Coal seam gas content dynamic prediction method based on Bayesian optimization XGBoost

Abstract

The invention discloses a coal seam gas content dynamic prediction method based on Bayesian optimization XGBoost, and belongs to the field of coal mine gas control, and the method comprises the steps: collecting a sample set U composed of influence factors of coal seam gas occurrence, preprocessing the sample set to obtain a sample set, and dividing the sample set in proportion to obtain a training set, a verification set and a test set; dividing the training set into a plurality of sample sets; an XGBoost model is constructed, the nth sample set serves as input, the actual coal seam gas content serves as output, the XGBoost model is trained, and an nth coal seam gas content prediction model is obtained; the influence factors of coal seam gas occurrence are calculated through a three-dimensional geologic model, geologic structure characteristics and coal seam roof and floor lithologic characteristics are comprehensively considered, and a Bayesian optimization algorithm is adopted to carry out adaptive search on hyper-parameters of an XGBoost model. The technical problem that a traditional gas content prediction method mostly depends on static geological parameters and cannot achieve real-time prediction of the gas occurrence state under the complex geological condition is solved.

Description

Technical Field

[0001] This invention belongs to the field of coal mine gas control and relates to a dynamic construction method for coal seam gas content attribute models, specifically a dynamic prediction method for coal seam gas content based on Bayesian optimization XGBoost. Background Technology In underground coal mining, accurate prediction of coal seam gas content is one of the core technical challenges in preventing coal and gas outbursts and ensuring safe mine production. With increasing mining depth and intensity, the occurrence of coal seam gas is significantly influenced by the combined effects of geological structure distribution and environmental parameters. Traditional gas content prediction models largely rely on static geological parameters, resulting in the following drawbacks: First, the rigidity of model hyperparameters leads to insufficient dynamic adaptability to gas emission characteristics under complex mining conditions; second, the nonlinear correlations between multi-source heterogeneous data are not fully integrated, resulting in insufficient generalization ability of the prediction model; and third, obtaining relevant geological parameters is difficult, leading to long update cycles for the prediction model.

[0002] Therefore, there is an urgent need to construct a coal seam gas content prediction method that integrates multi-source dynamic data driving and has adaptive parameter optimization capabilities, so as to realize real-time modeling of gas occurrence state under complex geological conditions. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a dynamic prediction method for coal seam gas content based on Bayesian optimization XGBoost, thereby solving the technical problem that traditional gas content prediction methods rely heavily on static geological parameters and cannot achieve real-time prediction of gas occurrence under complex geological conditions.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A dynamic prediction method for coal seam gas content based on Bayesian optimization XGBoost, using a three-dimensional geological model including a fault plane model and a stratigraphic model, includes the following steps: Step 1: Collect a sample set U consisting of influencing factors of coal seam gas occurrence, and preprocess it to obtain the sample set. The dataset is then divided proportionally to obtain a training set, a validation set, and a test set; the training set is further divided into multiple sample sets. ; The influencing factors of coal seam gas occurrence include geological structural features and the characteristics of the coal seam roof and floor. The geological structural features include fault distribution and fold distribution; The characteristics of the top and bottom plates of the coal seam include the thickness of the coal seam, the thickness of the overlying bedrock, the burial depth of the coal seam, the sand-mudstone ratio of the coal seam to the top interface of the Yan'an Formation, the mudstone thickness of the top and bottom plates of the coal seam, and the mudstone thickness within 20m of the top plate of the coal seam. Step 2: Construct the XGBoost model using the nth sample set. Using the actual coal seam gas content as the input and the actual coal seam gas content as the output, the XGBoost model is trained to obtain the nth coal seam gas content prediction model; among them, the Bayesian method is used for optimization. Step 3: Input the validation set into the nth coal seam gas content prediction model to obtain the predicted coal seam gas content. Use K-cross-validation to verify the error rate between the predicted coal seam gas content and the actual coal seam gas content corresponding to the validation set. Input the test set into the nth coal seam gas content prediction model to obtain the predicted coal seam gas content. Calculate the coefficient of determination using the predicted coal seam gas content and the actual coal seam gas content corresponding to the test set. ; Step 4: Select the coefficient of determination with the smallest error rate. The coal seam gas content prediction model that is closest to 1 is used as the final coal seam gas content prediction model. Step 5: Perform grid difference on the coal seam space to obtain the spatial location set of the coal seam space. ; Step 6: Use a three-dimensional geological model to calculate the set of spatial locations. Spatial position Factors influencing coalbed methane occurrence; The influencing factors of coal seam gas occurrence include geological structural features and the characteristics of the coal seam roof and floor. The geological structural features include fault distribution and fold distribution; The characteristics of the top and bottom plates of the coal seam include the thickness of the coal seam, the thickness of the overlying bedrock, the burial depth of the coal seam, the sand-mudstone ratio of the coal seam to the top interface of the Yan'an Formation, the mudstone thickness of the top and bottom plates of the coal seam, and the mudstone thickness within 20m of the top plate of the coal seam. Step 7: Combine the spatial location set obtained in Step 6. Spatial position The influencing factors of coal seam gas occurrence are input into the final coal seam gas content prediction model to obtain spatial location. The content of coal seam gas; Step 8: Measure the properties of the coal and rock strata in real time during the mining process, update the modeling parameters of the three-dimensional geological model based on the properties of the coal and rock strata, and return to Step 6.

