CN-121544836-B - Dynamic prediction method for coal bed gas content based on Bayesian optimization XGBoost

CN121544836BCN 121544836 BCN121544836 BCN 121544836BCN-121544836-B

Abstract

The invention discloses a dynamic prediction method for coal-bed gas content based on Bayesian optimization XGBoost, which belongs to the field of coal-mine gas control and comprises the steps of collecting and obtaining a sample set U consisting of influence factors of occurrence of coal-bed gas, and preprocessing the sample set U to obtain the sample set And divide it proportionally to obtain the training set, validation set, and testing set; Split the training set into multiple sample sets Constructing XGBoost model with nth sample set According to the method, the actual coal seam gas content is taken as output, a XGBoost model is trained to obtain an nth coal seam gas content prediction model, influence factors of coal seam gas occurrence are calculated through a three-dimensional geological model, geological structure characteristics and coal seam roof and floor lithology characteristics are comprehensively considered, a Bayesian optimization algorithm is adopted to conduct self-adaptive search on super parameters of the XGBoost model, and the technical problem that the traditional gas content prediction method is not capable of achieving real-time prediction of gas occurrence states under complex geological conditions depending on static geological parameters is solved.

Inventors

YANG HUI
YAN JUNSHENG
JI KANG
LIU ZAIBIN
LI PENG
FAN TAO
Ju Chaohui

Assignees

西安煤科透明地质科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260115

Claims (5)

1. A dynamic prediction method for gas content in a coal bed based on Bayesian optimization XGBoost is based on a three-dimensional geological model, wherein the three-dimensional geological model comprises a fault plane model and a stratum model, and is characterized by comprising the following steps: Step one, collecting a sample set U formed by influence factors of occurrence of coal bed gas, and preprocessing the sample set U to obtain a sample set Dividing the training set, the verification set and the test set according to the proportion to obtain the training set, the verification set and the test set; dividing a training set into a plurality of sample sets ; The influencing factors of the occurrence of the coal seam gas comprise geological structure characteristics and coal seam roof-floor characteristics; the geologic structural features include fault distribution and fold distribution; The characteristics of the top and bottom plates of the coal bed comprise the thickness of the coal bed, the thickness of overlying bedrock, the burial depth of the coal bed, the sand-shale ratio of the top interface of the coal bed to the Yan's set, the mudstone thickness of the top and bottom plates of the coal bed and the mudstone thickness of the top plate of the coal bed within 20 m; step two, constructing XGBoost models to form an nth sample set Training the XGBoost model by taking the actual coal seam gas content as output to obtain an nth coal seam gas content prediction model, wherein the model is optimized by adopting a Bayesian method; Inputting a verification set into an nth coal-bed gas content prediction model to obtain predicted coal-bed gas content, verifying the error rate of the actual coal-bed gas content corresponding to the verification set by adopting a K cross verification method, inputting a test set into the nth coal-bed gas content prediction model to obtain predicted coal-bed gas content, and calculating the actual coal-bed gas content corresponding to the test set by adopting the predicted coal-bed gas content and the actual coal-bed gas content to obtain a determination coefficient ; Step four, selecting the coefficient with the smallest error rate The coal bed gas content prediction model closest to 1 is taken as a final coal bed gas content prediction model; fifthly, mesh subdivision is carried out on the coal seam space, and a space position set of the coal seam space is obtained ; Step six, calculating by adopting a three-dimensional geological model to obtain a space position set Spatial position in (a) Influence factors of coalbed methane occurrence; the influencing factors of the occurrence of the coal seam gas comprise geological structure characteristics and coal seam roof-floor characteristics; the geologic structural features include fault distribution and fold distribution; The characteristics of the top and bottom plates of the coal bed comprise the thickness of the coal bed, the thickness of overlying bedrock, the burial depth of the coal bed, the sand-shale ratio of the top interface of the coal bed to the Yan's set, the mudstone thickness of the top and bottom plates of the coal bed and the mudstone thickness of the top plate of the coal bed within 20 m; Seventh, collecting the space position obtained in the sixth step Spatial position in (a) The influence factors of the occurrence of the coalbed methane are input into a final coalbed methane content prediction model to obtain a spatial position Is a coal seam gas content; And step eight, measuring the property of the coal stratum in the stoping process in real time, updating the modeling parameters of the three-dimensional geological model according to the property of the coal stratum, and returning to the step six.
2. The method for dynamically predicting gas content in coal seam based on Bayesian optimization XGBoost as recited in claim 1, wherein between the seventh step and the eighth step, a harmonic averaging method is adopted to perform spatial prediction And (3) processing the coal-bed gas content to obtain a coal-bed gas occurrence attribute model.
3. A method for dynamically predicting gas content in a coal seam based on bayesian optimization XGBoost as recited in claim 1, wherein the sixth step specifically includes the steps of: Step 6.1, calculating the inclination angle of the coal seam according to the formula (6-1) And takes the same as a wrinkling parameter; (6-1) Wherein: Representing spatial position A tangent vector at the position; Representing spatial position The mode length of the normal vector at the position; Step 6.2, respectively calculating the thickness of the coal bed, the thickness of overlying bedrock, the burial depth of the coal bed, the thickness of mudstone on the top and bottom plates of the coal bed, the thickness of sandstone on the top and bottom plates of the coal bed and the thickness of mudstone within the range of 20m of the top plate of the coal bed by adopting a formula (6-2); (6-2) Wherein: The representation of the geological formation is made, The value range is 1-6, and the value ranges are respectively and correspondingly represented by a coal bed, an overlying bedrock, a coal bed buried depth, a coal bed top and bottom plate mudstone, a coal bed top and bottom plate sandstone and a coal bed top plate mudstone within a range of 20 m; Representing spatial position Geological formation at the site Is a thickness of (2); Representing spatial position Geological formation at the site A corresponding top plate curved surface equation; Representing spatial position Geological formation at the site A corresponding bottom plate curved surface equation; Step 6.3, calculating the ratio of the thickness of the mudstone passing through the top and bottom plates of the coal bed to the thickness of the sandstone of the top and bottom plates of the coal bed, which is obtained in the step 6.2, and obtaining the ratio of the sandstone of the top interface of the coal bed to the Yanan group; step 6.4, calculating according to the formula (6-3) to obtain fault parameters Obtaining fault distribution; (6-3) Wherein: representing faults in coal seam space Is a drop of (2); Representing spatial position With faults in coal seam space Shortest distance between them.
4. A method for dynamically predicting gas content in a coal seam based on bayesian optimization XGBoost as recited in claim 1, wherein in step five, the mesh division method is Delaunay tetrahedral mesh division method.
5. The method for dynamically predicting gas content in coal seam based on Bayesian optimization XGBoost as set forth in claim 1, wherein the preprocessing comprises mapping the influence factors of occurrence of gas in each coal seam to the [0, 1] interval by linear transformation to obtain a sample set by adopting a maximum and minimum normalization method 。

