A drilling site arrangement dynamic optimization method based on conditional random field

Abstract

This invention discloses a dynamic optimization method for drilling site layout based on conditional random fields, belonging to the field of civil engineering exploration and design technology. The method first performs initial gridded drilling, generating multiple sets of three-dimensional conditional random fields considering depth trends and spatial variability based on sampled data. Through coupled finite element simulation of construction conditions, a set of high-risk random fields with excessive soil deformation is selected. Then, the similarity of parameters within the fields is calculated, and key areas most effective in reducing model uncertainty are intelligently identified as supplementary drilling target areas. Through a closed-loop feedback of "site layout-simulation-identification-optimization," high-risk scenarios are completely eliminated or their existence is confirmed. This invention achieves a shift from experience-driven to model-driven approaches, significantly improving the scientific rigor of site layout; the information entropy-based supplementary site mechanism maximizes exploration efficiency and economy, providing reliable exploration decision support for engineering construction under complex geological conditions.

Description

Technical Field

[0001] This invention relates to the field of civil engineering survey and design technology, and in particular to a dynamic optimization method for drilling point layout based on conditional random fields. Background Technology

[0002] With the rapid development of urbanization and infrastructure construction in my country, a large number of tunnel, subway, and underground space projects have been launched, with the scale of these projects continuously expanding and geological conditions becoming increasingly complex. In these projects, preliminary geological investigation is a crucial step in ensuring construction safety and the rationality of the design. Drilling point layout design, as the core work of geological investigation, directly relates to the accurate acquisition of geotechnical parameters, the precise identification of construction risks, and the reliability of subsequent design schemes. However, due to the significant spatial variability and uncertainty of the geological structure itself, how to scientifically determine the location of drilling points in an economical and efficient manner has always been a long-standing technical challenge in the field of engineering geological investigation.

[0003] Currently, the drilling site design methods used in engineering practice can be mainly divided into two categories, but both have obvious limitations:

[0004] 1. Site Selection Based on Engineering Experience: This method typically relies on the designer's personal experience or engineering analogies for site selection. Common practices include uniform grid placement or densification in "suspicious anomaly areas" identified based on topography. While simple to implement, this method is essentially qualitative, lacking objective and quantitative scientific evidence. Under complex and variable geological conditions, over-reliance on experience can easily lead to site selection that fails to accurately reflect the actual spatial variations of the strata, potentially overlooking key geological anomalies, creating blind spots in the exploration, and thus posing potential safety hazards to the project.

[0005] 2. Statistical or Numerical Simulation-Based Site Selection Methods: To overcome the shortcomings of empirical methods, another approach attempts to optimize site selection using geostatistical interpolation or sensitivity analysis based on numerical calculations. These methods introduce mathematical tools to some extent, improving the rationality of site selection. However, their inherent limitations are: First, most methods fail to comprehensively and deeply consider the correlation (i.e., spatial randomness) of soil and rock parameters in three-dimensional space, making it difficult to accurately characterize the true distribution characteristics of soil parameters; second, existing methods are usually independent of specific engineering construction conditions, failing to incorporate the effects of construction disturbances (such as excavation and loading) into site selection optimization. Therefore, it is difficult to assess the actual effectiveness of different site selection schemes for risk identification during the construction phase (such as soil deformation prediction), leading to a disconnect between the exploration work and the ultimate goal of engineering safety control.

[0006] In recent years, random field theory has been introduced into the field of geotechnical engineering, providing a powerful mathematical tool for describing the spatial variability and uncertainty of formation parameters, and gradually becoming the foundation for reliability analysis and parameter inversion. Conditional Random Fields (CRF), as a method capable of generating parameter fields that conform to actual statistical distributions under known measurement data (i.e., borehole point constraints), offer new insights for the scientific design of drilling site layout. It can simulate a large number of possible geological models that conform to statistical laws based on limited borehole data, thereby more fully revealing geological uncertainties.

[0007] However, existing research largely focuses on using random fields for parameter inversion or geological modeling, lacking studies that tightly integrate the generation of conditional random fields with drilling site optimization decisions in a closed loop. Furthermore, how to dynamically incorporate construction simulation into site selection design and guide optimization by quantifying the sensitivity of different site selection schemes to risk identification (such as soil deformation control) remains a pressing challenge.

