A high dip angle oil reservoir development mode selection method

Abstract

The present application relates to a kind of high dip angle reservoir development mode selection method, comprising the following steps: selecting domestic and foreign reservoir open data set, open data set includes high dip angle reservoir static parameter;Orthogonal experiment method is applied to each main reservoir parameter and the value level of other each main reservoir parameter is combined respectively;By different opening, the actual value set of each group of main reservoir parameter target parameter is obtained to form concept model library, then it is normalized to each level value;Respectively construct the prediction model based on neural network and based on regression decision tree;Using trained neural network and regression decision tree prediction model, training set and test set are calculated to obtain training set and test set, and the prediction model is trained using training set and test set and obtains the final trained high dip angle reservoir development mode prediction model.The present application can accurately predict each target parameter of high dip angle reservoir, and the predicted value of target parameter is combined with actual need to recommend the best development mode for actual engineering.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field development, and in particular to a method for selecting development modes for high-angle reservoirs. Background Technology

[0002] With the continuous discovery of high-dipping, thick-layered sandstone reservoirs, these reservoirs possess enormous geological reserves, and their efficient development is crucial for ensuring national energy security. Compared to low-dipping, layered reservoirs, high-dipping, thick-layered sandstone reservoirs exhibit multiple oil-bearing strata with significant differences in inter-strata properties and pressure coefficients. The reservoir dip angles are generally high (greater than 10°), significantly influencing crude oil flow due to gravity. Production well productivity varies greatly across different locations (bottom, waist, and top). Limited energy replenishment from bottom natural water bodies to waist and top wells results in low formation pressure and imbalances across different areas. For example, severe degassing occurs in top wells far from edge water, leading to poor development outcomes. Currently, there is no established framework for rationally developing high-dipping, thick-layered sandstone reservoirs, including optimal well types and networks, development methods, energy replenishment timing, injection media, and injection-production ratios. Furthermore, there are no mature development models available for reference for high-dipping reservoirs. Choosing natural energy development methods can reduce reservoir development costs, but prolonged extraction can lead to formation pressure depletion and reduced development effectiveness. While water and gas injection can replenish reservoir energy, water and gas channeling can also significantly impact development results. Some reservoirs are unsuitable for water and gas injection, resulting in poor development improvement, but the cost is substantial. Therefore, quickly selecting a suitable development method is crucial for the development of high-dipping reservoirs.

[0003] Chinese patent CN201810714510.3, entitled "Optimization and Adjustment Method for High-Inclination Reservoirs," proposes a method that combines comprehensive geological research, reservoir engineering research, and pilot experiment results, and uses numerical simulation results to determine areas of dense remaining oil distribution, then arranges and adjusts well networks within these areas. This patent does not address the issue of reservoir development technology policy decision-making methods.

[0004] Chinese patent CN201610355499.7, entitled "A Method for Developing High-Angle Heavy Oil Reservoirs Using Planar Gravity Drive," proposes a driving method that primarily uses steam injection supplemented by non-condensable gas injection. The injected medium, due to density differences, forms spatially differentiated "secondary gas caps" and "secondary water zones," thereby preventing the injected gas from penetrating production wells in lower structural areas. This patent does not mention any decision-making methods for developing high-angle oil reservoirs.

[0005] Chinese patent CN201611158240.X, entitled "A Method for Co-development of High-Angle Heavy Oil Reservoirs Using Fire Flooding and Flue Gas Reinjection in Power Flooding," proposes injecting air from fire flooding injection wells, extracting combustion flue gas from fire flooding production wells, separating and filtering the flue gas, and reinjecting it into flue gas injection wells to provide a gas source for top gas injection. This patent does not mention a method for deciding on the development approach for high-angle reservoirs. Summary of the Invention

[0006] In view of the above-mentioned problems in the existing technology, the technical problem to be solved by the present invention is: how to quickly make decisions on the development method of high-angle reservoirs.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A method for selecting development modes for high-angle reservoirs includes the following steps:

[0009] S01: Domestic and international open-source reservoir datasets are selected. These datasets include static parameters for high-dipping reservoirs. The methods for obtaining these static parameters include natural energy development, bottom water injection development, and bottom water injection + top gas injection development. I high-dipping reservoir static parameters are selected as the primary reservoir parameters. Each primary reservoir parameter includes... Each value level;

[0010] S02: An orthogonal experiment was conducted on I main reservoir parameters and J levels to obtain a total of Each set of conceptual data models corresponds to one orthogonal experiment.