[0005] This invention also includes the following technical features: Between steps seven and eight, the method of harmonic averaging is also used to determine the spatial location. The coal seam gas content was processed to obtain a coal seam gas occurrence attribute model.

[0006] Step six specifically includes the following steps: Step 6.1: Calculate the coal seam dip angle according to formula (6-1). And use it as a folding parameter; (6-1) in: Indicates spatial location The tangent vector at the point; Indicates spatial location The magnitude of the normal vector at that point; Step 6.2: Use formula (6-2) to calculate the coal seam thickness, overlying bedrock thickness, coal seam burial depth, mudstone thickness of the top and bottom plates of the coal seam, sandstone thickness of the top and bottom plates of the coal seam, and mudstone thickness within 20m of the top plate of the coal seam. (6-2) in: Indicates geological layers, The value range is 1 to 6, which respectively represent the coal seam, the overlying bedrock, the coal seam burial depth, the mudstone on the top and bottom of the coal seam, the sandstone on the top and bottom of the coal seam, and the mudstone within 20m of the top of the coal seam. Indicates spatial location geological strata The thickness; Indicates spatial location geological strata The corresponding equation for the top plate surface; Indicates spatial location geological strata The corresponding equation for the base plate surface; Step 6.3: Calculate the ratio of the thickness of mudstone at the top and bottom of the coal seam to the thickness of sandstone at the top and bottom of the coal seam obtained in Step 6.2, which gives the sandstone-mudstone ratio at the top interface of the coal seam to the Yan'an Formation. Step 6.4: Calculate the fault parameters according to formula (6-3). This yields the fault distribution; (6-3) in: Faults in coal seam space The difference in elevation; Indicates spatial location Faults in the coal seam space The shortest distance between them.

[0007] In step five, the mesh subdivision method is the Delaney tetrahedral meshing method.

[0008] In step one, the preprocessing specifically involves: using the maximum-minimum normalization method, mapping the influencing factors of gas occurrence in each coal seam to the [0, 1] interval through linear transformation to obtain the sample set. .

[0009] Compared with the prior art, the beneficial technical effects of this invention are: (I) In this invention, the influencing factors of coal seam gas occurrence are calculated by a three-dimensional geological model. Taking into account the geological structure characteristics and the lithological characteristics of the roof and floor of the coal seam, the hyperparameters of the XGBoost model are adaptively searched using the Bayesian optimization algorithm to achieve dynamic prediction of coal seam gas content. This solves the technical problem that traditional gas content prediction methods rely heavily on static geological parameters and cannot achieve real-time prediction of gas occurrence under complex geological conditions.

[0010] (II) Based on prediction, this invention constructs a coal seam gas content attribute model. At the same time, by continuously updating the modeling parameters of the three-dimensional geological model and iteratively updating the three-dimensional geological model, the influencing factors affecting coal seam gas occurrence are calculated, thus realizing the continuous updating of the coal seam gas content attribute model. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the coal seam space after grid differentiation according to the present invention; Figure 2 This is a comparison chart of the predicted and actual values ​​of the final coal seam gas content prediction model of this invention; Figure 3 This is a schematic diagram of the coal seam gas content attribute model of the present invention.