[0008] Therefore, there is an urgent need in this field for an innovative drilling site optimization method that can inherit the advantages of conditional random fields in characterizing spatial variability and further couple with the construction process. Through a dynamic and iterative mechanism, it can automatically identify the areas most critical to engineering safety for supplementary exploration, thereby achieving a fundamental shift from "experience-based site selection" to a "science-oriented, risk-driven" site selection model. This will maximize economic efficiency while ensuring exploration accuracy, and provide technical support for the safety of engineering construction under complex geological conditions. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a dynamic optimization method for drilling site layout based on conditional random fields.

[0010] The objective of this invention is achieved through the following technical solution: a dynamic optimization method for drilling site layout based on conditional random fields, comprising the following steps:

[0011] S1: Set the initial drilling points according to the gridded uniform sampling, and perform drilling sampling to obtain the sampling point test data;

[0012] S2: Based on the test data of the sampling points, calculate the coefficient of variation and mean of the soil parameters, as well as the lateral and longitudinal coefficients of variation, and generate multiple sets of conditional random fields with depth dependence based on the conditional random field theory.

[0013] S3: Introduce construction excavation or building conditions, calculate the soil displacement state under conditional random fields, count the maximum deformation value of the surface soil under each conditional random field, mark the conditional random fields with the maximum deformation value greater than the engineering design value as candidate samples, and form a candidate sample set P.

[0014] S4: Calculate the similarity of local regions among all candidate samples, and mark the regions that meet the similarity requirements as additional sampling points;

[0015] S5: Drill holes at the marked sampling points to be supplemented, conduct supplementary sampling and testing, and then repeat S2 to S4 until the candidate sample set P is empty or the candidate sample set P no longer decreases. At this point, it is considered that there is a geological risk at the sampling point.

[0016] Preferably, step S2 further includes the following steps:

[0017] S21: Denoise and remove outliers from the sampled test data, and organize it into layers according to the stratigraphic boundaries. Different groups are used for different stratigraphic boundaries. In subsequent operations, the different groups are independently statistically analyzed and conditional random fields are generated.

[0018] S22: Test and calculate the soil parameters at the sampling points, perform logarithmic transformation, and take the exponential value from the original data. ,in y These are the soil parameters after logarithmic transformation. The raw data obtained from the test; for different borehole coordinates The original material parameter sequence is ordered by depth. Decompose using the following formula:

[0019] In the formula The first i The horizontal, vertical, and axial coordinates of each borehole sampling point For the first i Material parameters of each borehole sampling point For fitting depth for all material parameters The trend of change, For the error term that cannot be fitted, the following formula is used to fit the original trend term curve:

[0020] In the formula , where is the trend term function of the parameter to be optimized. a , b and c For different coefficients to be optimized; The first sampling depth in the depth direction; This is the last sampling depth in the depth direction; For depth i Actual observed values;

[0021] For the error term that cannot be fitted The observation errors of soil parameters within the same stratum were statistically analyzed using a normal distribution, and a joint correlation matrix was established.

[0022] S23: Based on soil variability indices, according to the grid coordinates of finite element analysis. Generate multiple sets of log-normal random fields of unconditionally stationary soil observation residuals ,in n The number of predicted points; the set of borehole points is denoted as . , m Given the number of sampling points; calculate the covariance matrix between the known sampling points using the following formula:

[0023] In the formula The covariance matrix between sampling points; p , q For the indexes of two different borehole sampling points; The function for calculating covariance; These are the borehole sampling points. p Coordinates in the x, y, and z directions; These are the borehole sampling points. q Coordinates in the x, y, and z directions;

[0024] The covariance matrix between the predicted points and the sampled points is calculated using the following formula, with dimension 1. n × m :

[0025] In the formula i , j These are the indices of the prediction points and the borehole sampling points, respectively. This is the covariance matrix between the sampling points and the prediction points; These are the prediction points. i Coordinates in the x, y, and z directions; These are the borehole sampling points. j Coordinates in the x, y, and z directions;

[0026] S24: Calculate the residual vector d medium elements And based on the following formula, a conditional random field containing deep dependencies is obtained:

[0027] In the formula For the parameters of the random field under different coordinates in the simulation; This represents the fitted trend term as depth increases; These are the borehole sampling points. i Coordinates in the x, y, and z directions;

[0028] The original spatial parameters are converted into logarithmic spatial parameters according to the following formula:

[0029] In the formula, To convert the random field parameters to logarithmic space.

