A method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning

Abstract

The present invention relates to a method for predicting the wellbore collapse pressure of a multi-cleat coal seam based on machine learning, which belongs to the field of oil and gas drilling engineering, including: calculating key rock mechanical parameters of wellbore stability through a rock mechanics calculation model and a ground stress calculation model of a target coal seam; determining the cleat development characteristics of the target coal seam based on core observation and imaging logging data; calculating the wellbore stress based on the obtained ground stress, wellbore trajectory parameters, and formation pore pressure, and calculating the collapse pressure based on the Mogi-Coulomb intermediate principal stress law combined with specific cleat development characteristics; selecting 7 parameters as input parameters and selecting a machine learning method to carry out the collapse pressure calculation; establishing a fitting model between the collapse pressure and key influencing factors to achieve rapid prediction of the collapse pressure. The present invention combines machine learning to predict the wellbore collapse pressure of a multi-cleat coal seam, and the research results have important guiding significance for ensuring the stability of the coal seam wellbore and achieving efficient and safe drilling in the coal seam.

Description

Technical Field

[0001] The present invention relates to a method for predicting wellbore collapse pressure of a multi-cleat coal seam based on machine learning, and belongs to the technical field of oil and gas drilling engineering. Background Art

[0002] Coalbed methane is a clean energy source associated with coal and has broad development prospects. Due to the limitations of coalbed methane development technology, the main coalbed methane reservoirs currently developed are shallower than 1500m, while the geological resources of deep coalbed methane (1500-3000m burial depth) are approximately 30.37×10 12 m 3 The reserves are even greater, and will be an important resource foundation for the large-scale development of China's coalbed methane industry in the future. Due to the deep burial depth of deep coal seams, the development of cleats and fractures, high ground temperatures, strong ground stresses, and high formation pore pressure, the drilling process of coalbed methane wells is prone to wellbore instability, even triggering serious drilling accidents.

[0003] Coal is a typical dual-porosity medium, rich in cleat fracture networks, primarily consisting of face and butt cleats. The presence of cleats significantly affects the rock mechanical properties of the coal, even causing it to exhibit more complex failure characteristics after deformation. Existing shale wellbore stability models primarily use a continuum assumption, which is inadequate for analyzing wellbore stability in coal with cleats.

[0004] Wellbore stability analysis models primarily include analytical and numerical simulation methods. Analytical methods are more widely used due to their ease of calculation, but they require numerous input parameters and require iterative calculations of collapse pressure by continuously varying bottomhole pressure parameters. Numerical simulations are widely used for wellbore stability analysis. These methods can more fully and realistically reflect the characteristics of coal and rock cleats, particularly focusing on the impact of discontinuities such as mid- and face cleats on wellbore stability. However, numerical simulations are computationally intensive and complex, making them less suitable for focused studies of specific formations.

[0005] Machine learning studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills. Compared with traditional analytical and numerical simulation methods for wellbore stability analysis, it can quickly predict wellbore instability characteristics while reducing the operator's background knowledge requirements. However, machine learning for coal seam wellbore instability analysis depends on the quality of training sample data. Too low sample data quality will cause training and generalization errors, thereby reducing prediction accuracy.

[0006] Therefore, how to reduce the occurrence of drilling accidents induced by CBM well wall instability is an important issue that needs to be solved urgently in the safe and efficient development of CBM. Summary of the Invention

[0007] In order to address the shortcomings of the existing technology, the present invention provides a method for predicting the wellbore collapse pressure of multi-cleat coal seams based on machine learning. It uses geomechanics and the multi-cleat coal seam collapse pressure model to obtain the key between collapse pressure and key data, and further uses a machine learning method to train and test the input parameters to obtain the collapse pressure value of the formation, further providing basic support for the on-site coal seam drilling mud density window setting.

[0008] The present invention adopts the following technical solutions:

[0009] A method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning includes the following steps:

[0010] S1: Collect on-site logging, mud logging, and drilling data from coalbed methane wells. Calculate key rock mechanical parameters for wellbore stability using a rock mechanics calculation model and a geostress calculation model for the target coal seam (both models are derived from existing mechanical theory models combined with logging data). Key rock mechanical parameters include elastic modulus and Poisson's ratio. Geostress, including vertical principal stress, horizontal minimum principal stress, and horizontal maximum principal stress, is calculated.

[0011] S2: Determine the main cleat development characteristics of the target coal seam based on core observation and imaging logging data;

[0012] S3: Calculate the wellbore stress based on the obtained in-situ stress, wellbore trajectory parameters (including well inclination angle, well inclination azimuth), and formation pore pressure. Based on the Mogi-Coulomb intermediate principal stress law and the specific cleat development characteristics, calculate the collapse pressure p w , obtain the sample space database of collapse pressure calculation of the target block;

[0013] The wellbore instability was constrained and verified using the on-site wellbore expansion rate parameters. The average wellbore expansion rate parameter for all wells in the block was kept below 30% to meet quality requirements.

[0014] S4: Select the seven parameters of well inclination angle, well inclination azimuth, cleat 1 inclination, cleat 1 dip, cleat 2 inclination, cleat 2 dip, and drilling fluid density as input parameters, select the machine learning method to carry out machine learning modeling, and use the collapse pressure calculation result as the only output parameter;

[0015] S5: Establish a fitting model between collapse pressure and key influencing factors to achieve rapid prediction of collapse pressure.

[0016] This method uses model-driven and machine learning to predict the wellbore collapse pressure of multi-cleat coal seams. The research results have important guiding significance for ensuring the stability of the coal seam wellbore and achieving efficient and safe drilling in coal seams.