[0012] The specific content of the present invention will be further explained in detail below with reference to the embodiments. Detailed Implementation

[0013] It should be noted that, unless otherwise specified, all components in this invention are those known in the art.

[0014] The following are specific embodiments of the present invention. It should be noted that the present invention is not limited to the following specific embodiments. All equivalent modifications made based on the technical solutions of this application fall within the protection scope of the present invention.

[0015] This invention presents a dynamic prediction method for coal seam gas content based on Bayesian optimization XGBoost. Based on a three-dimensional geological model, which includes a fault plane model and a stratigraphic model, the method comprises the following steps: Step 1: Collect a sample set U consisting of influencing factors of coal seam gas occurrence, and preprocess it to obtain the sample set. The dataset is then divided proportionally to obtain a training set, a validation set, and a test set; the training set is further divided into multiple sample sets. ; Factors influencing coal seam gas occurrence include geological structural features and the characteristics of the coal seam roof and floor; Geological structural features include the distribution of faults and folds; The characteristics of the top and bottom plates of the coal seam include the thickness of the coal seam, the thickness of the overlying bedrock, the burial depth of the coal seam, the sand-mudstone ratio of the coal seam to the top interface of the Yan'an Formation, the mudstone thickness of the top and bottom plates of the coal seam, and the mudstone thickness within 20m of the top plate of the coal seam. Step 2: Construct the XGBoost model using the nth sample set. Using the actual coal seam gas content as the input and the actual coal seam gas content as the output, the XGBoost model is trained to obtain the nth coal seam gas content prediction model; among them, the Bayesian method is used for optimization. Step 3: Input the validation set into the nth coal seam gas content prediction model to obtain the predicted coal seam gas content. Use K-cross-validation to verify the error rate between the predicted coal seam gas content and the actual coal seam gas content corresponding to the validation set. Input the test set into the nth coal seam gas content prediction model to obtain the predicted coal seam gas content. Calculate the coefficient of determination using the predicted coal seam gas content and the actual coal seam gas content corresponding to the test set. ; Step 4: Select the coefficient of determination with the smallest error rate. The coal seam gas content prediction model that is closest to 1 is used as the final coal seam gas content prediction model. Step 5: Perform grid difference on the coal seam space to obtain the spatial location set of the coal seam space. ; Step 6: Use a three-dimensional geological model to calculate the set of spatial locations. Spatial position Factors influencing coalbed methane occurrence; Factors influencing coal seam gas occurrence include geological structural features and the characteristics of the coal seam roof and floor; Geological structural features include the distribution of faults and folds; The characteristics of the top and bottom plates of the coal seam include the thickness of the coal seam, the thickness of the overlying bedrock, the burial depth of the coal seam, the sand-mudstone ratio of the coal seam to the top interface of the Yan'an Formation, the mudstone thickness of the top and bottom plates of the coal seam, and the mudstone thickness within 20m of the top plate of the coal seam. Step 7: Combine the spatial location set obtained in Step 6. Spatial position The influencing factors of coal seam gas occurrence are input into the final coal seam gas content prediction model to obtain spatial location. The content of coal seam gas; Step 8: Measure the properties of the coal and rock strata in real time during the mining process, update the modeling parameters of the three-dimensional geological model based on the properties of the coal and rock strata, and return to Step 6.

[0016] In the above scheme, the influencing factors of coal seam gas occurrence are calculated by a three-dimensional geological model. Taking into account the geological structural characteristics and the lithological characteristics of the roof and floor of the coal seam, the hyperparameters of the XGBoost model are adaptively searched using a Bayesian optimization algorithm to achieve dynamic prediction of coal seam gas content. This solves the technical problem that traditional gas content prediction methods rely heavily on static geological parameters and cannot achieve real-time prediction of gas occurrence under complex geological conditions.

[0017] Ideally, the ratio of training set, validation set, and test set should be 80%, 10%, and 10%, respectively. The training set is used for training and learning feature parameters; the validation set is mainly used to tune the hyperparameters of the prediction model during training to confirm the optimal model; the test set is used for independent evaluation of the final model performance to verify the model's performance on unseen data; and K-cross-validation is used to improve the robustness and reliability of the evaluation results.

[0018] In this scheme, the XGBoost model based on the gradient boosting decision tree framework is adopted, which effectively constructs the complex mapping relationship between variables and avoids the possibility of highly nonlinear relationships between variables.

