A method for optimizing design of fracture parameter matching relationship of horizontal well in complex coal structure

Abstract

The application discloses a complex coal body structure horizontal well fracture parameter matching relationship optimization design method, including the following steps: step one: through the Petrel-Re numerical simulator, a fine numerical simulation model of a coal seam with complex coal body structure is established; step two: through a numerical simulation method, main control factors of a coalbed methane fracturing horizontal well development effect are screened out.The application combines various test results of coal rock cores, uses a Petrel-Re numerical simulation software to construct a numerical simulation model considering characteristics of a multi-layer superimposed coal seam, such as coalbed methane adsorption-desorption-diffusion, Lagmuir isothermal adsorption and interlayer heterogeneity, etc.Meanwhile, a fracturing horizontal well model is constructed by using artificial fractures, a numerical simulation method is adopted to screen out main control factors affecting coalbed methane productivity.Meanwhile, the application uses Python to call the numerical simulator to realize self-generation of numerical simulation samples, uses an XGboost algorithm to realize construction of a yield prediction proxy model, and then obtains an optimal solution of a coalbed methane fracturing horizontal well fracture parameter combination.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of coalbed methane pressure analysis, in particular to a complex coal structure horizontal well fracture parameter matching relationship optimization design method. BACKGROUND

[0002] Most of the existing coalbed methane reservoir development adopts hydraulic fracturing technology, and the parameter optimization of hydraulic fractures often adopts numerical simulation method. The current numerical simulation method is mainly based on continuous medium model (double medium, multiple medium and equivalent continuous medium model), which considers Knudsen diffusion, molecular diffusion and adsorbed gas desorption to predict the productivity of coalbed methane fracturing horizontal well and optimize the fracture parameters. The numerical simulation method has the advantages of repeatability, short time, low cost and simulation of various heterogeneous conditions, but the grid construction and numerical calculation process of the numerical simulation method considering the flow characteristics of the fracturing horizontal well are complex. Moreover, the numerical simulation technology always involves solving large-scale partial differential equations, which requires a lot of calculation time, making the optimization of the whole area fracturing parameters impractical.

[0003] At present, intelligentization of oil and gas exploration and development has become an industry hotspot and development trend. Machine learning methods such as support vector machine and neural network are used to process the geological parameters and field production parameters of coalbed methane reservoirs, and a productivity prediction model is established by analyzing the potential patterns in a large amount of data. Compared with the traditional numerical simulation method, the machine learning method not only shortens the modeling period and speeds up the calculation, but also has better scalability. Therefore, on the basis of identifying the main control factors affecting the productivity of coalbed methane, a coalbed methane reservoir productivity prediction model based on machine learning can be established to further optimize the fracturing parameters and improve the fracturing technology design level. Thus, the fracturing scheme design and the whole life cycle production management of coalbed methane wells are guided. SUMMARY

[0004] In order to solve the problems existing in the prior art, a complex coal structure horizontal well fracture parameter matching relationship optimization design method is provided.

[0005] The present application solves the above technical problems by the following technical solutions, and the present application comprises the following steps:

[0006] Step one: a fine numerical simulation model of complex coal structure coal seam is established by Petrel-Re numerical simulator;

[0007] Step two: the main control factors of coalbed methane fracturing horizontal well development effect are screened by numerical simulation method;

[0008] Step three: a numerical simulation sample self-generation model is constructed;

[0009] Step four: a coalbed methane productivity prediction proxy model based on XGBoost algorithm is established.

[0010] Further in, the specific process of step one is as follows:

[0011] The grid system of the basic model is established by using Petrel-Re reservoir numerical simulation software;

[0012] The fluid model adopts black oil model, considering Lagmuir isothermal adsorption and Fick diffusion;

[0013] The grid number of the conceptual model is 150*51*51, the grid step length in X and Y directions is 10m, the grid step length in Z direction is 0.5m, the model length is 1500m, and the model width is 510m;

[0014] A total of 8 sub-layers are divided, among which the second layer and the sixth layer are coal seams, and the remaining are rock and soil layers other than coal seams;

[0015] The input parameters of the model are determined based on coal rock core experiment data and mine data, the gas-water two-phase is taken as the research object, the dual-porosity and dual-permeability model is selected, and the Intersect simulator is used for numerical simulation operation.