[0017] Preferably, in step S1, the relationship between the dynamic elastic modulus and the dynamic Poisson's ratio is calculated based on the longitudinal wave velocity and the shear wave velocity (the longitudinal wave velocity and the shear wave velocity can be directly obtained by collecting on-site well logging data):

[0018]

[0019]

[0020] Where: E d is the dynamic elastic modulus, MPa; μ d is the dynamic Poisson's ratio, dimensionless; ρ is the formation density, g / cm 3 ; V p is the longitudinal wave velocity, km / s; V s is the shear wave velocity, km / s;

[0021] The calculation results of formula (1) (2) are the mechanical parameters under dynamic conditions during drilling operations, that is, dynamic mechanical parameters, which need to be converted into static mechanical parameters before they can be applied to the calculation of geomechanical models. The static elastic modulus E of the rock under static conditions can be obtained by conducting triaxial compression tests on rock mechanics using on-site coring. s and the static Poisson's ratio μ s , linear regression analysis method is used to establish linear representation of dynamic parameters and static parameters, and further combined with the ground stress model to carry out ground stress calculation. The calculation formula is as follows:

[0022]

[0023]

[0024]

[0025] Where σ H is the maximum horizontal principal stress, MPa; σ h is the horizontal minimum principal stress, MPa; σ z is the vertical principal stress, MPa; ρ is the formation bulk density, g / cm 3 ; h is the thickness of the formation, m; g is the acceleration of gravity, m / s 2 ; H is the depth, m; is the construction correction amount, E s is the static elastic modulus, GPa; μ s is the static Poisson's ratio; α is the effective stress coefficient; P p is the formation pore pressure.

[0026] Preferably, in step S2, through analysis methods such as SEM scanning electron microscopy or CT scanning, key parameters such as the degree of development of cleats and cracks, the number and connectivity of face cleats and end cleats, and the extension length of cleats can be intuitively observed. Well imaging (FMI) can provide high-resolution images of coal seams to intuitively identify the existence and distribution of cracks and cleats.

[0027] A cleat refers to a natural fracture system in a coal seam that is perpendicular to both the coal face and the surface. The longer, more continuous fracture is called a face cleat, while the less continuous fracture that terminates perpendicular to the face cleat is called an end cleat. Scanning electron microscopy images show that cleats are well-developed in the coal rock in this area, with face and end cleats intersecting each other. Consequently, the presence of these cleats can affect later coal rock failure, leading to tensile or shear failure of the bedding.

[0028] Preferably, the specific implementation process of step S3 is:

[0029] S31: Calculate the stress components around the well:

[0030]

[0031] Where: σ xx,i is the normal stress in the X direction of the X plane caused by the ground stress, MPa; σ yy,i is the normal stress in the Y direction of the Y surface caused by the ground stress, MPa; σ zz,i is the normal stress in the Z direction of the Z surface caused by the ground stress, MPa; τ xy,i is the shear stress in the Y direction of the X plane caused by the ground stress, MPa; τ yz,i is the shear stress in the Z direction of the Y surface caused by the ground stress, MPa; τ xz,i is the shear stress in the Z direction of the X plane caused by the ground stress, MPa; σ H is the maximum horizontal principal stress, MPa; σ h is the horizontal minimum principal stress, MPa; σ z is the vertical principal stress, MPa; ψ is the well inclination angle, °; Ω is the well inclination azimuth, °;

[0032] S32: Calculate the stress around the well caused by anisotropy:

[0033]

[0034] Where: σ xx,a is the normal stress in the X direction of the X plane caused by anisotropy, MPa; σ yy,a is the normal stress in the Y direction of the Y plane caused by anisotropy, MPa; σ zz,a is the normal stress in the Z direction of the Z plane caused by anisotropy, MPa; τ xy,ais the shear stress in the X-Y direction caused by anisotropy, MPa; τ yz,a is the shear stress in the Z direction of the Y plane caused by anisotropy, MPa; τ xz,a is the shear stress in the Z direction of the X plane caused by anisotropy, MPa;

[0035] B ij is the i×j element of the flexibility matrix, as shown in formula (8):

[0036]

[0037] Where: E is the elastic modulus in the isotropic surface, GPa; ν is the Poisson's ratio in the isotropic surface; E' is the elastic modulus in the normal direction of the isotropic surface, GPa; ν' is the Poisson's ratio in the normal direction of the isotropic surface; G' is the shear modulus of the plane perpendicular to the isotropic surface, GPa;

[0038] Re is the real part of an imaginary number; ξ i , χ i and φ i ′(z i ) is calculated as shown in equations (9)-(14), ξ i is the root of equation (9), equation (9) has 6 roots, and the two are conjugate, ξ i are the three roots whose imaginary parts are positive (i=1, 2, 3);

[0039]

[0040]

[0041]

[0042]

[0043] χ i Determined by equations (10) to (12), the expressions of φ1′(z1), φ2′(z2), and φ3′(z3) are shown in equation (13), where D′, E′, F′, and G k (k=1,2,3) is obtained from formula (14),

[0044]

[0045]

[0046] Where p w is the drilling fluid column pressure, MPa; θ is the wellbore angle, °;

[0047] Based on the above calculations and the principle of stress superposition, the wellbore stress component caused by geostress and the wellbore stress caused by anisotropy can be superimposed to obtain the wellbore stress, as shown in formula (15):

[0048]

[0049] Where: σ xx is the normal stress in the X direction of the X surface around the well, MPa; σ yy,a is the normal stress in the Y direction of the Y surface around each well, MPa; σ zz,a is the normal stress in the Z direction of the wellbore Z surface, MPa; τ xy,a is the shear stress in the X- and Y-direction around the well, MPa; τ yz,a is the shear stress in the Z direction of the Y plane around the well, MPa; τ xz,a is the shear stress in the Z direction of the X-plane around the well, MPa;

[0050] S33: Calculate wellbore stress:

[0051] Simplifying formula (15), we can get the wellbore stress in cylindrical coordinates:

[0052]

[0053] Where: σ rr is the radial stress, MPa; σ θθ is the circumferential stress, MPa; σ zz is the axial stress, MPa; τ rθ is the shear stress in the θ direction of the r plane, MPa; τ θz is the shear stress in the z direction of the θ plane, MPa; τ rz is the shear stress in the z direction of the r plane, MPa;

[0054] S34: Calculate principal stresses:

[0055]

[0056] Where σ i , σ j , σ k Represents three principal stresses. By sorting, the maximum value is the maximum principal stress σ1, the middle value is the middle principal stress σ2, and the minimum value is the minimum principal stress σ3;

[0057] S35: Mogi-Coulomb intermediate principal stress law:

[0058] Because there are a lot of cleats in coal seams, the coal rock is divided into two media: the main matrix and the cleats. The main matrix rock strength criterion uses the Mogi-Coulomb intermediate principal stress law to determine whether the rock has failed. Its expression is as follows:

[0059] τ oct =a+bσ m,2 (18)

[0060]

[0061]

[0062]

[0063]

[0064] Where σ1 is the maximum principal stress, σ2 is the intermediate principal stress, and σ3 is the minimum principal stress; m,2 is the effective mean stress, τ oct is the octahedral shear stress, c is the matrix cohesion, and φ is the matrix internal friction angle;

[0065] S36: Weak plane strength criterion:

[0066] Coal seams have a large number of cleats, which are defined as weak planes. Combined with the specific cleat development characteristics, mainly including the cohesion, internal friction angle, dip and inclination of the cleats, and considering that coal rock itself is a discontinuous rock, the Jaeger-Cook weak plane criterion is used to calculate the failure characteristics of coal rock in wellbore instability:

[0067]

[0068]

[0069] in,

[0070]

[0071] Where: c0 is the cohesion of the rock body; φ0 is the internal friction angle of the rock body; c bp,i is the cohesion of the i-th weak plane; φ bp,i is the internal friction angle of the i-th weak surface; β i is the angle between the loading stress of the i-th weak plane and the normal direction of the bedding plane; β i,1 and β i,2 is the bedding dip range of the i-th crack surface failure;

[0072] Combine formulas (6) to (25) and use p w To solve the unknown variables, MATLAB programming can be used to calculate p under different working conditions. w The value is the collapse pressure under this working condition. According to the actual drilling conditions of the target block, the parameters in the model, such as the well inclination angle ψ and the well inclination azimuth Ω, are adjusted to obtain a large number of collapse pressure values under different drilling conditions, and then obtain a sample space database for collapse pressure calculation of the target block.

[0073] Preferably, in step S4, data cleaning is carried out, and 7 parameters are selected as input parameters. The machine learning method is implemented by a machine learning module encapsulated in MATLAB. It is only necessary to input multiple parameter data into it and set the objective function to obtain the corresponding calculation results, wherein the basic input parameters include matrix elastic modulus (GPa), matrix Poisson's ratio, matrix internal friction angle (°), matrix cohesion (MPa), cleat 1 internal friction angle (°), cleat 2 internal friction angle (°), cleat 1 cohesion (MPa), cleat 2 cohesion (MPa), initial porosity, initial permeability (md), formation pore pressure (MPa), vertical principal stress (MPa), horizontal maximum principal stress (MPa) and horizontal minimum principal stress (MPa);

[0074] The input parameters of machine learning have a certain range and certain changes, while the basic input parameters are fixed values.

[0075] Among them, the matrix elastic modulus (GPa), i.e., the dynamic elastic modulus, and the matrix Poisson's ratio, i.e., the dynamic Poisson's ratio, are obtained by formulas (1) and (2), respectively; the matrix internal friction angle (°) and matrix cohesion (MPa) can be obtained through triaxial rock mechanics experiments; the cleat 1 internal friction angle (°), cleat 2 internal friction angle (°), cleat 1 cohesion (MPa), and cleat 2 cohesion (MPa) can be obtained through rock shear experiments; the initial porosity is obtained through rock porosity testing; the initial permeability is obtained through rock permeability experiments; the vertical principal stress (MPa), the horizontal maximum principal stress (MPa), and the horizontal minimum principal stress (MPa) are calculated by formulas (3)-(5);

[0076] The formation pore pressure is obtained from sonic logging data. The Eaton method is commonly used. The calculation process is as follows:

[0077] Acoustic transit time measures the propagation time of elastic waves in the formation. It primarily reflects formation lithology, compaction, and porosity. Except for gas-bearing zones where acoustic transit time may show high values or frequency skips, it is much less affected by changes in wellbore diameter, temperature, and formation water salinity than other logging methods. Using acoustic transit time is a relatively effective method for evaluating and calculating formation pore pressure.

[0078] In normally compacted formations, the relationship between acoustic transit time and well depth is as follows:

[0079] Δt=Δt0e cH (26)

[0080] Transforming the above formula we can get:

[0081] logΔt=AH+B (27)

[0082] Where Δt represents the formation acoustic wave time difference at depth H, μs / ft; Δt0 represents the acoustic wave time difference at depth 0, μs / ft; A, B, and c are coefficients, where A < 0, c < 0;

[0083] Formula (27) is the normal trend line of acoustic transit time. Log(Δt) is linearly related to H, with a slope of A (A<0). On the semi-logarithmic curve, the logarithmic value of acoustic transit time in a normally compacted formation decreases linearly with depth. If abnormally high pressure occurs, the acoustic transit time scatter points will significantly deviate from the normal trend line. When reading the acoustic transit time value, do not select a reduced diameter section or a wellbore with an excessively large diameter to avoid errors.

[0084] Eaton method calculation formula is as follows:

[0085] G p =G op -(G op -ρ w )(Δt n / Δt) n (28)

[0086] Where: G p is the equivalent density of the formation pore pressure at the well depth H, g / cm 3 ; G op is the pressure equivalent density of the overburden at the well depth H, g / cm 3 ρ w is the formation water density at the well depth H, g / cm 3 ;Δt n is the acoustic time difference value during normal compaction at the well depth H, μs / ft; Δt is the measured acoustic time difference value at the well depth H, μs / ft; n is the Eaton index;

[0087] A sample space database was established through numerical simulation, in which the area with a wellbore expansion rate greater than 30% was selected as the verification of the collapse pressure, and the limit value of wellbore collapse was analyzed.

[0088] Preferably, in step S4, the machine learning method includes but is not limited to support vector machine, KNN, adaptive boosting, bagging, multivariate linear regression, random forest algorithm, etc. Taking the random forest algorithm as an example, the algorithm mainly uses the bagging algorithm to construct n decision subtrees to generate a forest. At the same time, when classifying a new test sample, the results of all subtrees are summarized into a classification result, and the classification result with the most votes is selected as the output result of the random forest calculation, where the average value of all trees is the predicted value of the random forest. Among them, 80% of the simulated data is selected as the sample data, and 20% of the data is the analysis test data.