[0030] Preferably, step S3 further includes the following steps:

[0031] S31: Calculate the surface soil displacement under construction disturbance using the finite element method for random fields under different conditions;

[0032] S32: Calculate the set of maximum displacement values ​​for all surface soil masses. ,in For soil surface i The displacement value of the node; ns The number of nodes on the soil surface;

[0033] S33: Statistical Sets S The value of the middle element is greater than the design value. index I Set index I The conditional random field is used to obtain the candidate sample set P.

[0034] Preferably, step S4 further includes the following step:

[0035] S41: Group the conditional random fields in the candidate sample set P according to different coordinates to form a set of parameter matrices for different conditional random fields under the same coordinates:

[0036] In the formula For the first One candidate sample in Parameter vector in coordinates; These represent the first to the last coordinates in the conditional random field; The total number of elements in the candidate sample set P;

[0037] S42: For sets CS Each element in the sequence is calculated according to the following formula to determine the average variance of the parameters in the sequence. :

[0038] In the formula The dimension of the parameter vector in the element; For the target element i In the conditional random field, the th k The values ​​of the parameters; For the target element k The mean of each parameter is calculated using the following formula: ;

[0039] S43: Statistical Compilation Collection CS The average variance of all elements in the set is formed. CSS :

[0040] In the formula for Mean variance in coordinate system;

[0041] S44: Divide the engineering analysis region into grids according to the borehole sampling scale. For each grid, include all elements within the grid boundary. The set of coordinates CSS The elements in the formula are summed to measure the similarity of parameters within each grid region in all conditional random fields: ;

[0042] S45: Perform statistical analysis on the regional variance, arrange them from smallest to largest, and mark the regions less than or equal to the preset threshold as additional sampling points.

[0043] Preferably, when optimizing the hyperparameters of the trend term function, a genetic algorithm, a particle swarm optimization algorithm, or a gray wolf optimization algorithm is used.

[0044] The beneficial effects of this invention are:

[0045] 1) This invention differs from previous drilling site design methods that relied on experience or simple statistical methods. It introduces conditional random field theory and similarity analysis for the first time, realizing dynamic optimization of borehole layout. By establishing a depth-dependent three-dimensional conditional random field, this invention can generate a soil parameter field that better reflects actual geological variability under the constraint of limited borehole data. It can also simulate deformation in conjunction with construction disturbance conditions, thereby effectively identifying high-risk areas.

[0046] 2) This invention proposes a supplementary point placement strategy based on the local similarity of candidate samples, enabling the new point placement to maximize parameter identification results and avoiding the problems of blind, redundant, or insufficient point placement in traditional methods. Compared with existing techniques that rely solely on geostatistical interpolation, this invention comprehensively considers spatial variability, depth correlation, and construction condition effects, reducing unnecessary borehole counts while ensuring exploration accuracy and improving the economy and scientific rigor of the exploration. It provides a dynamic and iteratively optimized point placement approach for engineering exploration under complex geological conditions, which is of great significance for risk identification and safety control in tunnel, subway, and underground space engineering projects. Attached Figure Description

[0047] Figure 1 This is a flowchart of the method of the present invention;

[0048] Figure 2This is a schematic diagram of the initial drilling point;

[0049] Figure 3 For the generated conditional random field plot;

[0050] Figure 4 This is a plot showing the average variance of the random field and local supplementary test points. Detailed Implementation

[0051] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] See Figures 1-4 This invention provides a technical solution: a dynamic optimization method for drilling site layout based on conditional random fields, comprising the following steps:

[0053] S1: Set the initial drilling points according to the gridded uniform sampling, and perform drilling sampling to obtain the sampling point test data;

[0054] S2: Based on the test data of the sampling points, calculate the coefficient of variation and mean of the soil parameters, as well as the lateral and longitudinal coefficients of variation, and generate multiple sets of conditional random fields with depth dependence based on the conditional random field theory.

[0055] S3: Introduce construction excavation or building conditions, calculate the soil displacement state under conditional random fields, count the maximum deformation value of the surface soil under each conditional random field, mark the conditional random fields with the maximum deformation value greater than the engineering design value as candidate samples, and form a candidate sample set P.

[0056] S4: Calculate the similarity of local regions among all candidate samples, and mark the regions that meet the similarity requirements as additional sampling points;

[0057] S5: Drill holes at the marked sampling points to be supplemented, conduct supplementary sampling and testing, and then repeat S2 to S4 until the candidate sample set P is empty or the candidate sample set P no longer decreases. At this point, it is considered that there is a geological risk at the sampling point.

