic square root, along with the sample data in the prediction set of the embedded model.

[0117] That is, when the accuracy of the k-th DT1 is less than that of the k-th DT2, the squares of the predicted value (e.g., x1), the actual value (e.g., x2) of the DT2 at the k-th fold, and the actual value (e.g., x3) of the DT1 at the k-th fold are summed, and then the arithmetic square root is calculated. The x-value is used as the actual value of the recovery rate of the sample set of model Z. It is then matched one by one with the original sample data and used as part of the sample of model Z. This process is repeated to obtain the sample set 3N3 of model Z.

[0118] Because the embedded model cannot account for the influence of small cracks, steps i) and ii) above were set. The setting of these two steps ensures that the final sample set 3N3 data will not have large jump points, thus realizing the integration of the advantages of the dual-medium model and the embedded model in the simulation of buried hill fractured reservoirs, as well as the elimination of their respective disadvantages.

[0119] Step D4: Using the sample set of the random forest prediction model as the test set and training set, perform a second iteration of training on the random forest prediction model;

[0120] That is, using the sample set 3N3 obtained in step D3 as the test set and training set, the random forest prediction model is used for secondary iteration training;

[0121] Specifically, in the iterative training process, the K-fold cross-validation method is also used to divide the 3N3 data into K folds, which are then divided into the test set Test and the training set Train.

[0122] Step D5: Based on the second iteration training results of the random forest prediction model, select the prediction model with the highest accuracy as the final prediction model.

[0123] In step D4, the result of the second iteration training is a new recovery rate prediction value obtained based on the training of sample set 3N3, as well as the corresponding actual value and sample data in sample set 3N3;

[0124] Based on the above data, the accuracy of each Z prediction value is calculated. The Z with the highest accuracy is the final prediction model, which completes step S4. Then, step S5 is performed.

[0125] In step S5, the predicted oil production set is obtained based on the prediction of the gas injection parameter combination by the final prediction model.

[0126] Specifically, it includes:

[0127] a) Based on the actual situation, determine the selectable range of gas injection rate, timing, and injection stage. The gas injection rate is 52*10 3 -60*10 3 The timing for gas injection is 2 and 3, and the injection stages are 1 and 2.

[0128] b) Based on the range determined by the above injection parameters, generate all possible combinations of gas injection parameters;

[0129] In this embodiment of the invention, there are a total of 20 combinations of gas injection parameters, as detailed in Table 1;

[0130] c) Use the final prediction model to predict each gas injection parameter combination in the gas injection parameter combination set to obtain the corresponding predicted oil production. The corresponding predicted oil production is the predicted oil production set, where the predicted oil production is shown in Table 1.

[0131] Step S6: Based on the predicted oil production set, the optimized reservoir gas injection parameters are obtained. That is, according to the predicted oil production set obtained in step S5, the gas injection parameter combination corresponding to the maximum predicted oil production value is selected as the optimal gas injection parameter combination.

[0132] In this embodiment of the invention, the maximum crude oil recovery rate is 38.3% as shown in Table 1, and the selected gas injection rate is 58*10. 3 m 3 The gas injection section is 1, and the injection timing is 2, thus completing the optimization of reservoir gas injection parameters.

[0133] Table 1

[0134]

[0135]

[0136] like Figure 2 As shown, an intelligent optimization system for gas injection parameters in fractured bedrock buried hill reservoirs according to an embodiment of the present invention includes:

[0137] Physical model building module 1 is used to build dual-medium models and embedded models of buried hill reservoirs;

[0138] Sample set creation module 2 is used to obtain sample sets of multiple dual-medium models and multiple embedded models based on buried hill reservoir dual-medium models and buried hill reservoir embedded models;

[0139] The prediction set building module 3 is used for sample sets based on multiple dual-media models and multiple embedded models, and the prediction proxy model of decision tree to obtain the prediction set of dual-media model and the prediction set of embedded model.

[0140] The prediction model building module 4 is used to train the random forest prediction model for reservoir gas injection iterative training using the prediction sets of the dual-medium model and the embedded model as training sets; and to obtain the final prediction model based on the iterative training of the random forest prediction model.

[0141] Prediction module 5 is used to predict each set of steam injection parameters in the gas injection parameter combination based on the final prediction model, so as to obtain the predicted oil production set.

[0142] Optimization module 6 is used to obtain optimized reservoir gas injection parameters based on the predicted oil production set.