[0089] It is further preferred that the collapse pressure value is calculated through multiple sets of machine learning models, and the prediction stability and generalization performance are verified by calculating the error (80% of the simulation data are selected as sample data, and 20% of the data are analytical test data. The collapse pressure value is output by machine learning using the 80% simulation data as sample data, and compared with the 20% analytical test data, where the 20% data as analytical test data are regarded as accurate values, and the predicted collapse pressure is regarded as the calculated value. The error is the size of the deviation of the calculated value from the accurate value). The smaller the error, the more accurate the calculation, and the best machine learning algorithm suitable for target coal seam wellbore instability analysis is preferred.

[0090] Preferably, in step S5, by adjusting a single different input parameter or a combination of multiple input parameters, the influence of the output collapse pressure change amplitude is compared to determine the key influencing factors of the collapse pressure.

[0091] Where the present invention is not exhaustive, please refer to the prior art.

[0092] The beneficial effects of the present invention are:

[0093] The present invention uses a multi-weak plane cleat model to calculate the collapse pressure of coalbed methane wells to obtain a sample database. The coalbed collapse pressure calculation model takes into account the existence of multiple groups of cleat weak planes and the intermediate principal stress criterion to calculate the collapse pressure and scientifically evaluate the collapse pressure value.

[0094] The present invention utilizes a machine learning method to calculate the collapse pressure of a multi-cleat coal seam, which has the advantages of accurate calculation and fast solution speed, and is convenient for on-site personnel to perform rapid collapse pressure analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] The drawings in the specification, which constitute a part of this application, are used to provide further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute improper limitations on this application.

[0096] Figure 1 is a flow chart of the present invention;

[0097] Figure 2 The collapse pressure prediction analysis results for matrix instability, single cleat instability, and multiple cleat instability provided in one embodiment of the present invention, where (a) is complete coal matrix instability, (b) is single cleat instability, and (c) is dual cleat instability.

[0098] Figure 3The prediction analysis results of multi-cleat collapse pressure provided by an embodiment of the present invention, wherein (a) is a dip of 0° and a dip of 0°; (b) is a dip of 0° and a dip of 30°; (c) is a dip of 0° and a dip of 60°; (d) is a dip of 0° and a dip of 90°; (e) is a dip of 45° and a dip of 0°; (f) is a dip of 45° and a dip of 30°; (g) is a dip of 45° and a dip of 60°; (h) is a dip of 45° and a dip of 90°; (i) is a dip of 90° and a dip of 0°; (j) is a dip of 90° and a dip of 30°; (k) is a dip of 90° and a dip of 60°; (l) is a dip of 90° and a dip of 90°;

[0099] Figure 4 A comparison chart of the wellbore expansion rate using the random forest algorithm provided by an embodiment of the present invention;

[0100] Figure 5 A comparison chart of numerical calculation results and machine learning results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0101] In order to enable people in this technical field to better understand the technical solutions in this specification, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the drawings in the implementation of this specification, but are not limited to this. Anything not fully described in the present invention shall be based on the conventional technology in this field.

[0102] Example 1

[0103] A method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning includes the following steps:

[0104] S1: Collect on-site logging, mud logging, and drilling data from coalbed methane wells. Calculate key rock mechanical parameters for wellbore stability using a rock mechanics calculation model and a geostress calculation model for the target coal seam (both models are derived from existing mechanical theory models combined with logging data). Key rock mechanical parameters include elastic modulus and Poisson's ratio. Geostress, including vertical principal stress, horizontal minimum principal stress, and horizontal maximum principal stress, is calculated.

[0105] The relationship between the dynamic elastic modulus and the dynamic Poisson's ratio is calculated based on the P-wave velocity and S-wave velocity (the P-wave velocity and S-wave velocity can be directly obtained through field logging data collection):

[0106]

[0107]

[0108] Where: E d is the dynamic elastic modulus, MPa; μ d is the dynamic Poisson's ratio, dimensionless; ρ is the formation density, g / cm 3; V p is the longitudinal wave velocity, km / s; V s is the shear wave velocity, km / s;

[0109] The calculation results of formula (1) (2) are the mechanical parameters under dynamic conditions during drilling operations, that is, dynamic mechanical parameters, which need to be converted into static mechanical parameters before they can be applied to the calculation of geomechanical models. The static elastic modulus E of the rock under static conditions can be obtained by conducting triaxial compression tests on rock mechanics using on-site coring. s and the static Poisson's ratio μ s , linear regression analysis method is used to establish linear representation of dynamic parameters and static parameters, and further combined with the ground stress model to carry out ground stress calculation. The calculation formula is as follows:

[0110]

[0111]

[0112]

[0113] Where σ H is the maximum horizontal principal stress, MPa; σ h is the horizontal minimum principal stress, MPa; σ z is the vertical principal stress, MPa; ρ is the formation bulk density, g / cm 3 ; h is the thickness of the formation, m; g is the acceleration of gravity, m / s 2 ; H is the depth, m; is the construction correction amount, E s is the static elastic modulus, GPa; μ s is the static Poisson's ratio; α is the effective stress coefficient; P p is the formation pore pressure.

[0114] The calculation results of this embodiment are shown in Table 1:

[0115] Table 1 Calculation results of vertical principal stress, horizontal minimum principal stress and horizontal maximum principal stress

[0116]

[0117] S2: Determine the main cleat development characteristics of the target coal seam based on core observation and imaging logging data;

[0118] Through analytical methods such as SEM scanning electron microscopy or CT scanning, key parameters such as the degree of development of cleats and cracks, the number and connectivity of face cleats and end cleats, and the extension length of cleats can be intuitively observed. Well imaging logging (FMI) can provide high-resolution images of coal seams to intuitively identify the existence and distribution of cracks and cleats.

[0119] A cleat refers to a natural fracture system in a coal seam that is perpendicular to both the coal face and the surface. The longer, more continuous fracture is called a face cleat, while the less continuous fracture that terminates perpendicular to the face cleat is called an end cleat. Scanning electron microscopy images show that cleats are well-developed in the coal rock in this area, with face and end cleats intersecting each other. Consequently, the presence of these cleats can affect later coal rock failure, leading to tensile or shear failure of the bedding.