[0058] In this embodiment, S1 is as follows Figure 2 As shown, initial drilling points were set up according to a gridded uniform sampling method, and drilling sampling was carried out. The hole depth was no less than twice the construction depth. In S2, based on the test data of the sampling points, the coefficient of variation and mean of soil parameters were calculated, as well as the lateral and longitudinal coefficients of variation. The correlation between depth and the random field was also calculated. Figure 3As shown, based on the conditional random field theory, multiple conditional random fields with depth dependence are generated; in S3, construction excavation or building conditions are introduced to calculate the soil displacement state under random field conditions, and the maximum deformation of the surface soil under each conditional random field is statistically analyzed; the value of the maximum deformation of the random field is analyzed and compared with the engineering design value. If it is greater than the design value, it is marked as a candidate sample, and a new set of all candidate samples is formed as P; in S4, the similarity of local regions among all candidate samples is calculated, such as... Figure 4 As shown, in all samples, the local regions that meet the similarity requirements are marked as sampling points to be supplemented; S5, drill holes in the sampling points, and after the supplemented sampling and testing are completed, execute S2 again until P is an empty set.

[0059] In some embodiments, S2 further includes the following steps:

[0060] S21: Denoise and remove outliers from the sampled test data, and organize it into layers according to the stratigraphic boundaries. Different groups are used for different stratigraphic boundaries. In subsequent operations, the different groups are independently statistically analyzed and conditional random fields are generated.

[0061] S22: Test and calculate the soil parameters at the sampling points, perform logarithmic transformation, and take the exponential value from the original data. ,in y These are the soil parameters after logarithmic transformation. The raw data obtained from the test; for different borehole coordinates The original material parameter sequence is ordered by depth. Decompose using the following formula:

[0062] In the formula The first i The horizontal, vertical, and axial coordinates of each borehole sampling point For the first i Material parameters of each borehole sampling point For fitting depth for all material parameters The trend of change, For the error term that cannot be fitted, the following formula is used to fit the original trend term curve:

[0063] In the formula , where is the trend term function of the parameter to be optimized. a , b and c For different coefficients to be optimized; The first sampling depth in the depth direction; This is the last sampling depth in the depth direction; For depth iActual observed values;

[0064] For the error term that cannot be fitted The observation errors of soil parameters within the same stratum were statistically analyzed using a normal distribution, and a joint correlation matrix was established.

[0065] S23: Based on soil variability indices, according to the grid coordinates of finite element analysis. Generate multiple sets of log-normal random fields of unconditionally stationary soil observation residuals ,in n The number of predicted points; the set of borehole points is denoted as . , m Given the number of sampling points; calculate the covariance matrix between the known sampling points using the following formula:

[0066] In the formula The covariance matrix between sampling points; p , q For the indexes of two different borehole sampling points; The function for calculating covariance; These are the borehole sampling points. p Coordinates in the x, y, and z directions; These are the borehole sampling points. q Coordinates in the x, y, and z directions;

[0067] The covariance matrix between the predicted points and the sampled points is calculated using the following formula, with dimension 1. n × m :

[0068] In the formula i , j These are the indices of the prediction points and the borehole sampling points, respectively. This is the covariance matrix between the sampling points and the prediction points; These are the prediction points. i Coordinates in the x, y, and z directions; These are the borehole sampling points. j Coordinates in the x, y, and z directions;

[0069] S24: Calculate the residual vector d medium elements And based on the following formula, a conditional random field containing deep dependencies is obtained:

[0070] In the formula For the parameters of the random field under different coordinates in the simulation; This represents the fitted trend term as depth increases; These are the borehole sampling points.i Coordinates in the x, y, and z directions;

[0071] The original spatial parameters are converted into logarithmic spatial parameters according to the following formula:

[0072] In the formula, To convert the random field parameters to logarithmic space.

[0073] In this embodiment, when optimizing the hyperparameters of the trend term function, genetic algorithms, particle swarm optimization (PSO), and gray wolf optimization algorithms can be used. PSO is intuitive and easy to implement, requiring no complex mathematical derivations or gradient information, making it suitable for handling nonlinear, non-differentiable, and even black-box problems. Through a group-based co-evolution mechanism, it searches the solution space globally, possessing strong global optimization capabilities while avoiding getting trapped in local optima. Furthermore, it has fewer parameters, is easy to adjust, and converges quickly, especially in the early iteration stages, rapidly approaching a better solution. In addition, PSO has good versatility and scalability, allowing it to be combined with other optimization methods or domain knowledge, making it suitable for various scenarios such as engineering optimization, machine learning, and complex system modeling. PSO is used to optimize its hyperparameters.