[0011] Will The group concept data model is used as input to geological modeling software to establish... Group concept data model one-to-one correspondence A conceptual model of a high-dipping reservoir;

[0012] S03: Numerical simulation software was applied to analyze three development methods: natural energy development, bottom water injection development, and bottom water injection + top gas injection development. The target parameters of the high-dipping reservoir conceptual model were calculated sequentially to obtain the sample data set. Sample data set and sample data set ;

[0013] The , and To constitute a conceptual model library, and ;

[0014] The , or The data for the m-th sample includes the m-th group of high-dipping reservoir conceptual model and the corresponding actual values ​​of the target parameters;

[0015] S031: The above , and After normalizing all the value levels corresponding to each sample data in the dataset, the corresponding values ​​are obtained. , and The expression after normalization is as follows:

[0016] (1)

[0017] in, This represents the j-th level of the i-th major reservoir parameter in the m-th sample data. Let i represent the normalized data, where i = 1, 2, ..., I, j = 1, 2, ..., J, and m = 1, 2, ..., M;

[0018] S04: Construct a neural network-based predictive agent model (ANN) and initialize it;

[0019] S041: From respectively , and Sample data of the same proportion were selected and merged to form the initial training set Train0. , and The remaining sample data are merged into the initial test set Test0, where one sample data in Train0 is used as a training sample and one sample data in Test0 is used as a test sample.

[0020] S042: Use k-fold cross-validation to process Train0, dividing Train0 into k-fold data;

[0021] S043: Preset number of iterations, each fold of data corresponds to one ANN, and there are a total of k ANNs;

[0022] S044: Select the s-th ANN and its corresponding s-th fold data, take the other k-1 fold data except the s-th fold data as the input of the s-th ANN, train the s-th ANN, and obtain the trained s-th ANN when the training reaches the preset number of iterations.

[0023] S045: Use the s-th fold of data as the input of the trained s-th ANN, and output the predicted dataset of the s-th fold of data;

[0024] S046: Repeat S044 and S045 to train k ANNs, obtain k trained ANNs, and also obtain k prediction datasets. Stack the k prediction datasets to obtain new k-fold data. Combine the predicted value and the corresponding actual value of each training sample in the new k-fold data to obtain a new training set A1.

[0025] S047: Take Test0 from S041 as the input of k trained ANNs to obtain k prediction datasets, where the output of each ANN is a prediction dataset, and each prediction value in the prediction dataset corresponds one-to-one with each test sample data in the initial test set.

[0026] S048: Calculate the arithmetic mean of the k prediction data corresponding to each test sample in Test0 to obtain an average prediction dataset. Combine the average prediction value of each test sample with the corresponding actual value to form a new test set B1.

[0027] S05: Construct a predictive agent model RT based on a regression decision tree and initialize the RT;

[0028] Using the methods described in S041-S046, create a new training set A2; using the method described in S047, create a new test set B2.

[0029] S06: Constructing a predictive model W for the development mode of high-inclination reservoirs based on machine learning algorithms;

[0030] S061: Preset the number of training iterations, initialize W, and train W:

[0031] S062: Merge A1 and A2 as the training set for W. Combine B1 and B2 as the test set for W. ;

[0032] S063: Will The target parameter prediction value is calculated as input to W, and the parameters of W are updated by backpropagation. Training stops when the preset number of training iterations is reached, and the prediction model W' for the current high-angle reservoir development mode is obtained.

[0033] S064: Let t=1;

[0034] S065: From Select the t-th test sample data, input the t-th test sample data into W', and obtain the target parameter prediction value of the t-th test sample data;

[0035] S066: Calculate the correlation coefficient R between the predicted value of the target parameter of the t-th test sample data and the actual value of the target parameter of the t-th test sample data, and define R as the prediction accuracy;

[0036] S067: If R≥85%, the trained high-inclination reservoir development prediction model is obtained, and the next step is executed; otherwise, let t=t+1 in S064. When t is greater than or equal to the maximum number of iterations, the next step is executed; otherwise, return to S065.

[0037] S07: Calculate the predicted values ​​of the target parameters for the high-dipping reservoir using a trained high-dipping reservoir development prediction model. The target parameters are reservoir recovery rate and formation pressure level. The specific steps are as follows:

[0038] S071: The main parameters of the high-dipping reservoir to be predicted were labeled using natural energy development method, bottom water injection development method, and bottom water injection + top gas injection development method, respectively, to obtain three sets of data;

[0039] S072: Input the three sets of data into the trained high-inclination reservoir development mode prediction model to obtain the reservoir recovery rate prediction value and formation pressure level prediction value of the high-inclination reservoir to be predicted under the three development modes;

[0040] S08: Calculate the economic recovery rate of the high-dipping reservoir to be predicted under the three development methods. , The calculation expression is as follows:

[0041] (3)

[0042] in: F is the economic recovery rate; B is the oil-bearing area; S is the average total investment per well; N is the maximum well density; P is the geological reserves; R is the crude oil price; C is the tax rate; and D is the operating cost per ton of oil.

[0043] S09: By comparing the predicted economic recovery rate, reservoir recovery rate, and formation pressure level corresponding to the three development methods, the recommended development methods for the high-dipping reservoirs to be predicted are obtained. The specific steps are as follows:

[0044] S091: For natural energy development methods, if the predicted reservoir recovery rate of a natural energy development method is greater than or equal to the corresponding economic recovery rate, it shall be retained; otherwise, the development method shall be excluded.

