A method and system for microgravity recognition of underground spaces
By acquiring measured microgravity data and prior engineering geological information, and combining gravity anomaly analysis formulas and deep learning models, the problems of interference suppression and accuracy improvement in microgravity identification in urban environments were solved, achieving high-precision microgravity identification in complex environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST JIAOTONG UNIV
- Filing Date
- 2026-04-13
- Publication Date
- 2026-06-16
Smart Images
Figure CN122018028B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of gravity identification, and more specifically, to a microgravity identification method and system for underground spaces. Background Technology
[0002] Gravity exploration is a classic geophysical method. Its basic principle is to detect gravity anomalies caused by uneven density distribution of underground materials by measuring subtle changes in the Earth's surface gravity field. Based on this, the core applications of gravity exploration can be divided into two aspects: first, directly using gravity anomalies to identify underground anomalies; and second, through gravity inversion, quantitatively calculating the density differences, geometric shapes, and spatial locations of underground media, thereby inferring the characteristics of underground structures.
[0003] Gravity inversion, as a crucial step in inferring subsurface properties from observational data, has seen the development of numerous mature methods (such as regularization-based 3D inversion and Parker-Oldenburg density interface inversion). These methods are highly effective in handling regional or large-scale structures. However, when dealing with small-scale, shallow targets (such as shallow tunnels and underground cavities), they often face a dual dilemma: first, insufficient resolution makes it difficult to accurately characterize the target body; second, the inversion results exhibit severe non-uniqueness (i.e., multiple subsurface models may produce similar gravity anomalies), leading to increased uncertainty in the interpretation results.
[0004] Correspondingly, microgravity surveys aimed at detecting such small local targets place higher demands on data accuracy and processing methods. Especially in densely populated environments such as cities, interference signals from surface topography, buildings, and shallow bedrock interfaces can severely mask or confuse the weak gravity anomalies emanating from the target, further limiting the accuracy of identification.
[0005] Therefore, how to effectively suppress interference and improve the accuracy of gravity identification of shallow small targets has become a key technical problem that urgently needs to be solved in the field of microgravity exploration. Summary of the Invention
[0006] The purpose of this application is to provide a microgravity identification method and system for underground spaces, which can improve the accuracy of microgravity identification in complex shallow urban environments.
[0007] This application is implemented as follows:
[0008] In a first aspect, this application provides a microgravity identification method for underground spaces, comprising the following steps:
[0009] S1: Obtain the measured microgravity raw data of the target area and the engineering geological prior information including the geometric type of the target body. The engineering geological prior information includes geometric parameters.
[0010] S2: Define the key geometric parameters based on the geometric type of the target body, set the variation range of each geometric parameter, and generate multiple sets of geometric parameter combinations through traversal sampling; use the gravity anomaly analytical formula to perform forward modeling calculation on each set of geometric parameter combinations to generate a synthetic sample dataset composed of gravity anomaly sequences and corresponding geometric parameters.
[0011] S3: Perform tidal correction and zero drift correction on the measured microgravity raw data to obtain the basic gravity anomaly value. After determining the terrain correction range of the basic gravity anomaly value, perform terrain correction. Then, perform height correction and intermediate layer correction on the basic gravity anomaly value. After correction, calculate the Bouguer gravity anomaly.
[0012] S4: Determine the influence range of shallow disturbance, and based on the influence range of shallow disturbance, remove the shallow disturbance field from the Bouguer gravity anomaly to obtain the target gravity anomaly data for inversion.
[0013] S5: Construct a deep learning inversion model and train it using a synthetic sample dataset; input the target gravity anomaly data for inversion into the trained deep learning inversion model and output the prediction results of the geometric parameters and spatial location of the underground target.
[0014] Based on the first aspect, it also includes:
[0015] S6: Substitute the geometric parameters predicted by the deep learning inversion model into the forward model to calculate the theoretical gravity anomaly, compare it with the measured or processed gravity anomaly, and calculate the residual; verify the consistency between the predicted geometric parameters and the engineering survey data, and comprehensively evaluate the accuracy and confidence of the prediction results.
[0016] Based on the first aspect, the geometric parameters include the radius, burial depth, spacing, and horizontal position of the target body.