[0016] Further in, the specific process of step two is as follows:

[0017] First, the influencing factor analysis of the productivity of the coalbed methane fracturing horizontal well is carried out:

[0018] Including matrix porosity, gas saturation, artificial fracture conductivity, artificial fracture half length, fracture height, fracture spacing, and horizontal section length;

[0019] Second, the data range design of the main control factors is carried out:

[0020] Through parameter sensitivity analysis, four key influencing factors are identified, including the half length of the artificial fracture, the distance between the fractures, the conductivity of the fracture, and the height of the fracture;

[0021] The full-factor experimental design method involves the combination of all factors at all levels, and at least one test is carried out to estimate the main effect of each factor and its interaction.

[0022] Further in, the specific process of step three is as follows:

[0023] The numerical simulation file is automatically generated:

[0024] Based on the data range of the main control factors screened in step two, the simulation scheme is randomly combined, the ixf file of the intersect simulator is used, the function is defined by Python to assemble the keywords, the automatic filling of the fracture length, the fracture height, the fracture conductivity, and the fracture spacing is realized, and the AFI file is used to assemble the design scheme into a file;

[0025] After that, the numerical simulator operation automatic batch call and result extraction are carried out:

[0026] Based on the Python call of the Intersect environment in the computer, the AFI file automatically generated is batch processed, and the required numerical simulation result file is extracted and saved to a new folder for later simulation result extraction and combination and arrangement;

[0027] Based on the numerical simulation result file (PRT file), by marking the coalbed methane geological reserves of the first time node and the last time node of each numerical simulation result file, a loop function is defined, the cumulative gas production of different schemes is automatically extracted and formatted and saved to the CSV file, and the whole process automation of the simulator result extraction and saving is realized.

[0028] Further, the specific process of the fourth step includes: prediction sample set division, prediction model establishment and model prediction effect evaluation.

[0029] Further, the process of the prediction sample set division is as follows: the prediction sample set obtained from the numerical simulation sample self-generation model is divided into a training set and a validation set according to the related proportions of 80% and 20%;

[0030] The training set is mainly used for training the prediction model;

[0031] The validation set is mainly used for adjusting hyperparameters, including the number of decision trees, the learning rate, and the maximum depth of the tree;

[0032] Each set contains the input and output data of the model, wherein the input data includes the fracture height, the fracture length, the fracture conductivity and the fracture spacing, and the output data includes the cumulative gas production of the coalbed methane.

[0033] Further, the process of the prediction model establishment is as follows: the XGBoost algorithm is used for modeling, and the implementation process of the XGBoost algorithm is as follows:

[0034] (1) Let i represents the i-th sample in the training set (X, y)

[0035] (2) Satisfy 1≤t≤N k ;

[0036] ① Construct the cost function:

[0037]

[0038] Where, g i and h i respectively represent the first-order derivative and the second-order derivative of the loss function l at the sample i; y is the true label value of the sample; represents an XGBoost prediction model, x i is an input feature parameter of a decision tree model; r i represents a residual, N k is the number of decision tree models; T represents the number of leaves of the tree model, γ, λ represent regularization coefficients, and ω j represents a leaf node value.

[0039] ②Using the data set (X, r), a tree model with a maximum depth of d is trained

[0040] ③Updating the prediction value of the XGBoost model

[0041]

[0042] ④Updating the residual term r

[0043]

[0044] (3) Obtaining the final prediction model

[0045]

[0046] where ε represents a compression factor or a learning rate.