[0045] For bottom water injection development, if the predicted reservoir recovery rate is greater than or equal to the corresponding economic recovery rate and the predicted formation pressure level is greater than or equal to 50%, the development method will be retained; otherwise, the development method will be excluded.

[0046] For the bottom water injection + top gas injection development method, if the predicted reservoir recovery rate is greater than or equal to the corresponding economic recovery rate and the predicted formation pressure level is greater than or equal to 50%, it will be retained; otherwise, the development method will be excluded.

[0047] If all three development methods are excluded, it means that the reservoir is not suitable for development.

[0048] If two or three development methods are retained, the development method corresponding to the predicted maximum reservoir recovery rate will be selected as the recommended result.

[0049] Preferably, the geological modeling software used in S02 is PETREL geological modeling software.

[0050] Petrel is the only reservoir fine-grained description and modeling tool fully integrated into a complete reservoir description system. It features a 3D visualization platform for intuitive observation of reservoir changes, and Petrel allows for the simple and rapid creation of conceptual models of high-dipping reservoirs. The reservoir geological models created by Petrel are better suited for reservoir numerical simulation. In the process of creating the reservoir geological model, Petrel fully considers the impact of mesh spatial morphology and mesh structure characteristics on the computational speed of numerical simulation, resulting in geological models with optimal computational performance directly applicable to reservoir numerical simulation.

[0051] Preferably, the numerical simulation software used in S03 is ECLIPSE numerical simulation software.

[0052] The development of the reservoir numerical simulation software Eclipse can better complement Petrel for model simulation. Moreover, Eclipse has strong processing capabilities for black oil models and its technology is relatively mature.

[0053] Preferably, the target parameters in S03 include reservoir recovery rate and formation pressure level.

[0054] The level of reservoir recovery reflects the development level of an oil reservoir. An increase in reservoir recovery corresponds to an increase in the amount of crude oil produced, and it is a major indicator of oilfield development. Crude oil extraction mainly relies on reservoir energy, and the formation pressure level reflects the energy level of the reservoir. Reservoirs with higher formation pressure levels are easier to extract and require less investment. If the formation energy is insufficient, it needs to be supplemented, which will greatly increase the cost of reservoir extraction. Therefore, the energy level of the reservoir is also an important indicator of oilfield development. Thus, these two indicators can very intuitively and conveniently reflect the state of high-dipping reservoirs.

[0055] Preferably, the baseline model used in S06 to construct the high-inclination reservoir development prediction model W based on machine learning algorithm is the LR linear regression model.

[0056] To avoid overfitting, a simple model was chosen, so the LR linear regression model was selected as the baseline model for prediction.

[0057] Preferably, the expression for calculating R in S066 is as follows:

[0058] (2)

[0059] Where H represents the total number of prediction results obtained using the test samples, and z represents the z-th result in H. The actual value of the target parameter. The predicted value for the target parameter.

[0060] Compared with the prior art, the present invention has at least the following advantages:

[0061] 1. This invention discloses a method for selecting development modes for high-dipping reservoirs. By establishing multiple conceptual models considering reservoir static parameters and numerical simulation results, a conceptual model library is constructed. This library uses static parameters such as formation dip angle, thickness, porosity, permeability, and water ratio, along with the development mode, as independent variables, and recovery rate and original formation pressure maintenance level as dependent variables. Then, neural network algorithms and decision tree algorithms from machine learning are applied, using the constructed conceptual model library as the training and prediction sets to train a surrogate model. Stacking is then applied to integrate the two weak learners, neural network and decision tree, to obtain a surrogate model with higher fitting accuracy. Based on this trained model, the static parameters and development modes of other similar high-dipping reservoirs can be used as input variables to quickly predict the recovery rate and original formation pressure maintenance level of different development modes. Comparison of the development effects of different modes can guide oilfield development strategies.

[0062] 2. This invention integrates neural networks and decision tree regression, and uses conceptual model samples for training to construct a high-accuracy prediction model for oil recovery and pressure levels in high-inclination reservoirs, thereby achieving rapid prediction of oil recovery and pressure levels.

[0063] 3. The method disclosed in this invention solves the problems of high difficulty and long time consumption in numerical simulation of high-angle reservoirs. It can effectively guide the selection of development mode in the production area and quickly predict the recovery rate level, and has broad practical engineering application value. Attached Figure Description

[0064] Figure 1Flowchart for selecting development methods for high-angle reservoirs.

[0065] Figure 2a A schematic diagram of a conceptual model for different mining methods: Natural energy mining methods.

[0066] Figure 2b A schematic diagram of a conceptual model for different mining methods: bottom water injection mining method.

[0067] Figure 2c A schematic diagram of a conceptual model for different mining methods: bottom water injection + top gas injection mining method.