[0017] Based on the first aspect, the steps for performing forward modeling calculations on each combination of geometric parameters using the analytical formula for gravity anomalies specifically include:
[0018] Forward modeling was performed using the analytical formula for gravity anomalies;
[0019] ;
[0020] in, G is the gravitational anomaly value; G is the gravitational constant. h is the center burial depth of the underground cylinder; σ is the density difference between the cylinder and the surrounding rock strata; R is the radius of the cylinder; dx1 is the horizontal distance between the center of the underground left cylinder and the measuring point on the ground surface; dx2 is the horizontal distance between the center of the underground right cylinder and the measuring point on the ground surface; the gravity anomaly values of multiple observation points are calculated along the preset survey line to form a gravity anomaly sequence.
[0021] Based on the first aspect, the steps for generating a synthetic sample dataset consisting of gravity anomaly sequences and corresponding geometric parameters include: superimposing random Gaussian noise on the gravity anomaly sequences of some samples to improve the robustness of the trained model.
[0022] Based on the first aspect, terrain correction includes building correction and natural terrain correction. Buildings and surface terrain are discretized into multiple rectangular prism units. The gravity anomaly of each rectangular prism unit is calculated using the gravity anomaly calculation formula to determine the gravity anomaly generated by each terrain unit for the measuring points within the survey area. For buildings and natural terrain of different sizes, the decay curve of the gravity anomaly difference caused by the measuring points closest to and farthest from the buildings is calculated as a function of distance. When the gravity anomaly difference is <1μGal, the corresponding distance is the cutoff range of terrain correction.
[0023] Formula for calculating gravity anomalies:
[0024] ;
[0025] ;
[0026] ;
[0027] ;
[0028] ;
[0029] Where G is the gravitational constant, ; It is the density difference between the building and the air; the coordinates of the observation point are... The boundary of a rectangular prism in global coordinates x1 is the relative horizontal distance from the observation point to the left side of the prism; x2 is the relative horizontal distance from the observation point to the right side of the prism; y1 is the relative distance from the observation point to the front edge of the prism; y2 is the relative distance from the observation point to the rear edge of the prism; z1 is the height difference from the observation point to the top of the prism; z2 is the height difference from the observation point to the bottom of the prism. Let be the distances from the observation point to each of the eight vertices of the rectangular prism;
[0030] Finally, by integrating, the gravity anomaly generated by all rectangular prism elements within the truncation range of the terrain correction on the measuring point is obtained, yielding the terrain correction value TC.
[0031] Based on the first aspect, the adoption of highly corrective measures ;in, This is the height correction value. The height difference between the observation surface and the reference surface;
[0032] Intermediate layer correction adopted ;in, Here, G is the intermediate layer correction value, and G is the gravitational constant. ; ρ is the average density of the intermediate layer; t is the thickness of the intermediate layer;
[0033] The calculation of Bouguer gravity anomaly uses ;in, Based on the basic gravity anomaly, This is the terrain correction value. This is the height correction value. Intermediate layer correction value.
[0034] Based on the first aspect, the steps of determining the influence range of shallow disturbances, and then removing the shallow disturbance field from the Bouguer gravity anomaly based on the influence range of shallow disturbances to obtain the target gravity anomaly data for inversion include:
[0035] The influence of the undulating density interface on the surface gravity anomaly under different burial depths and amplitudes was analyzed to determine the range of potential field separation, which is the range of influence of shallow surface disturbance.
[0036] Bouguer gravity anomaly Based on this, the influence range of shallow interference is stripped away to obtain the regional field anomaly with shallow interference removed, which is then used for inversion of the target gravity anomaly; specifically including:
[0037] Setting field anomaly From regional field and local field Composition, namely:
[0038] set up Representatives and Points Distance The average value of a certain 4-point potential field, i.e.:
[0039]
[0040] In the formula, The cutting radius is; the region field is and The weighted average, i.e.
[0041]
[0042] in, and These are weighting coefficients, and ;
[0043] ;
[0044] ;
[0045] ;
[0046] ;
[0047] ;
[0048] ;
[0049] ;
[0050] ;
[0051] in, The gradient is in the horizontal direction; The gradient is in the vertical direction. The deviation between the center point Z(x,y) and half of the difference in the horizontal direction; The deviation between the center point Z(x,y) and half of the vertical difference;
[0052] The calculated region field is called the first-order cut region field. Indicate; to Repeat the calculation to obtain the second cutting region. ; continuing this iteration, eventually there will be
[0053] ;
[0054] Therefore, there is
[0055]
[0056] Through multiple iterations until the regional field changes tend to stabilize, the target gravity anomaly data used for inversion is finally obtained.
[0057] Based on the first aspect, the target body is one or more of the following: tunnel, cavity, or pipeline corridor.