[0120] S3: Calculate the wellbore stress based on the obtained in-situ stress, wellbore trajectory parameters (including well inclination angle, well inclination azimuth), and formation pore pressure. Based on the Mogi-Coulomb intermediate principal stress law and the specific cleat development characteristics, calculate the collapse pressure p w , obtain the sample space database of collapse pressure calculation of the target block;

[0121] The specific implementation process is:

[0122] S31: Calculate the stress components around the well:

[0123]

[0124] Where: σ xx,i is the normal stress in the X direction of the X plane caused by the ground stress, MPa; σ yy,i is the normal stress in the Y direction of the Y surface caused by the ground stress, MPa; σ zz,i is the normal stress in the Z direction of the Z surface caused by the ground stress, MPa; τ xy,i is the shear stress in the Y direction of the X plane caused by the ground stress, MPa; τ yz,i is the shear stress in the Z direction of the Y surface caused by the ground stress, MPa; τ xz,i is the shear stress in the Z direction of the X plane caused by the ground stress, MPa; σ H is the maximum horizontal principal stress, MPa; σ h is the horizontal minimum principal stress, MPa; σ z is the vertical principal stress, MPa; ψ is the well inclination angle, °; Ω is the well inclination azimuth, °;

[0125] S32: Calculate the stress around the well caused by anisotropy:

[0126]

[0127] Where: σ xx,a is the normal stress in the X direction of the X plane caused by anisotropy, MPa; σ yy,a is the normal stress in the Y direction of the Y plane caused by anisotropy, MPa; σ zz,a is the normal stress in the Z direction of the Z plane caused by anisotropy, MPa; τ xy,ais the shear stress in the X-Y direction caused by anisotropy, MPa; τ yz,a is the shear stress in the Z direction of the Y plane caused by anisotropy, MPa; τ xz,a is the shear stress in the Z direction of the X plane caused by anisotropy, MPa;

[0128] B ij is the i×j element of the flexibility matrix, as shown in formula (8):

[0129]

[0130] Where: E is the elastic modulus in the isotropic surface, GPa; ν is the Poisson's ratio in the isotropic surface; E' is the elastic modulus in the normal direction of the isotropic surface, GPa; ν' is the Poisson's ratio in the normal direction of the isotropic surface; G' is the shear modulus of the plane perpendicular to the isotropic surface, GPa;

[0131] Re is the real part of an imaginary number; ξ i , χ i and φ i ′(z i ) is calculated as shown in equations (9)-(14), ξ i is the root of equation (9), equation (9) has 6 roots, and the two are conjugate, ξ i are the three roots whose imaginary parts are positive (i=1, 2, 3);

[0132]

[0133]

[0134]

[0135]

[0136] χ i Determined by equations (10) to (12), the expressions of φ1′(z1), φ2′(z2), and φ3′(z3) are shown in equation (13), where D′, E′, F′, and G k (k=1,2,3) is obtained from formula (14),

[0137]

[0138]

[0139] Where p w is the drilling fluid column pressure, MPa; θ is the wellbore angle, °;

[0140] Based on the above calculations and the principle of stress superposition, the wellbore stress component caused by geostress and the wellbore stress caused by anisotropy can be superimposed to obtain the wellbore stress, as shown in formula (15):

[0141]

[0142] Where: σ xx is the normal stress in the X direction of the X surface around the well, MPa; σ yy,a is the normal stress in the Y direction of the Y surface around each well, MPa; σ zz,a is the normal stress in the Z direction of the wellbore Z surface, MPa; τ xy,a is the shear stress in the X- and Y-direction around the well, MPa; τ yz,a is the shear stress in the Z direction of the Y plane around the well, MPa; τ xz,a is the shear stress in the X-axis and Z-axis directions around the well, MPa;

[0143] S33: Calculate wellbore stress:

[0144] Simplifying formula (15), we can get the wellbore stress in cylindrical coordinates:

[0145]

[0146] Where: σ rr is the radial stress, MPa; σ θθ is the circumferential stress, MPa; σ zz is the axial stress, MPa; τ rθ is the shear stress in the θ direction of the r plane, MPa; τ θz is the shear stress in the z direction of the θ plane, MPa; τ rz is the shear stress in the z direction of the r plane, MPa;

[0147] S34: Calculate principal stresses:

[0148]

[0149] Where σ i , σ j , σ k Represents three principal stresses. By sorting, the maximum value is the maximum principal stress σ1, the middle value is the middle principal stress σ2, and the minimum value is the minimum principal stress σ3;

[0150] S35: Mogi-Coulomb intermediate principal stress law:

[0151] Because there are a lot of cleats in coal seams, the coal rock is divided into two media: the main matrix and the cleats. The main matrix rock strength criterion uses the Mogi-Coulomb intermediate principal stress law to determine whether the rock has failed. Its expression is as follows:

[0152] τ oct =a+bσ m,2 (18)

[0153]

[0154]

[0155]

[0156]

[0157] Where σ1 is the maximum principal stress, σ2 is the intermediate principal stress, and σ3 is the minimum principal stress; m,2 is the effective mean stress, τ oct is the octahedral shear stress, c is the matrix cohesion, and φ is the matrix internal friction angle;

[0158] S36: Weak plane strength criterion:

[0159] Coal seams have a large number of cleats, which are defined as weak planes. Combined with the specific cleat development characteristics, mainly including the cohesion, internal friction angle, dip and inclination of the cleats, and considering that coal rock itself is a discontinuous rock, the Jaeger-Cook weak plane criterion is used to calculate the failure characteristics of coal rock in wellbore instability:

[0160]

[0161]

[0162] in,

[0163]

[0164] Where: c0 is the cohesion of the rock body; φ0 is the internal friction angle of the rock body; c bp,i is the cohesion of the i-th weak plane; φ bp,i is the internal friction angle of the i-th weak surface; β i is the angle between the loading stress of the i-th weak plane and the normal direction of the bedding plane; β i,1 and β i,2 is the bedding dip range of the i-th crack surface failure;

[0165] Combine formulas (6) to (25) and use p w To solve the unknown variables, MATLAB programming can be used to calculate p under different working conditions. w The value is the collapse pressure under this working condition. According to the actual drilling conditions of the target block, the parameters in the model, such as the well inclination angle ψ and the well inclination azimuth Ω, are adjusted to obtain a large number of collapse pressure values under different drilling conditions, and then obtain a sample space database for collapse pressure calculation of the target block.