[0068] Figure 3 This is a diagram providing an overview of neural networks.

[0069] Figure 4 This is a diagram of the Stacking algorithm framework. Detailed Implementation

[0070] The present invention will now be described in further detail.

[0071] Neural networks can approximate any nonlinear continuous function with arbitrary precision, making them particularly suitable for solving problems with complex internal mechanisms. In high-dipping reservoirs, static parameters such as formation dip angle, thickness, porosity, and permeability, as well as development methods, all influence the reservoir's development effectiveness to varying degrees. Therefore, neural networks can be applied to predict the effects of different development methods for different high-dipping reservoirs. However, the prediction accuracy of a single individual learner is low and has a limit; regardless of the sample size, the accuracy cannot be improved. Therefore, integrating different weak learners into a strong learner can yield a prediction model with high accuracy and stability. Stacking integrates heterogeneous weak learners, allowing them to learn in parallel. By training a second-layer model, these learners are combined, and a final prediction result is output based on the prediction results of different weak models, resulting in a high-accuracy prediction model for further development method decisions. Conventional numerical simulation methods for development method decisions require collecting large amounts of data, which is time-consuming and labor-intensive. Compared to numerical simulation methods, this invention can quickly obtain the optimal development technology policy.

[0072] Example 1: See Figures 1-4 A method for selecting development modes for high-angle reservoirs includes the following steps:

[0073] The method for selecting development modes for high-dipping reservoirs provided by this invention, to further illustrate the effectiveness of this technique, takes a high-dipping reservoir as an example to provide a more detailed description of the implementation of this invention. Figure 1 As can be seen, the specific steps of the present invention are as follows:

[0074] S01: Domestic and international open-source reservoir datasets are selected. These datasets include static parameters for high-dipping reservoirs. The methods for obtaining these static parameters include natural energy development, bottom water injection development, and bottom water injection + top gas injection development. I high-dipping reservoir static parameters are selected as the primary reservoir parameters. Each primary reservoir parameter includes... When implementing the specific value level, parameters can be selected from all static parameters of high-dipping reservoirs according to the prediction target. The main parameters in the static parameters of high-dipping reservoirs include formation dip angle, formation thickness, permeability, porosity and water ratio.

[0075] In specific implementation, based on the reservoir parameters of the embodiment, the main reservoir parameters and their value levels of the conceptual model of the present invention are determined. The formation dip angles are 10°, 15°, 20°, 25°, and 30°, the formation thicknesses are 20, 60, 100, 140, and 180°, the porosities are 20%, 25%, 30%, 35%, and 40%, the permeabilities are 300 mD, 600 mD, 900 mD, 1200 mD, and 1500 mD, and the water volume ratios are 1, 5, 10, 25, and 50.

[0076] S02: An orthogonal experiment was conducted on I main reservoir parameters and J levels to obtain a total of A set of conceptual data models, each set of conceptual data models corresponds to one orthogonal experiment, and the orthogonal experiment method is the existing technology;

[0077] Will The group concept data model is used as input to geological modeling software to establish... Group concept data model one-to-one correspondence A conceptual model of a high-dipping reservoir was developed, and the geological modeling software used was existing technology.

[0078] Each high-inclination reservoir conceptual model has a size of a×b, where a and b represent the length and width of the oilfield represented by each model, respectively. Each oilfield contains c rows of wells, each row contains d production wells, and the well spacing between production wells is e. Under the natural energy development mode, the single-well production of each production well is lp. Under the bottom water injection development mode, the bottom row of production wells is converted into water injection wells, the single-well production of the production wells is lp, and the injection-production ratio is maintained at 1.1, where the injection-production ratio is the ratio of water injection volume to oil production volume, and the single-well injection volume of the water injection wells is lw. Under the bottom water injection + top gas injection development mode, the bottom row of production wells is converted into water injection wells, and the top row of production wells is converted into gas injection wells, the single-well production of the production wells is lp, the injection-production ratio is maintained at 1.1, the single-well injection volume of the water injection wells is lw, and the single-well injection volume of the gas injection wells is lq.

[0079] The geological modeling software used in S02 is PETREL geological modeling software.

[0080] In practice, orthogonal experimental design was used to establish 25 conceptual data models with 5 parameters and 5 levels, representing different parameter combinations. The PETREL geological modeling software was used to establish a conceptual model of a high-dipping reservoir. Each conceptual model was 1200×1200m in size, with 3 rows of wells, each row containing 3 wells, and a well spacing of 400m. Under natural energy development, the single-well production rate was 100m³. 3 / d. Under the bottom water injection development method, the three production wells at the bottom are converted into three water injection wells, with a single well production of 100m³. 3 / d, maintaining an injection-production ratio of 1.1, with a single-well injection volume of 220m³. 3 / d. Under the bottom water injection + top gas injection development method, the three production wells at the bottom are converted into three water injection wells, and the three production wells at the top are converted into three gas injection wells. The production well production is 100m³ / d. 3 / d, maintaining an injection-production ratio of 1.1, with a single-well injection volume of 55m³ / d. 3 / d, the injection rate of a single gas injection well is 6050m³. 3 / d.