[0058] Secondly, this application provides a microgravity identification system for underground spaces, comprising:
[0059] Acquisition module: used to acquire the measured microgravity raw data of the target area and the engineering geological prior information including the geometric type of the target body. The engineering geological prior information includes geometric parameters.
[0060] Forward modeling module: It is used to define the key geometric parameters of the target body based on its geometric type, set the variation range of each geometric parameter, and generate multiple sets of geometric parameter combinations by traversing sampling; it uses the gravity anomaly analytical formula to perform forward modeling calculations on each set of geometric parameter combinations, and generates a synthetic sample dataset composed of gravity anomaly sequences and corresponding geometric parameters.
[0061] The gravity anomaly calculation module is used to perform tidal correction and zero drift correction on the measured microgravity raw data to obtain the basic gravity anomaly value. After determining the terrain correction range of the basic gravity anomaly value, terrain correction is performed. The basic gravity anomaly value is then corrected for height and intermediate layer. After correction, the Bouguer gravity anomaly is calculated.
[0062] Stripping module: Used to determine the influence range of shallow disturbances, and strip the shallow disturbance field from the Bouguer gravity anomaly based on the influence range of shallow disturbances to obtain the target gravity anomaly data for inversion;
[0063] Prediction module: Used to build a deep learning inversion model, train the deep learning inversion model using a synthetic sample dataset; input the target gravity anomaly data for inversion into the trained deep learning inversion model, and output the prediction results of the geometric parameters and spatial location of the underground target.
[0064] Compared with the prior art, this application has at least the following advantages or beneficial effects:
[0065] This invention provides a method and system for microgravity identification in underground spaces. By acquiring measured raw microgravity data of the target area and prior engineering geological information including the geometric type of the target body, it ensures that the analysis is based on real-world physical observations rather than purely theoretical simulations, making the conclusions verifiable. This provides crucial data for subsequent forward modeling and inversion interpretation. Tidal and zero-drift corrections eliminate the periodic effects of instrument drift over time and celestial gravity, obtaining basic gravity anomaly values reflecting the static underground structure. Topographic, height, and intermediate layer corrections successively eliminate the influence of surface undulations, elevation differences at observation points, and the mass of rock layers between the measuring point and the reference surface. The resulting Bouguer gravity anomaly is theoretically a pure signal caused solely by lateral density inhomogeneities in the underground (target body). By determining the influence range of shallow interference, the shallow interference field is extracted from the Bouguer gravity anomaly based on this range, yielding target gravity anomaly data for inversion. This allows for the identification and correction of such weak interferences, significantly improving the accuracy and reliability of microgravity identification in underground spaces. By specifically stripping away shallow interference fields, essentially further refining the signal from its pure state, we can effectively extract weak residual anomalies caused by deep targets, greatly improving the effectiveness and reliability in areas of human activity. By training the deep learning model on massive amounts of samples, it internalizes the complex mapping relationship between the entire parameter space and the anomaly space, directly providing the most probable globally optimal solution, resulting in more stable and reliable results. Training the deep learning inversion model using synthetic sample datasets enables it to automatically complete the entire process from raw data to the final interpretation report. Attached Figure Description
[0066] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0067] Figure 1 This is a flowchart of an embodiment of a microgravity identification method for underground space according to this application;
[0068] Figure 2 This is a flowchart of another embodiment of a microgravity identification method for underground space according to this application;
[0069] Figure 3 This is a schematic diagram of an embodiment of a microgravity identification system for underground space according to this application.
[0070] icon:
[0071] 1. Acquisition module; 2. Forward modeling module; 3. Gravity anomaly calculation module; 4. Stripping module; 5. Prediction module. Detailed Implementation
[0072] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0073] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the various embodiments and features described below can be combined with each other.
[0074] Example
[0075] This application provides a microgravity identification method and system for underground spaces, which can improve the accuracy of microgravity identification in complex shallow urban environments.
[0076] Please refer to Figure 1 This method for microgravity identification in underground spaces includes the following steps:
[0077] S1: Obtain the measured microgravity raw data of the target area and the engineering geological prior information including the geometric type of the target body. The engineering geological prior information includes geometric parameters.
[0078] Specifically, obtaining raw, measured microgravity data ensures that the method is based on real-world physical observations rather than purely theoretical simulations, making the conclusions verifiable. Acquiring prior information containing the geometric type of the target body (such as "tunnels" or "cavities") provides crucial data for subsequent forward and inverse modeling interpretations.