[0047] Further, the process of establishing an XGBoost model includes the following steps:

[0048] First, by continuously adding trees, feature classification is performed, and each newly generated tree learns a new function f(x) to gradually grow a complete tree. In this way, each time a tree is added, the residual of the last prediction can be fitted, and the classification model can be continuously improved to improve the accuracy of the prediction;

[0049] Second, after training k trees, sample score prediction is needed. According to the characteristics of the sample, it is classified into the leaf nodes of each tree, and the corresponding scores are obtained;

[0050] Third, only the sum of the scores calculated by each tree is the final prediction value corresponding to the sample.

[0051] Further, the specific process of evaluating the prediction effect of the model is as follows:

[0052] The prediction effect of the production capacity prediction model is evaluated and analyzed by using the determination coefficient (R 2 ), mean square error (MSE).

[0053]

[0054] In the formula, y iand respectively, is the average value of the actual value, and N is the size of the actual value in the verification set.When R 2 The smaller the MSE is, the smaller the error between the predicted value and the actual value of the model is, and the better the performance of the model is.

[0055] Compared with the prior art, the complex coal body structure horizontal well fracture parameter matching relationship optimization design method has the following advantages: the numerical simulation sample self-generation program solves the problem of high labor cost in manual modification of numerical simulation parameters for numerical calculation, and based on the numerical simulation sample self-generation, a large number of accurate sample sets for coalbed methane production capacity prediction can be quickly generated. Moreover, based on the established coalbed methane production capacity prediction proxy model, the traditional optimization method can only screen out a local optimal solution, the time cost of simulation is reduced, the work efficiency is improved, and the system is more worthy of promotion and use. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 is a fine numerical simulation model of the complex coal body structure coal seam of the application;

[0057] Figure 2 is a numerical simulation sample self-generation model establishment flowchart of the application;

[0058] Figure 3 is a fracture parameter optimization flowchart based on the production capacity prediction proxy model of the application;

[0059] Figure 4 is a diagram showing the influence of the hydraulic fracture segment spacing on the daily gas production and the cumulative gas production of the application;

[0060] Figure 5 is a diagram showing the influence of the hydraulic fracture segment spacing on the daily water production and the cumulative water production of the application;

[0061] Figure 6 is a diagram showing the influence of the hydraulic fracture conductivity on the daily gas production and the cumulative gas production of the application;

[0062] Figure 7 is a diagram showing the influence of the hydraulic fracture conductivity on the daily water production and the cumulative water production of the application;

[0063] Figure 8 is a diagram showing the influence of the hydraulic fracture height on the cumulative gas production of the application;

[0064] Figure 9 is a diagram showing the influence of the hydraulic fracture height on the cumulative gas production of the application;

[0065] Figure 10 is a diagram showing the comparison result of the optimization method and the original scheme production capacity of the application. Detailed Implementation

[0066] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0067] like Figures 1-10 As shown, this embodiment provides a technical solution: a method for optimizing the matching relationship of fracture parameters in a horizontal well with a complex coal seam structure, comprising the following steps:

[0068] The first step is to establish a detailed numerical simulation model of the complex coal seam structure using the Petrel-Re numerical simulator.

[0069] A basic model grid system was established using Petrel-Re reservoir numerical simulation software. The fluid model adopted the black oil model, considering Lagmuir isothermal adsorption and Fick diffusion. The conceptual model has a grid size of 150×51×51, with a grid step size of 10m in the X and Y directions and 0.5m in the Z direction. The model is 1500m long and 510m wide. It is divided into eight sub-layers, with the second and sixth layers being coal seams, and the remainder being soil and rock layers other than coal seams. In the gas reservoir model, most horizontal well sections are located in the lower coal seams, with the remaining sections located in the intermediate layers. The predicted production cycle is 15 years. Hydraulic fractures are assumed to propagate along the direction of maximum principal stress. Local mesh refinement techniques and the equivalent conductivity method are used to accurately simulate artificial hydraulic fractures.