[0081] S03: Numerical simulation software was applied to three development methods: natural energy development, bottom water injection development, and bottom water injection + top gas injection development. The target parameters of the high-dipping reservoir conceptual model were calculated sequentially to obtain the sample data set. Sample data set and sample data set The sample data set was obtained. Sample data set and sample data set The process is existing technology;

[0082] The , and To constitute a conceptual model library, and ;

[0083] The , or The m-th sample data includes the m-th group of high-dipping reservoir conceptual model and the corresponding actual values ​​of target parameters; specifically, the number of sample data obtained using natural energy development methods. Number of sample data obtained by using the bottom water injection development method The number of sample data obtained by using the bottom water injection + top air injection development method The three different development methods are distinguished by setting different mining conditions, which will result in different extraction outcomes. The termination condition for the natural energy development method is that the production well's output falls below lp. min The bottom water injection method involves first extracting natural energy, and then switching to water injection at the bottom once the termination conditions are met. The termination condition is when the reservoir water cut reaches fw. max The top gas injection + bottom water injection development method involves first extracting natural energy, then developing with bottom water injection. Top gas injection begins once the bottom water injection termination condition is met, and ends when the gas-oil ratio in the production well reaches g. max ;

[0084] S031: The above , and After normalizing all the value levels corresponding to each sample data in the dataset, the corresponding values ​​are obtained. , and The normalization process uses the min-max normalization method, which is an existing technique. The normalization expression is as follows:

[0085] (1)

[0086] in, This represents the j-th level of the i-th major reservoir parameter in the m-th sample data. Let i represent the normalized data, where i = 1, 2, ..., I, j = 1, 2, ..., J, and m = 1, 2, ..., M;

[0087] The numerical simulation software used in S03 is ECLIPSE numerical simulation software, and the target parameters in S03 include reservoir recovery rate and formation pressure level.

[0088] In practical implementation, the ECLIPSE numerical simulation software is used to study the predicted end-of-period development under the above conceptual model, including natural energy development, bottom water injection, and top gas injection + bottom water injection development methods (the termination condition for natural energy development is that the production well's single-well production is less than 50m³). 3 / d; Bottom water injection method: first, natural energy extraction; when the termination condition is met, bottom water injection begins; the termination condition is when the reservoir water cut reaches 60%; Top gas injection + bottom water injection development method: first, natural energy extraction; then, bottom water injection development; when the bottom water injection termination condition is met, top gas injection begins; the termination condition is when the gas-oil ratio of the production well reaches 500m. 3 / m 3Reservoir recovery rate and pressure maintenance level. Reservoir recovery rate and formation pressure level under different combinations were obtained through numerical simulation and used as a conceptual model library. The obtained conceptual model library was then normalized using max-min standardization.

[0089] S04: Construct a predictive agent model ANN based on a neural network and initialize the ANN. Neural networks are an existing technology. This model can calculate the target parameters of each sample in the conceptual model library. The predictive model constructed by the neural network has a high degree of fit. Using the neural network, the response relationship between reservoir static parameters and development technology policies and target parameters can be found.

[0090] S041: From respectively , and Sample data of the same proportion were selected and merged to form the initial training set Train0. , and The remaining sample data are merged into the initial test set Test0, where one sample data in Train0 is used as a training sample and one sample data in Test0 is used as a test sample.

[0091] S042: Use k-fold cross-validation to process Train0. k-fold cross-validation is an existing technology, mainly used to reduce the risk of overfitting caused by direct training. Train0 is divided into k-fold data.

[0092] S043: Preset number of iterations, each fold of data corresponds to one ANN, and there are a total of k ANNs;

[0093] S044: Select the s-th ANN and its corresponding s-th fold data, and use the other k-1 fold data (excluding the s-th fold data) as the input of the s-th ANN to train the s-th ANN. When the training reaches the preset number of iterations, the trained s-th ANN is obtained; the training process of the neural network is the existing technology.

[0094] S045: Use the s-th fold of data as the input of the trained s-th ANN, and output the predicted dataset of the s-th fold of data;

[0095] S046: Repeat S044 and S045 to train k ANNs, obtaining k trained ANNs and k prediction datasets. Stack the k prediction datasets top to bottom to obtain new k-fold data. Combine the predicted value and the corresponding actual value of each training sample in the new k-fold data to obtain a new training set A1. The stacking method is as follows: Figure 4 As shown;

[0096] S047: Take Test0 from S041 as the input of k trained ANNs to obtain k prediction datasets, where the output of each ANN is a prediction dataset, and each prediction value in the prediction dataset corresponds one-to-one with each test sample data in the initial test set.