[0019] The spatial location calculated in this scheme Table 1 shows the influencing factors of coal seam gas occurrence and the corresponding actual coal seam gas content; the obtained spatial locations The influencing factors of coal seam gas occurrence are input into the final coal seam gas content prediction model to obtain spatial location. The predicted value of coal seam gas content, in this spatial location The comparison chart of predicted and actual coal seam gas content is shown below. Figure 2 As shown.

[0020] Between steps seven and eight, the method of harmonic averaging is also used to determine the spatial location. The coal seam gas content was processed to obtain a coal seam gas occurrence attribute model.

[0021] In the above scheme, a coal seam gas content attribute model is constructed based on prediction. At the same time, by continuously updating the modeling parameters of the three-dimensional geological model and iteratively updating the three-dimensional geological model, the influencing factors affecting coal seam gas occurrence are calculated, thus realizing the continuous updating of the coal seam gas content attribute model.

[0022] The constructed coal seam gas content attribute model is as follows Figure 3 As shown.

[0023] Step six specifically includes the following steps: Step 6.1: Calculate the coal seam dip angle according to formula (6-1). And use it as a folding parameter; (6-1) in: Indicates spatial location The tangent vector at the point; Indicates spatial location The magnitude of the normal vector at that point; Step 6.2: Use formula (6-2) to calculate the coal seam thickness, overlying bedrock thickness, coal seam burial depth, mudstone thickness of the top and bottom plates of the coal seam, sandstone thickness of the top and bottom plates of the coal seam, and mudstone thickness within 20m of the top plate of the coal seam. (6-2) in: Indicates geological layers, The value range is 1 to 6, which respectively represent the coal seam, the overlying bedrock, the coal seam burial depth, the mudstone on the top and bottom of the coal seam, the sandstone on the top and bottom of the coal seam, and the mudstone within 20m of the top of the coal seam. Indicates spatial location geological strata The thickness; Indicates spatial location geological strata The corresponding equation for the top plate surface; Indicates spatial location geological strata The corresponding equation for the base plate surface; Step 6.3: Calculate the ratio of the thickness of mudstone at the top and bottom of the coal seam to the thickness of sandstone at the top and bottom of the coal seam obtained in Step 6.2, which gives the sandstone-mudstone ratio at the top interface of the coal seam to the Yan'an Formation. Step 6.4: Calculate the fault parameters according to formula (6-3). This yields the fault distribution; (6-3) in: Faults in coal seam space The difference in elevation; Indicates spatial location Faults in the coal seam space The shortest distance between them; Among them, the fault plane model is used to calculate spatial location. Faults in the coal seam space The shortest distance between them; the stratigraphic model is used to derive the virtual borehole, and the spatial location is obtained by fitting the interface of each geological layer in the virtual borehole. geological strata The corresponding equations for the top and bottom surfaces.

[0024] In step five, the mesh subdivision method is the Delaney tetrahedral meshing method.

[0025] In the above technical solution, the coal seam space is divided into tetrahedral meshes after mesh difference as follows: Figure 3 As shown.

[0026] In step one, the preprocessing specifically involves: using the maximum-minimum normalization method, mapping the influencing factors of gas occurrence in each coal seam to the [0, 1] interval through linear transformation to obtain the sample set. .

[0027] In the above technical solution, dimensional differences of all quantitative characteristics are eliminated through preprocessing.