[0070] The input parameters of the model were determined using experimental data from coal and rock cores and mine data. Taking the gas-water two-phase system as the research object, a dual-pore dual-permeability model was selected, and numerical simulation calculations were performed using the Intersect simulator.

[0071] The second step involves using numerical simulation to screen the main controlling factors of the development effect of coalbed methane fracturing horizontal wells and analyzing the factors affecting the productivity of coalbed methane fracturing horizontal wells.

[0072] Many factors influence coalbed methane (CBM) production, such as matrix porosity, gas saturation, conductivity of artificial fractures, half-length of artificial fractures, fracture height, fracture spacing, and length of horizontal sections. This invention starts with factors directly related to CBM production capacity and identifies the main controlling factors affecting CBM development effectiveness. Using numerical simulation methods, a sensitivity scheme is set up to conduct sensitivity studies on different parameters, thereby revealing the main factors affecting coal seam production capacity.

[0073] Data range design of controlling factors

[0074] Four key factors were identified through careful parameter sensitivity analysis, including the half-length of artificial fractures, the distance between fractures, the fracture conductivity and the fracture height. Full factorial design involves all combinations of factors at all levels, at least one test to estimate the main effect of each factor and its interaction. This design method can obtain a large amount of information, accurately estimate the main influence of test factors and the interaction between factors. The value range of each influencing factor is based on the field data and parameter sensitivity analysis in the study area. Specifically, the fracture half-length range is set to 60-150 m, the fracture spacing range is 10-110 m, and the fracture conductivity range is 50-300 mD·m. The fracture height design uses two methods, one is to penetrate two coal seams with a fracture height of 25.5 m, and the other is to only penetrate the lower coal seam with a fracture extending to the floor of the upper coal seam, with a fracture height of 20 m. This methodology aims to explore and quantify the impact of artificial fracture characteristics on coalbed methane production capacity, to guide the design and optimization of fractures in practice.

[0075] Step 3, build a numerical simulation sample self-generation model

[0076] 1. Automatic generation of numerical simulation files

[0077] Based on the data range of the main control factors selected in the second step, randomly combine to generate simulation schemes. Use the ixf file of the intersect simulator, define functions through Python to assemble keywords, realize automatic filling of fracture length, fracture height, fracture conductivity and fracture spacing, and use AFI files to assemble the design scheme files.

[0078] 2. Automatic batch call of numerical simulator operation and result extraction

[0079] Based on Python, call the Intersect environment in the computer to batch process the automatically generated AFI files, and extract and save the required numerical simulation result files (PRT files) to a new folder for later extraction and combination of simulator results. Based on the numerical simulation result file (PRT file), by marking the coalbed methane geological reserves at the first time node and the coalbed methane geological reserves at the last time node of each numerical simulation result file, define a loop function to automatically extract and format the cumulative gas production of different schemes into a CSV file, realize the whole process automation of simulator result extraction and saving, so as to significantly improve the running efficiency of single model multi-thread processing. This ensures the automation of the whole process from the deployment of design parameters in numerical simulation, parallel computation of multiple schemes to the extraction and saving of simulation results, and realizes the fast and accurate construction process of database.

[0080] Step 4: Establishing a coalbed methane production capacity prediction proxy model based on the XGBoost algorithm

[0081] 1. Prediction sample set division

[0082] The prediction sample set obtained from the self-generation model of the numerical simulation sample is divided into a training set and a validation set according to the relevant proportions of 80% and 20%. The training set is mainly used to train the prediction model, and the validation set is mainly used to adjust the hyperparameters, such as the number of decision trees, the learning rate, the maximum depth of the tree, etc. Each set contains the input and output data of the model, where the input data includes the fracture height, fracture length, fracture conductivity, and fracture spacing, and the output data includes the cumulative gas production of coalbed methane.