[0079] As one implementation method, the target body can be one or more of a tunnel, cavity, or pipeline corridor. Geometric parameters include the target body's radius, depth, spacing, and horizontal position. For example, the target could be two tunnels with the following geometric parameters: tunnel radius r = 3.350m, depth h = 29.280m, spacing d = 15.920m, and horizontal position D = 35.000m.
[0080] S2: Define the key geometric parameters based on the geometric type of the target body, set the variation range of each geometric parameter, and generate multiple sets of geometric parameter combinations through traversal sampling; use the gravity anomaly analytical formula to perform forward modeling calculation on each set of geometric parameter combinations to generate a synthetic sample dataset composed of gravity anomaly sequences and corresponding geometric parameters.
[0081] Specifically, considering the massive data required for deep learning, while real-world engineering test cases are scarce and costly, this step automatically generates tens of thousands of high-quality "parameter-anomaly" pairings through parameterized traversal and physical forward modeling for model training. By setting the parameter variation range and traversing the sampling, the generated dataset covers various possible geological conditions (targets of different sizes, burial depths, and locations). The model trained with this dataset possesses strong generalization ability to handle unknown and complex scenarios. Using the analytical formula for gravity anomalies for forward modeling ensures that every generated data point strictly conforms to Newton's law of universal gravitation, fundamentally endowing the deep learning model with physical interpretability and avoiding absurd predictions that might occur with purely data-driven models.
[0082] As one implementation method, taking the two tunnels as an example in step S1, the range of geometric parameters can be set as follows: tunnel radius r∈[1,5]m, step size 0.2m; burial depth h∈[22,30]m, step size 1m; spacing d∈[12,20]m, step size 0.5m; center position D∈[15,75]m, step size 5m. Traversing the parameter space generates approximately 50,000 parameter combinations.
[0083] As one implementation method, the steps of performing forward modeling calculations on each combination of geometric parameters using the analytical formula for gravity anomalies specifically include:
[0084] Forward modeling was performed using the analytical formula for gravity anomalies;
[0085] ;
[0086] in, G is the gravitational anomaly value; G is the gravitational constant. h is the center burial depth of the underground cylinder; σ is the density difference between the cylinder and the surrounding rock strata; R is the radius of the cylinder; dx1 is the horizontal distance between the center of the underground left cylinder and the measuring point on the ground surface; dx2 is the horizontal distance between the center of the underground right cylinder and the measuring point on the ground surface; the gravity anomaly values of multiple observation points are calculated along the preset survey line to form a gravity anomaly sequence.
[0087] Furthermore, the steps for generating a synthetic sample dataset consisting of gravity anomaly sequences and corresponding geometric parameters include: superimposing random Gaussian noise on the gravity anomaly sequences of a portion of the samples to improve the robustness of the trained model. For example, superimposing Gaussian noise (amplitude ±5~10μGal) on 10% of the samples can improve model robustness. A dataset is constructed that consists of input (gravity anomaly values at 25 points) and output (r, h, d, D corresponding to the survey line).
[0088] S3: Perform tidal correction and zero drift correction on the measured microgravity raw data to obtain the basic gravity anomaly value. After determining the terrain correction range of the basic gravity anomaly value, perform terrain correction. Then, perform height correction and intermediate layer correction on the basic gravity anomaly value. After correction, calculate the Bouguer gravity anomaly.
[0089] It is preferable to use a CG-6 gravimeter to perform tidal correction and zero drift correction on the measured microgravity raw data.
[0090] Specifically, tidal and zero-drift corrections eliminate the periodic effects of instrument drift over time and celestial gravity, yielding fundamental gravity anomaly values reflecting the static subsurface structure. Topographic, elevation, and intermediate-layer corrections successively eliminate the influence of surface undulations, elevation differences at observation points, and the mass of rock strata between the measuring point and the reference surface. The resulting Bouguer gravity anomaly is theoretically a pure signal caused solely by lateral density inhomogeneities in the subsurface (the target body). Removing the largest, calculable noise source makes subsequent detailed analysis of the target body possible.
[0091] As one implementation method, terrain correction includes building correction and natural terrain correction. Buildings and surface terrain are discretized into multiple rectangular prism units. Preferably, buildings are divided into 1m*1m*1m rectangular prism units, and undulating surface terrain is divided into 5m*5m units (5m*5m depends on the spacing between measuring points). ( A rectangular prism terrain unit is defined as the height difference between the undulating surface and the reference surface. For each rectangular prism unit, the gravity anomaly generated by that unit at the measuring points within the survey area is calculated using the gravity anomaly calculation formula. For buildings of different sizes and natural terrain, the decay curves of the gravity anomaly difference caused by the measuring points closest to and farthest from the buildings are calculated as a function of distance. When the gravity anomaly difference is <1 μGal, the corresponding distance is the cutoff range for terrain correction.