[0097] S048: Calculate the arithmetic mean of the k prediction data corresponding to each test sample in Test0 to obtain an average prediction dataset. Combine the average prediction value of each test sample with the corresponding actual value to form a new test set B1.

[0098] In practice, the normalized concept model library is imported into Matlab and a neural network model ANN is built. 80% of the samples in the concept model library are used as the training set (20 groups), and 20% of the samples are used as the test set (5 groups). The training set is divided into four folds, with three folds used as training data and the other fold as prediction data. The resulting four folds of prediction data are stacked top to bottom to obtain a new training set A1 for the second layer model.

[0099] A high-fitting neural network was established to find the response relationship between reservoir static parameters, development technology policies, recovery rate, and formation pressure level. The established neural network prediction model was used to predict the test set, and the arithmetic mean of the prediction values ​​of the four test sets was taken as the test set B1 of the second layer model.

[0100] S05: Construct a predictive agent model RT based on a regression decision tree and initialize the RT. The regression decision tree is an existing technology. This model can calculate the predicted value of the target parameter for each sample in the concept model library. Using the methods described in S041-S046, create a new training set A2 and use the method described in S047 to create a new test set B2.

[0101] In practice, the normalized conceptual model library is imported and a regression decision tree model is built. Similarly, 80% of the samples in the conceptual model library are used as the training set (20 groups), and 20% are used as the test set (5 groups). The training set is divided into four folds, with three folds used as training data and the other fold as prediction data. The resulting four folds of prediction data are stacked vertically to obtain a new training set A2 for the second-layer model. A decision tree regression model is established to find the response relationship between reservoir static parameters, development technology policies, and recovery rate and formation pressure level. The established decision tree regression model is used to predict the test set, and the arithmetic mean of the four test set predictions is taken as the test set B2 for the second-layer model.

[0102] S06: Construct a prediction model W for the development mode of high-dipping reservoirs based on machine learning algorithms, where the machine learning algorithm is an existing technology;

[0103] The baseline model used in S06 to construct the high-inclination reservoir development prediction model W based on machine learning algorithms is the LR linear regression model.

[0104] S061: Preset the number of training iterations, initialize W, and train W:

[0105] S062: Merge A1 and A2 as the training set for W. Combine B1 and B2 as the test set for W. ;

[0106] S063: Will The target parameter prediction value is calculated as input to W, and the parameters of W are updated by backpropagation. Training stops when the preset number of training iterations is reached, and the prediction model W' for the current high-angle reservoir development mode is obtained.

[0107] S064: Let t=1;

[0108] S065: From Select the t-th test sample data, input the t-th test sample data into W', and obtain the target parameter prediction value of the t-th test sample data;

[0109] S066: Calculate the correlation coefficient R between the predicted value of the target parameter of the t-th test sample data and the actual value of the target parameter of the t-th test sample data, and define R as the prediction accuracy;

[0110] The expression for calculating R in S066 is as follows:

[0111] (2)

[0112] Where H represents the total number of prediction results obtained using the test samples, and z represents the z-th result in H. The actual value of the target parameter. The predicted value for the target parameter.

[0113] S067: If R≥85%, the trained high-inclination reservoir development prediction model is obtained, and the next step is executed; otherwise, let t=t+1 in S064. When t is greater than or equal to the maximum number of iterations, the next step is executed; otherwise, return to S065.

[0114] S07: Calculate the predicted values ​​of the target parameters for the high-dipping reservoir using a trained high-dipping reservoir development prediction model. The target parameters are reservoir recovery rate and formation pressure level. The specific steps are as follows:

[0115] S071: The main parameters of the high-dipping reservoir to be predicted were labeled using natural energy development method, bottom water injection development method, and bottom water injection + top gas injection development method, respectively, to obtain three sets of data;

[0116] S072: Input the three sets of data into the trained high-inclination reservoir development mode prediction model to obtain the reservoir recovery rate prediction value and formation pressure level prediction value of the high-inclination reservoir to be predicted under the three development modes;

[0117] In specific implementation, the static parameters and development technology policies of the high-angle reservoir in the example are imported into the high-angle reservoir development mode prediction model to obtain the reservoir recovery rate and formation pressure level of three modes: the recovery rate of natural energy development is 8%, the recovery rate of bottom water injection development is 34% while maintaining 60% of the original formation pressure, and the recovery rate of bottom water injection + top gas injection development mode is 63%, while the formation pressure of this mode can be restored to the original formation pressure.