[0083] 2. Prediction model establishment

[0084] XGBoost is a machine learning system that uses decision trees as base learners and combines the learning algorithms of linear scale solvers and classification regression trees. The training process of XGBoost is to continuously add trees, and in each process of adding trees, a new tree is used to fit the prediction residual, thereby improving the prediction accuracy. The characteristics of the XGBoost tree model allow us to quantify the importance of each feature and perform feature selection. The basic implementation process of the XGBoost algorithm is as follows:

[0085] (1) Let i represent the ith sample in the training set (X, y)

[0086] (2) Satisfy 1 ≤ t ≤ N k

[0087] ① Construct the cost function:

[0088]

[0089] where g i and h i represent the first and second derivatives of the loss function l at sample i, respectively; y is the true label value of the sample; represents the XGBoost prediction model, x i is the input feature parameter of the decision tree model; r i represents the residual, N k is the number of decision tree models; T represents the number of leaves of the tree model, γ, λ represent the regularization coefficients, and ω j represents the leaf node value.

[0090] ② Use the dataset (X, r) to train a tree model with a maximum depth of d

[0091] ③ Update the prediction value of the XGBoost model

[0092]

[0093] ④ Update the residual term r

[0094]

[0095] (3) Obtain the final prediction model

[0096]

[0097] where ε represents the compression factor or learning rate.

[0098] The process of the XGBoost prediction model is as follows:

[0099] First, by continuously adding trees, the features are classified, and each newly generated tree learns a new function f(x) to gradually grow a complete tree. In this way, each time a tree is added, the residual error of the last prediction can be fitted, and the classification model can be continuously improved to improve the accuracy of the prediction.

[0100] Second, after training k trees, sample score prediction is needed. According to the characteristics of the sample, it is classified into the leaf nodes of each tree and the corresponding scores are obtained from them;

[0101] Third, only the sum of the scores calculated by each tree is the final prediction value corresponding to the sample.

[0102] 3. Model prediction effect evaluation

[0103] The prediction effect of the production capacity prediction model is evaluated and analyzed by using the coefficient of determination (R 2 ), mean square error (MSE).

[0104]

[0105] where y i and are the actual value and the predicted value, respectively, is the average value of the actual value, and N is the size of the actual value in the validation set. When R 2 is larger and MSE is smaller, the error between the predicted value and the actual value of the model is smaller, and the performance of the model is better.

[0106] 4. Model prediction effect evaluation

[0107] Output the fracture parameters, fracture spacing, etc. corresponding to the optimal development scheme.

[0108] Taking a coal mine block in Liupanshui, Guizhou as an example, the coalbed methane field mainly adopts fractured horizontal wells for coalbed methane development. The target coal seam has two layers, 10 coal and 12 coal respectively, the main body of the horizontal well is designed to be located in the 12 coal seam, the coal seam is transformed by hydraulic fracturing, and the two layers of coal are longitudinally communicated through the hydraulic fracture. Simulate which fracture parameter combination can maximize the coalbed methane production capacity resources.

[0109] The specific steps of using the present application to optimize the fracture parameters of the fractured horizontal well are as follows:

[0110] Step one: using Petrel-Re reservoir numerical simulation software, based on the core data and reservoir physical property data of the block as the data basis, using the black oil simulator of Petrel-Re commercial numerical simulation software to establish a complex superimposed coal gas reservoir model. The middle depth of the reservoir is 743m, the formation temperature is 29℃, the reservoir pressure gradient of 10 coal is 0.88MPa / 100m, the reservoir pressure gradient of 12 coal is 0.89MPa / 100m, and the water saturation in the fracture is 99%.

[0111] Based on the natural fracture development of the coal seam, and the matrix permeability and porosity are extremely low, the fracture permeability is much larger than the matrix permeability, and the gas mainly flows in the fracture, so the dual medium model is used to characterize the coal seam. The model adopts Cartesian grid, the grid size is 10*10*0.5m, and the model size is 1500*510*25.5m. The hydraulic fracture is set to expand along the maximum principal stress direction. The local grid refinement technology and equivalent conductivity method are used to accurately simulate the artificial hydraulic fracture.