[0166] The wellbore instability was constrained and verified using the on-site wellbore expansion rate parameters. The average wellbore expansion rate parameter for all wells in the block was kept below 30% to meet quality requirements.

[0167] Table 2 Collapse stress and average wellbore expansion rate

[0168]

[0169] Through the well diameter expansion rate of the study block, it can be found that the average well diameter expansion rate of all wells in the study block is kept below 25%, and the statistical average value is 9.03%, which meets the quality requirements.

[0170] S4: Select the seven parameters of well inclination angle, well inclination azimuth, cleat 1 inclination, cleat 1 dip, cleat 2 inclination, cleat 2 dip, and drilling fluid density as input parameters, select the machine learning method to carry out machine learning modeling, and use the collapse pressure calculation result as the only output parameter;

[0171] Data cleaning was carried out, and seven parameters were selected as input parameters. The machine learning method was implemented through the machine learning module encapsulated in MATLAB. It only needs to input multiple parameter data into it and set the objective function to obtain the corresponding calculation results. The basic input parameters include matrix elastic modulus (GPa), matrix Poisson's ratio, matrix internal friction angle (°), matrix cohesion (MPa), cleat 1 internal friction angle (°), cleat 2 internal friction angle (°), cleat 1 cohesion (MPa), cleat 2 cohesion (MPa), initial porosity, initial permeability (md), formation pore pressure (MPa), vertical principal stress (MPa), horizontal maximum principal stress (MPa), and horizontal minimum principal stress (MPa).

[0172] The input parameters of machine learning have a certain range and certain changes, while the basic input parameters are fixed values.

[0173] Among them, the matrix elastic modulus (GPa), i.e., the dynamic elastic modulus, and the matrix Poisson's ratio, i.e., the dynamic Poisson's ratio, are obtained by formulas (1) and (2), respectively; the matrix internal friction angle (°) and matrix cohesion (MPa) can be obtained through triaxial rock mechanics experiments; the cleat 1 internal friction angle (°), cleat 2 internal friction angle (°), cleat 1 cohesion (MPa), and cleat 2 cohesion (MPa) can be obtained through rock shear experiments; the initial porosity is obtained through rock porosity testing; the initial permeability is obtained through rock permeability experiments; the vertical principal stress (MPa), the horizontal maximum principal stress (MPa), and the horizontal minimum principal stress (MPa) are calculated by formulas (3)-(5);

[0174] The formation pore pressure is obtained from sonic logging data. The Eaton method is commonly used. The calculation process is as follows:

[0175] Acoustic transit time measures the propagation time of elastic waves in the formation. It primarily reflects formation lithology, compaction, and porosity. Except for gas-bearing zones where acoustic transit time may show high values or frequency skips, it is much less affected by changes in wellbore diameter, temperature, and formation water salinity than other logging methods. Using acoustic transit time is a relatively effective method for evaluating and calculating formation pore pressure.

[0176] In normally compacted formations, the relationship between acoustic transit time and well depth is as follows:

[0177] Δt=Δt0e cH (26)

[0178] Transforming the above formula we can get:

[0179] logΔt=AH+B (27)

[0180] Where Δt represents the formation acoustic wave time difference at depth H, μs / ft; Δt0 represents the acoustic wave time difference at depth 0, μs / ft; A, B, and c are coefficients, where A < 0, c < 0;

[0181] Formula (27) is the normal trend line of acoustic transit time. Log(Δt) is linearly related to H, with a slope of A (A<0). On the semi-logarithmic curve, the logarithmic value of acoustic transit time in a normally compacted formation decreases linearly with depth. If abnormally high pressure occurs, the acoustic transit time scatter points will significantly deviate from the normal trend line. When reading the acoustic transit time value, do not select a reduced diameter section or a wellbore with an excessively large diameter to avoid errors.

[0182] Eaton method calculation formula is as follows:

[0183] G p =G op -(G op -ρ w )(Δt n / Δt) n (28)

[0184] Where: G p is the equivalent density of the formation pore pressure at the well depth H, g / cm 3 ; G op is the pressure equivalent density of the overburden at the well depth H, g / cm 3 ρ w is the formation water density at the well depth H, g / cm 3 ;Δt n is the acoustic time difference value during normal compaction at the well depth H, μs / ft; Δt is the measured acoustic time difference value at the well depth H, μs / ft; n is the Eaton index;

[0185] A sample space database was established through numerical simulation, in which the area with a wellbore expansion rate greater than 30% was selected as the verification of the collapse pressure, and the limit value of wellbore collapse was analyzed.

[0186] In this embodiment, the machine learning method uses the random forest algorithm. This algorithm primarily uses a bagging algorithm to construct n decision subtrees to generate a forest. When classifying a new test sample, the results of all subtrees are combined into a single classification result. The classification result with the most votes is selected as the output of the random forest calculation, with the average of all trees being the random forest's predicted value. 80% of the simulated data is used as sample data, and 20% of the data is used for analytical testing.

[0187] This embodiment selects the random forest algorithm and uses the grid search algorithm to configure the optimal parameters of the random forest. The exhaustive search method of the grid search algorithm is used to select the maximum depth d as 10 and the number of regression trees m as 10 to obtain the final optimal model. The test set data is used to verify the effect diagram of the predicted formation collapse pressure and the actual pressure.

[0188] S5: Establish a fitting model between collapse pressure and key influencing factors to achieve rapid prediction of collapse pressure;

[0189] By adjusting a single different input parameter or a combination of multiple input parameters and comparing the impact of the output collapse pressure change amplitude, the key influencing factors of the collapse pressure are determined.

[0190] By predicting the formation collapse pressure of multi-cleat coalbed methane wells, the lower limit of the drilling fluid density window can be determined (usually the collapse pressure is regarded as the lower limit of the drilling fluid density window). The calculated collapse pressure data is compared with the mud density in the well history data. If the mud density is within the drilling fluid density window, it means that the risk of accidents in this well section is not great, otherwise it is not. The results are then compared with the on-site conditions.

[0191] For example, at a depth of 869 m in a coalbed methane well in the Ordos Basin, the predicted collapse pressure is 0.725 g / cm 3 According to the well history data, the mud density value here is 1.0g / cm 3 , which is within the density window of the CBM drilling fluid, so the probability of risk occurrence is low;

[0192] The present invention was applied to a coalbed methane well in a certain block of the Ordos Basin. Through the wellbore expansion rate, it was found that varying degrees of wellbore wall collapse occurred in the Shanxi Formation to the Taiyuan Formation in the area where the study well was located. The wellbore expansion rate was relatively large, and the prediction accuracy reached 90.15%.