[0092] Formula for calculating gravity anomalies:
[0093]
[0094]
[0095]
[0096]
[0097] ;
[0098] Where G is the gravitational constant, ; It is the density difference between the building and the air; the coordinates of the observation point are... The boundary of a rectangular prism in global coordinates x1 is the relative horizontal distance from the observation point to the left side of the prism; x2 is the relative horizontal distance from the observation point to the right side of the prism; y1 is the relative distance from the observation point to the front edge of the prism; y2 is the relative distance from the observation point to the rear edge of the prism; z1 is the height difference from the observation point to the top of the prism; z2 is the height difference from the observation point to the bottom of the prism. Let be the distances from the observation point to each of the eight vertices of the rectangular prism;
[0099] Finally, by integrating, the gravity anomaly generated by all rectangular prism elements within the truncation range of the terrain correction on the measuring point is obtained, yielding the terrain correction value TC.
[0100] As one implementation method, highly corrected use ;in, This is the height correction value. The height difference between the observation surface and the reference surface;
[0101] Intermediate layer correction adopted ;in, Here, G is the intermediate layer correction value, and G is the gravitational constant. ; ρ is the average density of the intermediate layer; t is the thickness of the intermediate layer;
[0102] The calculation of Bouguer gravity anomaly uses ;in, Based on the basic gravity anomaly, This is the terrain correction value. This is the height correction value. Intermediate layer correction value.
[0103] S4: Determine the influence range of shallow disturbance, and based on the influence range of shallow disturbance, remove the shallow disturbance field from the Bouguer gravity anomaly to obtain the target gravity anomaly data for inversion.
[0104] Specifically, drastic density changes within a few meters of the surface can generate high-frequency, high-amplitude interference anomalies, easily masking or distorting signals from deeper targets. This step can strip away shallow interference. Since shallow undulating interfaces typically have a meter-scale spatial dimension, the gravity anomalies they cause are on the microgal level, similar to the anomalies generated by targets in underground space. This step can identify and correct such weak interference, significantly improving the accuracy and reliability of microgravity identification in underground spaces. By specifically stripping away the shallow interference field, it's equivalent to further purifying the pure signal, effectively extracting weak residual anomalies caused by deep targets, greatly enhancing the effectiveness and reliability in areas with human activity.
[0105] As one implementation method, the steps of determining the influence range of shallow disturbance and stripping the shallow disturbance field from the Bouguer gravity anomaly based on the influence range of shallow disturbance to obtain the target gravity anomaly data for inversion include: setting the cutting radius and calculating the average value of the potential field grid point and its surrounding points; calculating the gradient component based on the potential field value of the point and the surrounding points, and dynamically determining the weighting coefficients of the regional field and the local field accordingly; iteratively calculating the regional field using the weighting coefficients until the result converges, and using the converged result as the stripped regional background field to obtain the remaining gravity anomaly. Specifically:
[0106] The influence of undulating density interfaces on surface gravity anomalies under different burial depths and amplitudes was analyzed to determine the range of potential field separation, which is the influence range of shallow disturbance. Specific steps included: when the interface depth was h, the gravity anomaly caused by the interface was greater than 1 / 3 of that caused by the underground anomaly; when the interface depth was greater than h, the amplitude of the anomaly caused by the interface was less than 1 / 3 of that caused by the underground anomaly, and its influence on surface gravity anomalies was relatively small and negligible. Therefore, the range of potential field separation was determined to be h.
[0107] Bouguer gravity anomaly Based on this, the influence range of shallow interference is stripped away to obtain the regional field anomaly with shallow interference removed, which is then used for inversion of the target gravity anomaly; specifically including:
[0108] Setting field anomaly From regional field and local field Composition, namely:
[0109] set up Representatives and Points Distance The average value of a certain 4-point potential field, i.e.:
[0110]
[0111] In the formula, The cutting radius is; the region field is and The weighted average, i.e.