[0112] Step two: the main control factors of the development effect of the coalbed methane fractured horizontal well are screened by the numerical simulation method, the sensitivity scheme is set, the sensitivity of different parameters is studied, and the main factors affecting the coal seam productivity are obtained. The main factors are fracture spacing, fracture height, fracture conductivity and fracture length.

[0113] Table 1 parameter setting of influencing factors

[0114]

[0115] Step three: based on the constructed fracture parameter numerical simulation sample self-generation model, the coalbed methane production capacity prediction sample set is automatically constructed by calling the intersect numerical simulator based on Python, the input parameters of the model are fracture height, fracture length, fracture conductivity and fracture spacing, and the output data is cumulative gas production.

[0116] Step four, based on the XGBoost algorithm to establish the coalbed methane production capacity prediction proxy model, the prediction sample set obtained from the self-generation model is divided into training set and validation set according to the correlation ratio of 80%, 20%. The total sample generated is 500, and the sample numbers of the training set and the test set are 400 and 100 respectively. The comparison of the prediction results and the actual results on the training set and the validation set is shown in Figure 9 2 respectively. When R 2 is larger, the MSE is smaller, which means that the error between the predicted value and the actual value of the model is smaller, and the performance of the model is better. Based on the optimized production capacity prediction model, the optimal solution of the production capacity is found, so as to determine the global optimal solution of the fracture parameters.

[0117] Table 2 shows the effect of the model on the evaluation index

[0118]

[0119] The example application proves that the method can well solve the coalbed methane fracture parameter optimization problem of the target block, and the scheme obtained by the optimization calculation has good performance in development effect, which shows that the production capacity prediction proxy model based on numerical simulation has high credibility and can well assist the field fracturing decision.

[0120] The present application combines various test results of coal rock cores, uses Petrel-Re numerical simulation software to construct a numerical simulation model considering the characteristics of multi-layer superimposed coal seams such as coalbed methane adsorption-desorption-diffusion, Lagmuir isothermal adsorption and interlayer heterogeneity. At the same time, the artificial fracture is used to construct a fracturing horizontal well model, and the numerical simulation method is used to screen the main control factors affecting the coalbed methane production capacity. At the same time, the present application uses Python to call the numerical simulator to realize the self-generation of numerical simulation samples, and uses the XGboost algorithm to realize the construction of the production capacity prediction proxy model, and then obtains the optimal solution of the fracture parameter combination of the coalbed methane fracturing horizontal well.

[0121] In addition, the terms "first", "second" are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, such as two, three, etc., unless otherwise specifically limited.

[0122] ​In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are contained in at least one embodiment or example of the present application. In the specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0123] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments within the scope of the present application.