[0193] like Figure 2 Shown are the prediction analysis results of collapse pressure for matrix instability, single cleat instability, and multiple cleat instability, in MPa. Figure 3 is the collapse stress analysis result of multiple cleats;

[0194] Figure 4 This is a comparison chart of the well diameter expansion rate obtained by the random forest algorithm, with the horizontal axis being the well number;

[0195] like Figure 5 As shown, the collapse stresses obtained using the machine learning algorithm in this embodiment are not significantly different from those calculated using conventional techniques. However, the algorithm of the present invention simplifies the calculation process and achieves a faster solution speed, enabling rapid prediction. This method utilizes model-driven methods combined with machine learning to predict the wellbore collapse pressure in multi-cleat coal seams. The research results have important guiding significance for ensuring wellbore stability and achieving efficient and safe drilling in coal seams.

[0196] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning, characterized in that: The steps include: S1: Collect on-site logging, mud logging, and drilling data from coalbed methane wells. Calculate key rock mechanical parameters for wellbore stability using the rock mechanical calculation model and geostress calculation model for the target coal seam. Key rock mechanical parameters include elastic modulus and Poisson's ratio. Geostress, including vertical principal stress, horizontal minimum principal stress, and horizontal maximum principal stress, is calculated. S2: Determine the cleat development characteristics of the target coal seam based on core observation and imaging logging data; S3: Calculate the stress around the wellbore based on the obtained ground stress, wellbore trajectory parameters, and formation pore pressure. Calculate the collapse pressure p based on the Mogi-Coulomb intermediate principal stress law and the specific cleat development characteristics. w , obtain the target block collapse pressure calculation sample space database; The wellbore instability was constrained and verified using the on-site wellbore expansion rate parameters. The average wellbore expansion rate parameter for all wells in the block was kept below 30% to meet quality requirements. S4: Select the seven parameters of well inclination angle, well inclination azimuth, cleat 1 inclination, cleat 1 dip, cleat 2 inclination, cleat 2 dip, and drilling fluid density as input parameters, select the machine learning method to carry out machine learning modeling, and use the collapse pressure calculation result as the only output parameter; S5: Establish a fitting model between collapse pressure and key influencing factors to achieve rapid prediction of collapse pressure; In step S1, the relationship between the dynamic elastic modulus and the dynamic Poisson's ratio is calculated based on the longitudinal wave velocity and the shear wave velocity: Where: E d is the dynamic elastic modulus, MPa; μ d is the dynamic Poisson's ratio, dimensionless; ρ is the formation density, g / cm 3 ; V p is the longitudinal wave velocity, km / s; V s is the shear wave velocity, km / s; The static elastic modulus E of the rock under static conditions can be obtained by conducting triaxial compression tests on rock mechanics using on-site coring. s and the static Poisson's ratio μ s , linear regression analysis method is used to establish linear representation of dynamic parameters and static parameters, and further combined with the ground stress model to carry out ground stress calculation. The calculation formula is as follows: Where σ H is the maximum horizontal principal stress, MPa; σ h is the horizontal minimum principal stress, MPa; σ z is the vertical principal stress, MPa; ρ is the formation bulk density, g / cm 3 ; h is the thickness of the formation, m; g is the acceleration of gravity, m / s 2 ; H is the depth, m; ζ1, ζ2 are the structural corrections, E s is the static elastic modulus, GPa; μ s is the static Poisson's ratio; α is the effective stress coefficient; P p is the formation pore pressure.

2. The method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning according to claim 1, characterized in that: In step S2, the degree of development of cleats and cracks, the number and connectivity of face cleats and end cleats, and the parameters of cleat extension length can be visually observed through SEM scanning electron microscopy or CT scanning analysis. Imaging logging provides high-resolution images of the coal seam, which can visually identify the existence and distribution of cracks and cleats.