[0112]
[0113] in, and These are weighting coefficients, and ;
[0114] ;
[0115] ;
[0116] ;
[0117] ;
[0118] ;
[0119] ;
[0120] ;
[0121] ;
[0122] in, The gradient is in the horizontal direction; The gradient is in the vertical direction. The deviation between the center point Z(x,y) and half of the difference in the horizontal direction; The deviation between the center point Z(x,y) and half of the vertical difference;
[0123] The calculated region field is called the first-order cut region field. Indicate; to Repeat the calculation to obtain the second cutting region. ; continuing this iteration, eventually there will be
[0124] ;
[0125] Therefore:
[0126]
[0127] Through multiple iterations until the regional field changes tend to stabilize, the target gravity anomaly data used for inversion is finally obtained.
[0128] S5: Construct a deep learning inversion model and train it using a synthetic sample dataset; input the target gravity anomaly data for inversion into the trained deep learning inversion model and output the prediction results of the geometric parameters and spatial location of the underground target.
[0129] Specifically, the inversion process of the deep learning model involves only one forward propagation computation, which can output results in milliseconds, improving efficiency. By learning from massive amounts of samples, it internalizes the complex mapping relationship between the entire parameter space and the anomaly space, enabling it to directly provide the most likely global optimal solution, resulting in more stable and reliable results. The physical laws (forward modeling knowledge) from step S2 and the data processing experience (correction and knowledge stripping) from steps S3 and S4 are fed into the deep learning model. This allows the deep learning model to automatically complete the entire process from raw data to the final interpretation report.
[0130] As one implementation method, a deep learning inversion model is constructed using a fully connected neural network (FCN) network structure. This network structure has an input size of 25, an output size of 4, and 5 hidden layers, with the corresponding number of neurons being 256→512→1024→1024→512, and the activation function being LeakyReLU (α=0.1).
[0131] During training, the input and output are normalized to 0–1 respectively:
[0132] ;
[0133] in, and These are the maximum and minimum values of the corresponding features, respectively.
[0134] The synthetic sample dataset is divided into a training set and a test set, with a 7:3 ratio. The mean squared error (MSE) is used as the loss function.
[0135] ;
[0136] Use the Adam optimizer; train until the validation error is minimized and save the weights. After training, input the target gravity anomaly data to be inverted into the model according to the survey line, output the predicted parameters and inversely normalize them into physical quantities.
[0137] Please refer to Figure 2Furthermore, in order to evaluate the accuracy and confidence of the recognition results of the deep learning model, the following also includes:
[0138] S6: Substitute the geometric parameters predicted by the deep learning inversion model into the forward model to calculate the theoretical gravity anomaly, compare it with the measured or processed gravity anomaly, and calculate the residual; verify the consistency between the predicted geometric parameters and the engineering survey data, and comprehensively evaluate the accuracy and confidence of the prediction results.
[0139] Specifically, the average error, maximum error, and RMSE of the six survey lines were statistically analyzed, and consistency verification was performed in conjunction with engineering data (drilling data, design drawings). Please refer to Table 1 for the statistics of geometric parameter errors.
[0140] Table 1: Statistics of Geometric Parameter Errors
[0141]
[0142] The statistical analysis of geometric parameter errors (as shown in Table 1) provides users with direct expectations. For example, knowing that the error in "buried depth h" is relatively large allows for a more cautious approach to this parameter in practical applications, or feedback can be given to the R&D team for targeted model optimization.
[0143] The geometric parameters predicted by the deep learning inversion model are substituted into the forward model to calculate the theoretical gravity anomaly. This calculation is then compared with the measured or processed gravity anomaly to calculate the residuals. By calculating the residuals and analyzing their distribution, the physical correctness of the entire technology chain can be ultimately verified. If the residuals are too large, the source of the problem can be traced back, such as incomplete data preprocessing, inadequate removal of shallow interference, or defects in the deep learning model itself. This provides a clear diagnostic path for the continuous optimization of the method.
[0144] Simultaneously, the consistency of the predicted geometric parameters with engineering survey data is verified, and the accuracy and confidence of the prediction results are comprehensively evaluated, which greatly enhances users' trust in this deep learning inversion model. Through quantitative statistics, physical verification, and engineering comparison, the effectiveness and high accuracy of the method itself are not only proven, but more importantly, it endows the results with the necessary confidence level in major engineering decisions.
[0145] Please refer to Figure 3 Secondly, this application also provides a microgravity identification system for underground spaces, comprising:
[0146] Acquisition Module 1: Used to acquire the measured microgravity raw data of the target area and the engineering geological prior information including the geometric type of the target body. The engineering geological prior information includes geometric parameters.