Claims

1. A complex coal structure horizontal well fracture parameter matching relationship optimization design method, characterized in that, It comprises the following steps: Step one: Establish a fine numerical simulation model of coal seam with complex structure through Petrel-Re numerical simulator; Step two: Screen the main control factors of the development effect of CBM fracturing horizontal well through numerical simulation method; Step three: Build a self-generating model of numerical simulation sample; Step four: Establish a CBM productivity prediction proxy model based on XGBoost algorithm; The specific process of step one is as follows: A grid system of the basic model is established by using Petrel-Re reservoir numerical simulation software; The fluid model adopts black oil model, considering Lagmuir temperature adsorption and Fick diffusion; The grid number of the conceptual model is 150*51*51, the grid step length in X and Y directions is 10m, the grid step length in Z direction is 0.5m, the model length is 1500m, and the model width is 510m; A total of 8 sub-layers are divided, among which the second layer and the sixth layer are coal seams, and the remaining are rock and soil layers other than coal seams; The input parameters of the model are determined based on coal rock core experiment data and mine site data, the gas-water two-phase is taken as the research object, the double-porosity and double-permeability model is selected, and the Intersect simulator is used for numerical simulation operation; The specific process of step two is as follows: First, analyze the influencing factors of CBM fracturing horizontal well productivity: Including matrix porosity, gas saturation, artificial fracture conductivity, artificial fracture half-length, fracture height, fracture spacing, and horizontal section length; Second, design the data range of the main control factors: Through parameter sensitivity analysis, four key influencing factors are identified, including the half-length of artificial fracture, the distance between fractures, the conductivity of fracture, and the height of fracture; The full-factorial design method involves the combination of all factors at all levels, at least one test is conducted to estimate the main effect of each factor and its interaction; The specific process of step three is as follows: Automatic generation of numerical simulation file: Based on the data range of the main control factors screened in step two, the simulation schemes are randomly combined, the ixf file of the intersect simulator is used, the function is defined by Python to assemble the keywords, the automatic filling of the fracture length, fracture height, fracture conductivity, and fracture spacing is realized, and the AFI file is used to assemble the design scheme into a file; Then, automatic batch calling of numerical simulator operation and result extraction: Based on Python, the Intersect environment in the computer is called to batch process the automatically generated AFI file, and the numerical simulation result file is extracted and saved to a new folder for later extraction and combination of the simulator results; Based on the numerical simulation result file, the cumulative gas production of different schemes is automatically extracted and saved in CSV file by defining a loop function, which realizes the full automation of the simulator result extraction and saving; The specific process of step four includes prediction sample set division, prediction model establishment, and model prediction effect evaluation.

2. The method according to claim 1, wherein the method is characterized in that: The process of dividing the prediction sample set is as follows: the prediction sample set obtained from the self-generation model of the numerical simulation sample is divided into a training set and a validation set according to the related proportions of 80% and 20%; The training set is mainly used for training the prediction model; The validation set is mainly used for adjusting hyperparameters, including the number of decision trees, the learning rate, and the maximum depth of the tree; Each set contains the input and output data of the model, wherein the input data includes the crack height, crack length, crack conductivity, and crack spacing, and the output data includes the cumulative gas production of the coalbed gas.

3. The method according to claim 1, wherein the method is characterized in that: The process of establishing the prediction model is as follows: the XGBoost algorithm is used for modeling, and the implementation process of the XGBoost algorithm is as follows: (1) Let i denotes the i-th sample in the training set (X, y) (2) satisfy 1 < t < N k ; ① Constructing a cost function: where g i and h i denote the first and second derivatives of the loss function l at sample i, respectively; y is the true label value of the sample; denotes the XGBoost prediction model, x i is the input feature parameter of the decision tree model; r i denotes the residual, N k is the number of decision tree models; T denotes the number of leaves of the tree model, γ, λ denote the regularization coefficients, ω j denotes the leaf node value; ii. using the dataset (X, r), training to obtain a tree model with maximum depth d ③ Updating the prediction value of the XGBoost model ④ Updating the residual term r (3) Obtaining the final prediction model Wherein, ε represents the compression factor or learning rate.

4. The method according to claim 3, wherein the method is characterized in that: The process of establishing the XGBoost model includes the following steps: First, by continuously adding trees, feature classification is performed, and each newly generated tree learns a new function f(x) to gradually grow a complete tree; Second, after training k trees, sample score prediction is needed, and according to the characteristics of the sample, it is classified into the leaf nodes of each tree and obtains the corresponding score from it; Third, only the sum of the scores calculated by each tree is the final prediction value corresponding to the sample.

5. The method according to claim 1, wherein, The specific process of evaluating the prediction effect of the model is as follows: The coefficient of determination R 2 The mean square error MSE is used to evaluate and analyze the prediction effect of the productivity prediction model. where y i and are the actual and predicted values, respectively, is the average of the actual values, and N is the size of the actual values in the validation set. When R 2 is larger, the MSE is smaller, indicating that the error between the predicted and actual values is smaller, and the model performs better.

Images