[0118] S08: Calculate the economic recovery rate for each of the three methods based on actual economic indicators, and calculate the economic recovery rate of the predicted high-angle reservoir under each of the three development methods. , The calculation expression is as follows:

[0119] (3)

[0120] in: Economic recovery rate (%); F represents oil-bearing area (km²); B represents average total investment per well (in ten units). 4 Yuan / well; S is the ultimate well network density, unit: wells / km 2 N represents geological reserves, in units of 10. 4 t; P is the price of crude oil, in yuan / t; R is the tax rate, in yuan / t; C is the operating cost per ton of oil, in yuan / t; different types of development methods involve different types and numbers of single wells, so the relevant parameter values ​​also differ;

[0121] S09: By comparing the predicted economic recovery rate, reservoir recovery rate, and formation pressure level corresponding to the three development methods, the recommended development methods for the high-dipping reservoirs to be predicted are obtained. The specific steps are as follows:

[0122] S091: For natural energy development methods, if the predicted reservoir recovery rate of a natural energy development method is greater than or equal to the corresponding economic recovery rate, it shall be retained; otherwise, the development method shall be excluded.

[0123] For bottom water injection development, if the predicted reservoir recovery rate is greater than or equal to the corresponding economic recovery rate and the predicted formation pressure level is greater than or equal to 50%, the development method will be retained; otherwise, the development method will be excluded.

[0124] For the bottom water injection + top gas injection development method, if the predicted reservoir recovery rate is greater than or equal to the corresponding economic recovery rate and the predicted formation pressure level is greater than or equal to 50%, it will be retained; otherwise, the development method will be excluded.

[0125] If all three development methods are excluded, it means that the reservoir is not suitable for development.

[0126] If two or three development methods are retained, the development method corresponding to the predicted maximum reservoir recovery rate will be selected as the recommended result.