[0074] For the error term that cannot be fitted The observation errors of soil parameters within the same stratum are statistically analyzed using a normal distribution, and a joint correlation matrix is ​​established; for the th l The coefficient of variation for each layer can be calculated using the following formula. COV :

[0075] ;

[0076] When calculating the fluctuation distance, different soil deformability models can be used for fitting. First, the half-variogram is calculated using the following formula: In the formula For a certain direction at a relevant length of h The semivariance over time; This represents the number of sample point logs under this correlation length; This represents the error at different sampling points within the corresponding length. After the semivariogram is calculated, a function model is used for simulation to obtain the fluctuation distance. An exponential model is commonly used for fitting, and the exponential function is shown below:

[0077] In the formula The nugget value represents noise or microscale variation. The stratification difference represents a portion that contributes to the total variance. For direction d The relevant length, i.e., the fluctuation distance. The distance parameter can be optimized using the PSO method.

[0078] When calculating the covariance matrix, the covariance can be calculated using the following formula:

[0079] In the formula For residual variance; For correlation coefficient functions, the following formula can be used to calculate them for exponential functions:

[0080] ;

[0081] In the formula, For horizontal correlation length, This refers to the vertical related length.

[0082] In some embodiments, S3 further includes the following steps:

[0083] S31: Calculate the surface soil displacement under construction disturbance using the finite element method for random fields under different conditions;

[0084] S32: Calculate the set of maximum displacement values ​​for all surface soil masses. ,in For soil surface i The displacement value of the node; ns The number of nodes on the soil surface;

[0085] S33: Statistical Sets S The value of the middle element is greater than the design value. index I Set index I The conditional random field is used to obtain the candidate sample set P.

[0086] In this embodiment, the finite element method is used in S31 to calculate the surface soil displacement under construction disturbance under different random fields. First, the ground stress balance is performed, and different construction conditions are set according to the construction steps under the stress field state, such as excavation and building construction, to calculate the new equilibrium state of the soil under external disturbance.

[0087] In some embodiments, S4 further includes the following step:

[0088] S41: Group the conditional random fields in the candidate sample set P according to different coordinates to form a set of parameter matrices for different conditional random fields under the same coordinates:

[0089] In the formula For the first One candidate sample in Parameter vector in coordinates; These represent the first to the last coordinates in the conditional random field; The total number of elements in the candidate sample set P;

[0090] S42: For sets CS Each element in the sequence is calculated according to the following formula to determine the average variance of the parameters in the sequence. :

[0091] In the formula The dimension of the parameter vector in the element; For the target element i In the conditional random field, the th k The values ​​of the parameters; For the target element k The mean of each parameter is calculated using the following formula: ;

[0092] S43: Statistical Compilation Collection CS The average variance of all elements in the set is formed. CSS :

[0093] In the formula for Mean variance in coordinate system;

[0094] S44: Divide the engineering analysis region into grids according to the borehole sampling scale. For each grid, include all elements within the grid boundary. The set of coordinates CSS The elements in the formula are summed to measure the similarity of parameters within each grid region in all conditional random fields: ;

[0095] S45: Perform statistical analysis on the regional variance, arrange them from smallest to largest, and mark the regions less than or equal to the preset threshold as additional sampling points.

[0096] In some embodiments, when optimizing the hyperparameters of the trend term function, a genetic algorithm, a particle swarm optimization algorithm, or a gray wolf optimization algorithm is used.