Description

Dynamic prediction method for coal bed gas content based on Bayesian optimization XGBoost Technical Field The invention belongs to the field of coal mine gas control, relates to a dynamic construction method of a coal seam gas content attribute model, and particularly relates to a dynamic prediction method of coal seam gas content based on Bayesian optimization XGBoost. Background In the exploitation of an underground coal mine, accurate prediction of the gas content of a coal bed is one of the core technical challenges of preventing and controlling the outburst of coal and gas and guaranteeing the safe production of the mine. With the increase of the mining depth and intensity of the coal mine, the occurrence state of coal seam gas is obviously enhanced by the influence of the composite action of geological structure distribution and environmental parameters. The traditional gas content prediction model method is mostly dependent on static geological parameters, and has the defects that firstly, dynamic adaptability to gas emission characteristics under complex mining conditions is insufficient due to model super-parameter solidification, secondly, nonlinear association among multi-source heterogeneous data is not fully fused, prediction model generalization capability is insufficient, thirdly, the related geological parameters are difficult to obtain, and the prediction model updating period is long. Therefore, it is needed to construct a coal seam gas content prediction method with adaptive parameter optimization capability by fusing multisource dynamic data driving so as to realize real-time modeling of gas occurrence state under complex geological conditions. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide a dynamic prediction method for the gas content of a coal seam based on Bayesian optimization XGBoost, which solves the technical problem that the traditional gas content prediction method is dependent on static geological parameters to realize the real-time prediction of the gas occurrence state under complex geological conditions. In order to solve the technical problems, the invention adopts the following technical scheme: A dynamic prediction method for gas content of a coal bed based on Bayesian optimization XGBoost is based on a three-dimensional geological model, wherein the three-dimensional geological model comprises a fault plane model and a stratum model, and comprises the following steps: Step one, collecting a sample set U formed by influence factors of occurrence of coal bed gas, and preprocessing the sample set U to obtain a sample set Dividing the training set, the verification set and the test set according to the proportion to obtain the training set, the verification set and the test set; dividing a training set into a plurality of sample sets; The influencing factors of the occurrence of the coal seam gas comprise geological structure characteristics and coal seam roof-floor characteristics; the geologic structural features include fault distribution and fold distribution; The characteristics of the top and bottom plates of the coal bed comprise the thickness of the coal bed, the thickness of overlying bedrock, the burial depth of the coal bed, the sand-shale ratio of the top interface of the coal bed to the Yan's set, the mudstone thickness of the top and bottom plates of the coal bed and the mudstone thickness of the top plate of the coal bed within 20 m; step two, constructing XGBoost models to form an nth sample set Training the XGBoost model by taking the actual coal seam gas content as output to obtain an nth coal seam gas content prediction model, wherein the model is optimized by adopting a Bayesian method; Inputting a verification set into an nth coal-bed gas content prediction model to obtain predicted coal-bed gas content, verifying the error rate of the actual coal-bed gas content corresponding to the verification set by adopting a K cross verification method, inputting a test set into the nth coal-bed gas content prediction model to obtain predicted coal-bed gas content, and calculating the actual coal-bed gas content corresponding to the test set by adopting the predicted coal-bed gas content and the actual coal-bed gas content to obtain a determination coefficient ; Step four, selecting the coefficient with the smallest error rateThe coal bed gas content prediction model closest to 1 is taken as a final coal bed gas content prediction model; fifthly, mesh subdivision is carried out on the coal seam space, and a space position set of the coal seam space is obtained ; Step six, calculating by adopting a three-dimensional geological model to obtain a space position setSpatial position in (a)Influence factors of coalbed methane occurrence; the influencing factors of the occurrence of the coal seam gas comprise geological structure characteristics and coal seam roof-floor characteristics; the geologic structural feat