[0147] Forward Modeling Module 2: This module is used to define the key geometric parameters of the target body based on its geometric type, set the range of variation for each geometric parameter, and generate multiple sets of geometric parameter combinations through traversal sampling; it uses the analytical formula of gravity anomalies to perform forward modeling calculations on each set of geometric parameter combinations, generating a synthetic sample dataset composed of gravity anomaly sequences and corresponding geometric parameters.
[0148] Module 3 for calculating gravity anomalies: It is used to perform tidal correction and zero drift correction on the measured microgravity raw data to obtain the basic gravity anomaly value, determine the terrain correction range of the basic gravity anomaly value, perform terrain correction, and perform height correction and intermediate layer correction on the basic gravity anomaly value. After correction, the Bouguer gravity anomaly is calculated.
[0149] Stripping Module 4: Used to determine the influence range of shallow disturbances, and strip the shallow disturbance field from the Bouguer gravity anomaly based on the influence range of shallow disturbances to obtain the target gravity anomaly data for inversion.
[0150] Prediction Module 5: Used to build a deep learning inversion model, train the deep learning inversion model using a synthetic sample dataset; input the target gravity anomaly data for inversion into the trained deep learning inversion model, and output the prediction results of the geometric parameters and spatial location of the underground target.
[0151] For a detailed implementation of the microgravity identification system for underground space, please refer to a detailed implementation of a microgravity identification method for underground space; further details will not be provided here.
[0152] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. A microgravity identification method for underground spaces, characterized in that, Includes the following steps: S1: Obtain the measured microgravity raw data of the target area and the engineering geological prior information including the geometric type of the target body, wherein the engineering geological prior information includes geometric parameters; S2: Define the key geometric parameters based on the geometric type of the target body, set the variation range of each geometric parameter, and generate multiple sets of geometric parameter combinations by traversing sampling; Forward modeling is performed on each combination of geometric parameters using the analytical formula for gravity anomalies, generating a synthetic sample dataset consisting of gravity anomaly sequences and corresponding geometric parameters; S3: Perform tidal correction and zero drift correction on the measured microgravity raw data to obtain the basic gravity anomaly value. After determining the terrain correction range of the basic gravity anomaly value, perform terrain correction. Then, perform height correction and intermediate layer correction on the basic gravity anomaly value. After correction, calculate the Bouguer gravity anomaly. S4: Determine the influence range of shallow disturbance, and based on the influence range of shallow disturbance, remove the shallow disturbance field from the Bouguer gravity anomaly to obtain the target gravity anomaly data for inversion. S5: Construct a deep learning inversion model and train the deep learning inversion model using the synthetic sample dataset; input the target gravity anomaly data used for inversion into the trained deep learning inversion model and output the prediction results of the geometric parameters and spatial location of the underground target body.
2. The microgravity identification method for underground space according to claim 1, characterized in that, Also includes: S6: Substitute the geometric parameters predicted by the deep learning inversion model into the forward model to calculate the theoretical gravity anomaly, compare it with the measured or processed gravity anomaly, and calculate the residual; verify the consistency between the predicted geometric parameters and the engineering survey data, and comprehensively evaluate the accuracy and confidence of the prediction results.
3. The microgravity identification method for underground space according to claim 1, characterized in that, The geometric parameters include the radius, burial depth, spacing, and horizontal position of the target.
4. The microgravity identification method for underground space according to claim 3, characterized in that, The steps of performing forward modeling calculations for each combination of geometric parameters using the analytical formula for gravity anomalies specifically include: Forward modeling was performed using the analytical formula for gravity anomalies; ； in, G is the gravitational anomaly value; G is the gravitational constant. h is the center burial depth of the underground cylinder; σ is the density difference between the cylinder and the surrounding rock strata; R is the radius of the cylinder; dx1 is the horizontal distance between the center of the underground left cylinder and the surface measuring point; dx2 is the horizontal distance between the center of the underground right cylinder and the surface measuring point; the gravity anomaly values of multiple observation points are calculated along the preset measuring line to form the gravity anomaly sequence.
5. The microgravity identification method for underground space according to claim 4, characterized in that, The step of generating a synthetic sample dataset consisting of gravity anomaly sequences and corresponding geometric parameters includes: superimposing random Gaussian noise on the gravity anomaly sequences of some samples to improve the robustness of the trained model.