[0097] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A dynamic optimization method for drilling site layout based on conditional random fields, characterized in that: Includes the following steps: S1: Set the initial drilling points according to the gridded uniform sampling, and perform drilling sampling to obtain the sampling point test data; S2: Based on the test data of the sampling points, calculate the coefficient of variation and mean of the soil parameters, as well as the lateral and longitudinal coefficients of variation, and generate multiple sets of conditional random fields with depth dependence based on the conditional random field theory. S3: Introduce construction excavation or building conditions, calculate the soil displacement state under conditional random fields, count the maximum deformation value of the surface soil under each conditional random field, mark the conditional random fields with the maximum deformation value greater than the engineering design value as candidate samples, and form a candidate sample set P. S4: Calculate the similarity of local regions among all candidate samples, and mark the regions that meet the similarity requirements as additional sampling points; S5: Drill holes at the marked sampling points to be supplemented, conduct supplementary sampling and testing, and then repeat S2 to S4 until the candidate sample set P is empty; or the candidate sample set P no longer decreases, at which point it is considered that there is a geological risk at the sampling site; The S2 further includes the following steps: S21: Denoise and remove outliers from the sampled test data, and organize it into layers according to the stratigraphic boundaries. Different groups are used for different stratigraphic boundaries. In subsequent operations, the different groups are independently statistically analyzed and conditional random fields are generated. S22: Test and calculate the soil parameters at the sampling points, perform logarithmic transformation, and take the exponential value from the original data. ,in y These are the soil parameters after logarithmic transformation. The raw data obtained from the test; for different borehole coordinates The original material parameter sequence is ordered by depth. Decompose using the following formula: In the formula The first i The horizontal, vertical, and axial coordinates of each borehole sampling point For the first i Material parameters of each borehole sampling point For fitting depth for all material parameters The trend of change, For the error term that cannot be fitted, the following formula is used to fit the original trend term curve: In the formula , where is the trend term function of the parameter to be optimized. a , b and c For different coefficients to be optimized; The first sampling depth in the depth direction; This is the last sampling depth in the depth direction; For depth i Actual observed values; For the error term that cannot be fitted The observation errors of soil parameters within the same stratum were statistically analyzed using a normal distribution, and a joint correlation matrix was established. S23: Based on soil variability indices, according to the grid coordinates of finite element analysis. Generate multiple sets of log-normal random fields of unconditionally stationary soil observation residuals ,in n The number of predicted points; the set of borehole points is denoted as . , m Given the number of sampling points; calculate the covariance matrix between the known sampling points using the following formula: In the formula The covariance matrix between sampling points; p , q For the indexes of two different borehole sampling points; The function for calculating covariance; These are the borehole sampling points. p Coordinates in the x, y, and z directions; These are the borehole sampling points. q Coordinates in the x, y, and z directions; The covariance matrix between the predicted points and the sampled points is calculated using the following formula, with dimension 1. n × m : In the formula i , j These are the indices of the prediction points and the borehole sampling points, respectively. This is the covariance matrix between the sampling points and the prediction points; Prediction points i Coordinates in the x, y, and z directions; These are the borehole sampling points. j Coordinates in the x, y, and z directions; S24: Calculate the residual vector d Middle elements And based on the following formula, a conditional random field containing deep dependencies is obtained: In the formula For the parameters of the random field under different coordinates in the simulation; This represents the fitted trend term as depth increases; These are the borehole sampling points. i Coordinates in the x, y, and z directions; The original spatial parameters are converted into logarithmic spatial parameters according to the following formula: In the formula, To convert to random field parameters in logarithmic space; The S4 further includes the following steps: S41: Group the conditional random fields in the candidate sample set P according to different coordinates to form a set of parameter matrices for different conditional random fields under the same coordinates: In the formula For the first One candidate sample in Parameter vector in coordinates; These represent the first to the last coordinates in the conditional random field; The total number of elements in the candidate sample set P; S42: For sets CS Each element in the sequence is calculated according to the following formula to determine the average variance of the parameters in the sequence. : In the formula The dimension of the parameter vector in the element; For the target element i In the conditional random field, the th k The values ​​of the parameters; For the target element k The mean of each parameter is calculated using the following formula: ; S43: Statistical Compilation Collection CS The average variance of all elements in the set is formed. CSS : In the formula for Mean variance in coordinate system; S44: Divide the engineering analysis region into grids according to the borehole sampling scale. For each grid, include all elements within the grid boundary. The set of coordinates CSS The elements in the formula are summed to measure the similarity of parameters within each grid region in all conditional random fields: ; S45: Perform statistical analysis on the regional variance, arrange them from smallest to largest, and mark the regions less than or equal to the preset threshold as additional sampling points.

2. The dynamic optimization method for drilling site layout based on conditional random fields according to claim 1, characterized in that: The S3 further includes the following steps: S31: Calculate the surface soil displacement under construction disturbance using the finite element method for random fields under different conditions; S32: Calculate the set of maximum displacement values ​​for all surface soil masses. ,in For soil surface i The displacement value of the node; ns The number of nodes on the soil surface; S33: Statistical Sets S The value of the middle element is greater than the design value. index I Set index I The conditional random field is used to obtain the candidate sample set P.

3. The dynamic optimization method for drilling site layout based on conditional random fields according to claim 1, characterized in that: When optimizing the hyperparameters of the trend term function, genetic algorithms, particle swarm optimization algorithms, or gray wolf optimization algorithms can be used.

Images