[0143] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent optimization of gas injection parameters in fractured bedrock buried hill reservoirs, characterized in that, include: Establish a dual-medium model and an embedded model of a buried hill reservoir; Based on the dual-medium model and the embedded model of the buried hill reservoir, sample sets of multiple dual-medium models and sample sets of multiple embedded models were obtained. Based on sample sets from multiple dual-media models and multiple embedded models, as well as a prediction surrogate model using a decision tree, the prediction sets for the dual-media models and the embedded models are obtained, including: Using sample sets from multiple dual-medium models as test and training sets, a predictive surrogate model based on decision trees is used for iterative training of reservoir gas injection. The prediction set of the dual-media model is obtained from the iterative training results of the prediction agent model based on the decision tree. Using sample sets of multiple embedded models as test and training sets, the predictive agent model of decision tree is used for iterative training of reservoir gas injection. The prediction set of the embedded model is obtained from the iterative training results of the prediction agent model based on the decision tree; Based on the prediction sets of the dual-media model and the embedded model, as well as the random forest prediction model, the final prediction model is obtained, including: The prediction sets of the dual-media model and the embedded model were used as the test set and training set, respectively, and the random forest prediction model was trained in one iteration. The accuracy of the dual-media model and the embedded model were obtained from the dataset of training results of a single iteration of the random forest prediction model. Based on the comparison and processing of the accuracy of the dual-media model and the embedded model, a sample set for the random forest prediction model is obtained, including: If the accuracy of the dual-media model is greater than or equal to that of the embedded model, then the predicted values ​​obtained by training the random forest prediction model with the sample data in the prediction set of the dual-media model, along with the corresponding sample data in the prediction set, are used as the sample set of the random forest prediction model. If the accuracy of the dual-media model is less than that of the embedded model, then the sample set of the embedded model is obtained by training the random forest prediction model with the sample data in the prediction set, the actual value in the prediction set of the embedded model, and the actual value in the prediction set of the dual-media model. The sum of these three values ​​is then calculated as the square root, and the sample data in the prediction set of the embedded model is used as the sample set of the random forest prediction model. Using the sample set of the random forest prediction model as the test set and training set, the random forest prediction model is trained in two iterations. Based on the results of the second iteration training of the random forest prediction model, the prediction model with the highest accuracy is selected as the final prediction model. Based on the prediction of each set of steam injection parameters in the final prediction model, the predicted oil production set is obtained. Based on the predicted oil production set, the optimized reservoir gas injection parameters are obtained.

2. The intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to claim 1, characterized in that, The buried hill reservoir dual-medium model includes a weathering zone, a fracture-vuggy zone, and a semi-filled fracture zone. The embedded model of the buried hill reservoir includes weathering zone, fracture-vuggy zone and semi-filled fracture zone; The weathering zone in the embedded model of the buried hill reservoir has no cracks, but cracks exist in both the fracture-cavity development zone and the semi-filled fracture development zone.

3. The intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to claim 1, characterized in that, Based on the dual-medium model and embedded model of buried hill reservoirs, sample sets of multiple dual-medium models and sample sets of multiple embedded models are obtained, including: Based on the dual-medium model or embedded model of buried hill reservoir, multiple different injection timings are determined; Based on multiple different injection timings and gas injection parameters for buried hill reservoirs, multiple reservoir gas injection schemes were constructed. Simulations were performed on multiple reservoir gas injection schemes using both dual-medium and embedded models, resulting in sample sets for both dual-medium and embedded models.

4. The intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to claim 3, characterized in that, The method, based on the dual-medium model or embedded model of buried hill reservoirs, determines multiple different injection timings, including: Simulations of natural energy extraction are performed on dual-medium models or embedded models of buried hill reservoirs. During the simulation, the simulation is stopped when the reservoir pressure drops to a preset value of the initial pressure. Based on the simulation results of natural energy extraction, a formation pressure-cumulative oil production curve is plotted. Different injection timings are determined based on the slope changes in the formation pressure-cumulative oil production curve.

5. The intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to claim 1, characterized in that, The process of using sample sets from multiple dual-media models as the test and training sets is as follows: The sample sets of multiple dual-media models are divided into K-fold data using hierarchical K-fold cross-validation, with the sample sets of multiple dual-media models evenly distributed in each fold. Use 1 fold of the K-fold data as the test set and the remaining K-1 fold data as the training set.

6. The intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to claim 1, characterized in that, During the iterative training, the injected parameters are used as input values, and the recovery rate is used as the output value.

7. The intelligent optimization method for gas injection parameters in fractured bedrock buried hill reservoirs according to any one of claims 1-6, characterized in that, The construction of the gas injection parameter combination includes: Determine the range of injection parameters; Based on the range of injection parameters, a combination of gas injection parameters is constructed.

8. A smart optimization system for gas injection parameters in fractured bedrock buried hill reservoirs, characterized in that, A method for intelligently optimizing gas injection parameters in fractured bedrock buried hill reservoirs as described in claim 1 includes: The physical model building module is used to build dual-medium models and embedded models of buried hill reservoirs; The sample set creation module is used to obtain sample sets of multiple dual-medium models and multiple embedded models based on buried hill reservoir dual-medium models and buried hill reservoir embedded models; The prediction set building module is used for sample sets based on multiple dual-media models and multiple embedded models, and the prediction proxy model of decision tree, to obtain the prediction set of dual-media model and the prediction set of embedded model. The prediction model building module is used to train the random forest prediction model for reservoir gas injection iterative training using the prediction sets of the dual-medium model and the embedded model as training sets; and to obtain the final prediction model through iterative training based on the random forest prediction model. The prediction module is used to predict each set of steam injection parameters in the combination of gas injection parameters based on the final prediction model, so as to obtain the predicted oil production set. The optimization module is used to obtain optimized reservoir gas injection parameters based on the predicted oil production set.

Images