6. The microgravity identification method for underground space according to claim 1, characterized in that, The terrain correction includes building correction and natural terrain correction. Buildings and surface terrain are discretized into multiple rectangular prism units. The gravity anomaly of each terrain unit to the measuring points in the survey area is calculated using the gravity anomaly calculation formula. For buildings of different sizes and natural terrain, calculate the attenuation curve of gravity anomaly difference caused by the nearest and farthest measuring points from the building as a function of distance. When the gravity anomaly difference is <1μGal, the corresponding distance is the cutoff range for terrain correction. Formula for calculating gravity anomalies: ； ； ； ； ； Where G is the gravitational constant, ; It is the density difference between the building and the air; the coordinates of the observation point are... The boundary of a rectangular prism in global coordinates x1 is the relative horizontal distance from the observation point to the left side of the prism; x2 is the relative horizontal distance from the observation point to the right side of the prism; y1 is the relative distance from the observation point to the front edge of the prism; y2 is the relative distance from the observation point to the rear edge of the prism; z1 is the height difference from the observation point to the top of the prism; z2 is the height difference from the observation point to the bottom of the prism. Let be the distances from the observation point to each of the eight vertices of the rectangular prism; Finally, by integrating, the gravity anomaly generated by all rectangular prism elements within the truncation range of the terrain correction on the measuring point is obtained, yielding the terrain correction value TC.
7. The microgravity identification method for underground space according to claim 6, characterized in that, The height correction adopts ;in, This is the height correction value. The height difference between the observation surface and the reference surface; Intermediate layer correction adopted ;in, Here, G is the intermediate layer correction value, and G is the gravitational constant. ρ is the average density of the intermediate layer; t is the thickness of the intermediate layer; The calculation of Bouguer gravity anomaly uses ;in, Based on the basic gravity anomaly, This is the terrain correction value. This is the height correction value. Intermediate layer correction value.
8. The microgravity identification method for underground spaces according to claim 1, characterized in that, The steps of determining the influence range of shallow disturbances, and based on this influence range, removing the shallow disturbance field from the Bouguer gravity anomaly to obtain the target gravity anomaly data for inversion specifically include: The influence of the undulating density interface on the surface gravity anomaly under different burial depths and amplitudes was analyzed to determine the range of potential field separation, which is the range of influence of shallow surface disturbance. Bouguer gravity anomaly Based on this, the influence range of shallow interference is stripped away to obtain the regional field anomaly with shallow interference removed, which is then used for inversion of the target gravity anomaly; specifically including: Setting field anomaly From regional field and local field Composition, namely: ; set up Representatives and Points Distance The average value of a certain 4-point potential field, i.e.: In the formula, The cutting radius is; the region field is and The weighted average, i.e. in, and These are weighting coefficients, and ; ； ； ； ； ； ； ； ； in, The gradient is in the horizontal direction; The gradient is in the vertical direction. The deviation between the center point Z(x,y) and half of the difference in the horizontal direction; The deviation between the center point Z(x,y) and half of the vertical difference; The calculated region field is called the first-order cut region field. Indicate; to Repeat the calculation to obtain the second cutting region. ; continuing this iteration, eventually there will be ； Therefore, there is Through multiple iterations until the regional field changes tend to stabilize, the target gravity anomaly data used for inversion is finally obtained.
9. A microgravity identification method for underground spaces according to any one of claims 1-8, characterized in that, The target body is one or more of the following: tunnel, cavity, or pipeline corridor.
10. A microgravity identification system for underground spaces, characterized in that, include: Acquisition module: used to acquire the measured microgravity raw data of the target area and the engineering geological prior information including the geometric type of the target body, wherein the engineering geological prior information includes geometric parameters; Forward modeling module: used to define key geometric parameters based on the geometric type of the target body, set the variation range of each geometric parameter, and generate multiple sets of geometric parameter combinations by traversing sampling; Forward modeling is performed on each combination of geometric parameters using the analytical formula for gravity anomalies, generating a synthetic sample dataset consisting of gravity anomaly sequences and corresponding geometric parameters; The gravity anomaly calculation module is used to perform tidal correction and zero drift correction on the measured microgravity raw data to obtain the basic gravity anomaly value, determine the terrain correction range of the basic gravity anomaly value and then perform terrain correction, and perform height correction and intermediate layer correction on the basic gravity anomaly value, and calculate the Bouguer gravity anomaly after correction. Stripping module: used to determine the influence range of shallow disturbance, and strip the shallow disturbance field from the Bouguer gravity anomaly based on the influence range of shallow disturbance to obtain the target gravity anomaly data for inversion; Prediction module: used to construct a deep learning inversion model, train the deep learning inversion model using the synthetic sample dataset; input the target gravity anomaly data used for inversion into the trained deep learning inversion model, and output the prediction results of the geometric parameters and spatial location of the underground target body.