3. The method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning according to claim 2, characterized in that: The specific implementation process of step S3 is: S31: Calculate the stress components around the well: Where: σ xx,i is the normal stress in the X direction of the X plane caused by the ground stress, MPa; σ yy,i is the normal stress in the Y direction of the Y surface caused by the ground stress, MPa; σ zz,i is the normal stress in the Z direction of the Z surface caused by the ground stress, MPa; τ xy,i is the shear stress in the Y direction of the X plane caused by the ground stress, MPa; τ yz,i is the shear stress in the Z direction of the Y surface caused by the ground stress, MPa; τ xz,i is the shear stress in the Z direction of the X plane caused by the ground stress, MPa; σ H is the maximum horizontal principal stress, MPa; σ h is the horizontal minimum principal stress, MPa; σ z is the vertical principal stress, MPa; ψ is the well inclination angle, °; Ω is the well inclination azimuth, °; S32: Calculate the stress around the well caused by anisotropy: Where: σ xx,a is the normal stress in the X direction of the X plane caused by anisotropy, MPa; σ yy,a is the normal stress in the Y direction of the Y plane caused by anisotropy, MPa; σ zz,a is the normal stress in the Z direction of the Z plane caused by anisotropy, MPa; τ xy,a is the shear stress in the X-Y direction caused by anisotropy, MPa; τ yz,a is the shear stress in the Z direction of the Y plane caused by anisotropy, MPa; τ xz,a is the shear stress in the Z direction of the X plane caused by anisotropy, MPa; B ij is the i×j element of the flexibility matrix, as shown in formula (8): Where: E is the elastic modulus in the isotropic surface, GPa; ν is the Poisson's ratio in the isotropic surface; E' is the elastic modulus in the normal direction of the isotropic surface, GPa; ν' is the Poisson's ratio in the normal direction of the isotropic surface; G' is the shear modulus of the plane perpendicular to the isotropic surface, GPa; Re is the real part of the imaginary number; ξ i , χ i and φ′ i (z i ) is calculated as shown in equations (9)-(14), ξ i is the root of equation (9). Equation (9) has 6 roots, and they are conjugated to each other. i The three roots whose imaginary parts are positive when i=1, 2, 3; χ i Determined by equations (10) to (12), the expressions of φ′1(z1), φ′2(z2), and φ′3(z3) are shown in equation (13), where D′, E′, F′, and G k From formula (14), we get k = 1, 2, 3; Where p w is the drilling fluid column pressure, MPa; θ is the wellbore angle, °; Based on the above calculations, the wellbore stress component caused by the ground stress and the wellbore stress caused by anisotropy are superimposed to obtain the wellbore stress, as shown in formula (15): Where: σ xx is the normal stress in the X direction of the X surface around the well, MPa; σ yy,a is the normal stress in the Y direction of the Y surface around each well, MPa; σ zz,a is the normal stress in the Z direction of the wellbore Z surface, MPa; τ xy,a is the shear stress in the X- and Y-direction around the well, MPa; τ yz,a is the shear stress in the Z direction of the Y plane around the well, MPa; τ xz,a is the shear stress in the Z direction of the X-plane around the well, MPa; S33: Calculate wellbore stress: Simplifying formula (15), we can get the wellbore stress in cylindrical coordinates: Where: σ rr is the radial stress, MPa; σ θθ is the circumferential stress, MPa; σ zz is the axial stress, MPa; τ rθ is the shear stress in the θ direction of the r plane, MPa; τ θz is the shear stress in the z direction of the θ plane, MPa; τ rz is the shear stress in the z direction of the r plane, MPa; S34: Calculate principal stresses: Where σ i , σ j , σ k Represents three principal stresses. By sorting, the maximum value is the maximum principal stress σ1, the middle value is the middle principal stress σ2, and the minimum value is the minimum principal stress σ3; S35: Mogi-Coulomb intermediate principal stress law: Because there are a lot of cleats in coal seams, the coal rock is divided into two media: the main matrix and the cleats. The main matrix rock strength criterion uses the Mogi-Coulomb intermediate principal stress law to determine whether the rock has failed. Its expression is as follows: t oct =a+bσ m,2 (18) Where σ1 is the maximum principal stress, σ2 is the intermediate principal stress, and σ3 is the minimum principal stress; m,2 is the effective mean stress, τ oct is the octahedral shear stress, c is the matrix cohesion, and φ is the matrix internal friction angle; S36: Weak plane strength criterion: Coal seams have a large number of cleats, which are defined as weak planes. Combined with the cleat development characteristics, including the cohesion, internal friction angle, dip, and inclination of the cleats, and considering that coal rock itself is a discontinuous rock, the Jaeger-Cook weak plane criterion is used to calculate the failure characteristics of coal rock in wellbore instability: Where: c0 is the cohesion of the rock body; φ0 is the internal friction angle of the rock body; c bp,i is the cohesion of the i-th weak plane; φ bp,i is the internal friction angle of the i-th weak surface; β i is the angle between the loading stress of the i-th weak plane and the normal direction of the bedding plane; β i,1 and β i,2 is the bedding dip range of the i-th crack surface failure; Combine formulas (6) to (25) and use p w To solve the unknown variables, MATLAB programming can be used to calculate p under different working conditions. w The value is the collapse pressure under this working condition. By adjusting the parameters in the model according to the actual drilling conditions of the target block, a large number of collapse pressure values under different drilling conditions can be obtained, and then a sample space database for collapse pressure calculation of the target block can be obtained.

4. The method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning according to claim 3, characterized in that: In step S4, data cleaning is carried out, and seven parameters are selected as input parameters. The machine learning method is implemented through a machine learning module encapsulated in MATLAB, where the basic input parameters include matrix elastic modulus, matrix Poisson's ratio, matrix internal friction angle, matrix cohesion, cleat 1 internal friction angle, cleat 2 internal friction angle, cleat 1 cohesion, cleat 2 cohesion, initial porosity, initial permeability, formation pore pressure, vertical principal stress, horizontal maximum principal stress, and horizontal minimum principal stress; Among them, the matrix elastic modulus is the dynamic elastic modulus, and the matrix Poisson's ratio is the dynamic Poisson's ratio, which are obtained by formulas (1) and (2) respectively; the matrix internal friction angle and matrix cohesion can be obtained through rock triaxial mechanics experiments; the cleat 1 internal friction angle, cleat 2 internal friction angle, cleat 1 cohesion, and cleat 2 cohesion can be obtained through rock shear experiments; the initial porosity is obtained through rock porosity testing; the initial permeability is obtained through rock permeability experiments; the vertical principal stress, horizontal maximum principal stress, and horizontal minimum principal stress are calculated using formulas (3)-(5); The formation pore pressure is obtained through sonic logging data, and the calculation process is as follows: In normally compacted formations, the relationship between acoustic transit time and well depth is as follows: Transforming the above formula we can get: logΔt=AH+B m (27) Where Δt represents the formation acoustic wave time difference at depth H, μs / ft; Δt0 represents the acoustic wave time difference at depth 0, μs / ft; A, B m 、c m is the coefficient, where A<0, c m <0; Formula (27) is the normal trend line of acoustic wave transit time. Log(Δt) is linearly related to H, with a slope of A. On the semi-logarithmic curve, the logarithmic value of acoustic wave transit time in a normally compacted formation decreases linearly with depth. Eaton method calculation formula is as follows: Where: G p is the equivalent density of the formation pore pressure at the well depth H, g / cm 3 ; G op is the pressure equivalent density of the overburden at the well depth H, g / cm 3 ρ w is the formation water density at the well depth H, g / cm 3 ;Δt n is the acoustic time difference value during normal compaction at the well depth H, μs / ft; Δt is the measured acoustic time difference value at the well depth H, μs / ft; and n is the Eaton index.

5. The method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning according to claim 4, characterized in that: In step S4, machine learning methods include but are not limited to support vector machines, KNN, adaptive boosting, bagging, multivariate linear regression, and random forest algorithms. 80% of the simulated data are selected as sample data, and 20% of the data are used as analysis test data.

6. The method for predicting wellbore collapse pressure in multi-cleat coal seams based on machine learning according to claim 5, characterized in that: In step S5, by adjusting a single different input parameter or a combination of multiple input parameters, the influence of the output collapse pressure change range is compared to determine the key influencing factors of the collapse pressure.

Images