[0127] In specific implementation, in this embodiment, the oil recovery rate of the reservoir developed using natural energy is 8%, and the formation energy level remains low, below the bubble point pressure of the reservoir, thus failing to restore the formation pressure level. The oil recovery rate of the reservoir developed using bottom water injection is 34%, which is higher than the economic recovery rate of this development method. At this oil recovery rate, the cost of the water injection process can be guaranteed, and the formation pressure level can be restored, maintaining 60% of the original formation pressure. Therefore, the bottom water injection method is feasible. The oil recovery rate of the reservoir developed using bottom water injection + top gas injection is 63%, also higher than the corresponding economic recovery rate. At this oil recovery rate, the cost of the water injection and gas injection process can be guaranteed, and the formation pressure level can be restored to the original formation pressure level. Therefore, the bottom water injection + top gas injection method is feasible. Comparing the two development methods, the bottom water injection + top gas injection method is clearly the most effective. Therefore, the bottom water injection + top gas injection method is selected as the recommended development method for this project, thus completing the rapid decision-making and recommendation of the development method for this high-dipping reservoir.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for selecting development modes for high-angle reservoirs, characterized in that: The steps include the following: S01: Domestic and international open-source reservoir datasets are selected. These datasets include static parameters for high-dipping reservoirs. The methods for obtaining these static parameters include natural energy development, bottom water injection development, and bottom water injection + top gas injection development. I high-dipping reservoir static parameters are selected as the primary reservoir parameters. Each primary reservoir parameter includes... The main reservoir parameters include formation dip angle, formation thickness, permeability, porosity, and water ratio, and are available in several value levels. S02: An orthogonal experiment was conducted on I main reservoir parameters and J levels to obtain a total of Each set of conceptual data models corresponds to one orthogonal experiment. Will The group concept data model is used as input to geological modeling software to establish... Group concept data model one-to-one correspondence A conceptual model of a high-dipping reservoir; S03: Numerical simulation software was applied to analyze three development methods: natural energy development, bottom water injection development, and bottom water injection + top gas injection development. The target parameters of the high-dipping reservoir conceptual model were calculated sequentially to obtain the sample data set. Sample data set and sample data set ; The , and To constitute a conceptual model library, and ; The , or The data for the m-th sample includes the m-th group of high-dipping reservoir conceptual model and the corresponding actual values ​​of the target parameters; S031: The above , and After normalizing all the value levels corresponding to each sample data in the dataset, the corresponding values ​​are obtained. , and The expression after normalization is as follows: ；（1） in, This represents the j-th level of the i-th major reservoir parameter in the m-th sample data. Let i = 1, 2, ..., I, j = 1, 2, ..., J, m = 1, 2, ..., M; S04: Construct a neural network-based predictive agent model (ANN) and initialize it; S041: From respectively , and Sample data of the same proportion were selected and merged to form the initial training set Train0. , and The remaining sample data are merged into the initial test set Test0, where one sample data in Train0 is used as a training sample and one sample data in Test0 is used as a test sample. S042: Use k-fold cross-validation to process Train0, dividing Train0 into k-fold data; S043: Preset number of iterations, each fold of data corresponds to one ANN, and there are a total of k ANNs; S044: Select the s-th ANN and its corresponding s-th fold data, take the other k-1 fold data except the s-th fold data as the input of the s-th ANN, train the s-th ANN, and obtain the trained s-th ANN when the training reaches the preset number of iterations. S045: Use the s-th fold of data as the input of the trained s-th ANN, and output the predicted dataset of the s-th fold of data; S046: Repeat S044 and S045 to train k ANNs, obtain k trained ANNs, and also obtain k prediction datasets. Stack the k prediction datasets to obtain new k-fold data. Combine the predicted value and the corresponding actual value of each training sample in the new k-fold data to obtain a new training set A1. S047: Take Test0 from S041 as the input of k trained ANNs to obtain k prediction datasets, where the output of each ANN is a prediction dataset, and each prediction value in the prediction dataset corresponds one-to-one with each test sample data in the initial test set. S048: Calculate the arithmetic mean of the k prediction data corresponding to each test sample in Test0 to obtain an average prediction dataset. Combine the average prediction value of each test sample with the corresponding actual value to form a new test set B1. S05: Construct a predictive agent model RT based on a regression decision tree and initialize the RT; Using the methods described in S041-S046, create a new training set A2; using the method described in S047, create a new test set B2. S06: Constructing a predictive model W for the development mode of high-inclination reservoirs based on machine learning algorithms; S061: Preset the number of training iterations, initialize W, and train W: S062: Merge A1 and A2 as the training set for W. Combine B1 and B2 as the test set for W. ; S063: Will The target parameter prediction value is calculated as input to W, and the parameters of W are updated by backpropagation. Training stops when the preset number of training iterations is reached, and the prediction model W' for the current high-angle reservoir development mode is obtained. S064: Let t=1; S065: From Select the t-th test sample data, input the t-th test sample data into W', and obtain the target parameter prediction value of the t-th test sample data; S066: Calculate the correlation coefficient R' between the predicted value of the target parameter of the t-th test sample data and the actual value of the target parameter of the t-th test sample data, and define R' as the prediction accuracy; S067: If R'≥85%, then the trained high-dip reservoir development prediction model is obtained, and the next step is executed; otherwise, let t=t+1 in S064 and return to S065. S07: Calculate the predicted values ​​of the target parameters for the high-dipping reservoir using a trained high-dipping reservoir development prediction model. The target parameters are reservoir recovery rate and formation pressure level. The specific steps are as follows: S071: The main parameters of the high-dipping reservoir to be predicted were labeled using natural energy development method, bottom water injection development method, and bottom water injection + top gas injection development method, respectively, to obtain three sets of data; S072: Input the three sets of data into the trained high-inclination reservoir development mode prediction model to obtain the reservoir recovery rate prediction value and formation pressure level prediction value of the high-inclination reservoir to be predicted under the three development modes; S08: Calculate the economic recovery rate of the high-dipping reservoir to be predicted under the three development methods. , The calculation expression is as follows: ；（3） in: F is the economic recovery rate; B is the oil-bearing area; S is the average total investment per well; N is the maximum well density; P is the geological reserves; R is the crude oil price; C is the tax rate; and D is the operating cost per ton of oil. S09: By comparing the predicted economic recovery rate, reservoir recovery rate, and formation pressure level corresponding to the three development methods, the recommended development methods for the high-dipping reservoirs to be predicted are obtained. The specific steps are as follows: S091: For natural energy development methods, if the predicted reservoir recovery rate of a natural energy development method is greater than or equal to the corresponding economic recovery rate, it shall be retained; otherwise, the development method shall be excluded. For bottom water injection development, if the predicted reservoir recovery rate is greater than or equal to the corresponding economic recovery rate and the predicted formation pressure level is greater than or equal to 50%, the development method will be retained; otherwise, the development method will be excluded. For the bottom water injection + top gas injection development method, if the predicted reservoir recovery rate is greater than or equal to the corresponding economic recovery rate and the predicted formation pressure level is greater than or equal to 50%, it will be retained; otherwise, the development method will be excluded. If all three development methods are excluded, it means that the reservoir is not suitable for development. If two or three development methods are retained, the development method corresponding to the predicted maximum reservoir recovery rate will be selected as the recommended result.

2. The method for selecting a development mode for a high-angle reservoir as described in claim 1, characterized in that: The geological modeling software used in S02 is PETREL geological modeling software.

3. The method for selecting a development mode for a high-angle reservoir as described in claim 2, characterized in that: The numerical simulation software used in S03 is ECLIPSE numerical simulation software.

4. The method for selecting a development mode for a high-angle reservoir as described in claim 3, characterized in that: The target parameters in S03 include reservoir recovery rate and formation pressure level.

5. The method for selecting a development mode for a high-angle reservoir as described in claim 4, characterized in that: The baseline model used in S06 to construct the high-inclination reservoir development prediction model W based on machine learning algorithms is the LR linear regression model.

6. A method for selecting a development mode for a high-angle reservoir as described in any one of claims 1-5, characterized in that: In S066 The calculation expression is as follows: ；（2） Where H represents the total number of prediction results obtained using the test samples, and z represents the z-th result in H. The actual value of the target parameter. The predicted value for the target parameter.

Images