[0028] Table 1

Claims

1. A coal seam gas content dynamic prediction method based on Bayesian optimization XGBoost, based on a three-dimensional geological model, the three-dimensional geological model comprising a fault plane model and a stratum model; characterized in that, The method comprises the following steps: Step one, a sample set U of the influencing factors of coal seam gas occurrence is collected and preprocessed to obtain a sample set , and is divided in proportion to obtain a training set, a validation set and a test set; the training set is divided into multiple sample sets . The influencing factors of the coal seam gas occurrence include geological structure features and coal seam roof and floor features; The geological structure features include fault distribution and fold distribution; The coal seam roof and floor features include coal seam thickness, overlying bedrock thickness, coal seam buried depth, sandstone-mudstone ratio from the coal seam to the top interface of Yan'an Formation, mudstone thickness of the coal seam roof and floor, and mudstone thickness within 20m of the coal seam roof; Step two, build the XGBoost model to the nth sample set The XGBoost model is trained with the actual coal seam gas content as input and output, and the nth coal seam gas content prediction model is obtained; wherein the Bayesian method is used for optimization; Step three, input the verification set into the nth coal seam gas content prediction model to obtain the predicted coal seam gas content, and use the K cross-validation method to verify the error rate of the predicted coal seam gas content and the actual coal seam gas content corresponding to the verification set; input the test set into the nth coal seam gas content prediction model to obtain the predicted coal seam gas content, and use the predicted coal seam gas content and the actual coal seam gas content corresponding to the test set to calculate the determination coefficient ; Step four, select the minimum error rate and the coefficient of determination The coal seam gas content prediction model corresponding to the closest 1 is taken as the final coal seam gas content prediction model. Step five, grid difference is made to the coal seam space to obtain a spatial position set of the coal seam space ; Step six, using the three-dimensional geological model to calculate the spatial position set of the coal seam gas occurrence influencing factors of the spatial position The influencing factors of the coal seam gas occurrence include geological structure features and coal seam roof and floor features; The geological structure features include fault distribution and fold distribution; The coal seam roof and floor features include coal seam thickness, overlying bedrock thickness, coal seam buried depth, sandstone-mudstone ratio from the coal seam to the top interface of Yan'an Formation, mudstone thickness of the coal seam roof and floor, and mudstone thickness within 20m of the coal seam roof; Step seven, input the spatial position set obtained in step six and the influencing factors of coal seam gas occurrence in the spatial position into the final coal seam gas content prediction model to obtain the coal seam gas content in the spatial position . Step eight, real-time measuring the coal and rock stratum properties in the mining process, updating the modeling parameters of the three-dimensional geological model according to the coal and rock stratum properties, and returning to step six.

2. The coal seam gas content dynamic prediction method based on Bayesian optimization XGBoost of claim 1, characterized in that, Between step seven and step eight, there is also: using harmonic mean method to process the spatial position of coal seam gas content, and obtain the coal seam gas occurrence attribute model. of coal seam gas content, and obtain the coal seam gas occurrence attribute model.

3. The coal seam gas content dynamic prediction method based on Bayesian optimization XGBoost of claim 1, characterized in that, Step six specifically comprises the following steps: Step 6.1 Calculate the coal seam dip angle according to formula (6-1) and take it as the fold parameter; （6-1） Wherein: tangent vector representing a spatial position tangent vector representing a spatial position a length of a normal vector representing a spatial position at a point; Step 6.2, calculating the coal seam thickness, overlying bedrock thickness, coal seam buried depth, mudstone thickness of the coal seam roof and floor, sandstone thickness of the coal seam roof and floor, and mudstone thickness within 20m of the coal seam roof by using formula (6-2) respectively; （6-2） Wherein: representing a geological layer, The value range is 1~6, which respectively corresponds to representing coal seam, overlying bedrock, coal seam buried depth, coal seam roof and floor mudstone, coal seam roof and floor sandstone and coal seam roof 20m range mudstone; representing a spatial location of a geological layer at a location; representing a spatial location geologic layer at the location corresponding roof surface equation; representing a spatial location geologic layer at the location corresponding floor surface equation; Step 6.3, calculating the ratio of the mudstone thickness of the coal seam roof and floor to the sandstone thickness of the coal seam roof and floor obtained in step 6.2, that is, obtaining the sandstone-mudstone ratio from the coal seam to the top interface of Yan'an Formation; Step 6.

4. Calculate fault parameters according to equation (6-3) i.e. obtain fault distribution; （6-3） Wherein: representing a fault in the space of the coal seam gap of the fault; representing spatial locations the shortest distance between the coal seam and the fault in space. the shortest distance between the coal seam and the fault in space.

4. The coal seam gas content dynamic prediction method based on Bayesian optimization XGBoost of claim 1, characterized in that, In step five, the grid difference method is Delaney tetrahedral mesh division method.

5. The coal seam gas content dynamic prediction method based on Bayesian optimization XGBoost of claim 1, characterized in that, In step one, the pretreatment is specifically: using the maximum and minimum value normalization method, the influencing factors of each coal seam gas occurrence are mapped to the [0, 1] interval through linear transformation to obtain the sample set .

Images