Method and device for predicting structure of surface karst zone by fusing multi-source data detection

Abstract

This invention discloses a method and apparatus for predicting the structure of surface karst zones by integrating multi-source data detection. The method includes: Step S1. Acquiring apparent resistivity data detected by high-density electrical resistivity method, electromagnetic wave data detected by ground penetrating radar, and remote sensing data of the surface environment; Step S2. Extracting the spatial distribution of surface environmental factors based on remote sensing images; Step S3. Preliminarily determining the soil and surface karst zone thickness data of the karst area under test based on the apparent resistivity data, electromagnetic wave data, and corresponding spatial distribution data; Step S4. Constructing a training dataset using the apparent resistivity data, electromagnetic wave data, surface environmental factors, and corresponding thickness data, and training the model to obtain a surface karst zone structure prediction model; Step S5. Inputting the real-time acquired data into the surface karst zone structure prediction model to obtain the prediction result. This invention has the advantages of simple implementation, low cost, high prediction accuracy, and high efficiency.

Description

Technical Field

[0001] This invention relates to the field of karst geological exploration technology, and in particular to a method and apparatus for predicting the structure of surface karst zones by integrating multi-source data detection. Background Technology

[0002] The surface karst zone is an important component of karst landforms. Its unique binary hydrogeological structure, high heterogeneity, and anisotropy make quantitative characterization of its structure challenging. As a crucial area for groundwater and surface water exchange, the accurate inversion of the surface karst zone structure is of great significance for understanding karst hydrogeological processes, assessing geological hazard risks, and guiding engineering construction. Traditional geological exploration methods, such as drilling and remote sensing, are insufficient due to the complex dynamic changes of groundwater in the surface karst zone. Traditional methods often suffer from high labor costs, low efficiency, and insufficient accuracy, making it difficult to meet the demand for high-precision and rapid acquisition of surface karst zone structural information.

[0003] Machine learning methods possess powerful data processing and pattern recognition capabilities, automatically learning and extracting hidden features and patterns from massive datasets with little or no human intervention, enabling in-depth data mining and efficient processing. Applying machine learning methods to predict the structure of surface karst zones can significantly improve work efficiency, shorten exploration cycles, and reduce costs. However, the accuracy of machine learning methods depends on precise feature parameter extraction and extensive data training. Since surface karst zones are underground and structurally complex, it is difficult to directly extract feature parameters that effectively characterize their structure. Therefore, existing machine learning algorithms are not suitable for predicting the structure of surface karst zones. Summary of the Invention

[0004] The technical problem to be solved by this invention is: in view of the technical problems existing in the prior art, this invention provides a method and device for predicting the structure of surface karst zones by integrating multi-source data detection, which is simple to implement, low in cost, and has high prediction accuracy and efficiency.

[0005] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows:

[0006] A method for predicting the structure of surface karst zones by integrating multi-source data detection, comprising the following steps:

[0007] Step S1. Multi-source detection data acquisition: Acquire apparent resistivity data of the underground soil layer in the karst area under test by high-density electrical resistivity tomography, acquire electromagnetic wave data of the underground karst zone in the karst area under test by ground penetrating radar, and acquire remote sensing image data of the surface environment of the karst area under test. Based on the apparent resistivity data and electromagnetic wave data, obtain the corresponding spatial distribution data to obtain apparent resistivity spatial distribution data and electromagnetic wave spatial distribution data.

[0008] Step S2. Comprehensive feature extraction: Extract surface environmental factors that affect the structure of the surface karst zone from the remote sensing image data of the above-ground environment;

[0009] Step S3. Preliminary structural identification: Based on the apparent resistivity data, electromagnetic wave data and corresponding spatial distribution data, the layering interface of the soil and rock mass, the interface between the surface karst zone and the bedrock are identified, and the soil thickness data and surface karst zone thickness data of the karst area under test are preliminarily determined.

[0010] Step S4. Prediction Model Construction: The apparent resistivity data, electromagnetic wave data, surface environmental factors, and corresponding soil thickness data and surface karst zone thickness data are used to construct a training dataset. The training dataset is used to train a deep learning model to obtain a surface karst zone structure prediction model, so as to establish a nonlinear mapping relationship between apparent resistivity and soil thickness, and between electromagnetic waves and surface karst zone thickness.

[0011] Step S5. Real-time structural prediction: Acquire real-time data on apparent resistivity, electromagnetic waves, and remote sensing data of the surface environment of the karst area under test, and extract real-time surface environmental factors. Input the real-time acquired apparent resistivity data, electromagnetic waves, and surface environmental factors into the surface karst zone structure prediction model to obtain the predicted results of soil thickness and surface karst zone thickness in the karst area under test. Combine the predicted soil thickness data and surface karst zone thickness data to obtain the surface karst zone structure distribution status.

[0012] Further, in step S1, the high-density electrical resistivity tomography (EDT) uses a Winner device for detection. The A, M, N, and B electrodes of the Winner device are arranged in a straight line with equal spacing, where A and B are power supply electrodes, and M and N are measurement electrodes. By controlling the power supply state of each electrode, the distance between the electrodes is made to satisfy AM = MN = NB = n*x = a during each data acquisition, where AM, MN, and NB are the distances between the electrodes, n is the isolation coefficient, and a is the electrode spacing. When using the Winner device for detection, AM = MN = NB = x, and A, M, N, and B move point by point along the measurement line to obtain the next profile. Then, the distance between two adjacent electrodes is increased by one unit electrode spacing, and A, M, N, and B move point by point along the measurement line again to obtain the next profile. This process is repeated until an inverted trapezoidal cross-section is finally obtained.

[0013] Furthermore, based on the data obtained from high-density electrical resistivity tomography (EDT), the apparent resistivity data of the underground structure in the karst area under test is obtained according to the following formula:

[0014]

[0015] Where ρ is the apparent resistivity, K is the device coefficient, ΔV is the potential difference, I is the supply current, and AN, BM, and BN are the distances between the electrodes, respectively.

[0016] The electromagnetic wave data of the underground structure of the karst area being measured is obtained from the data obtained by ground penetrating radar according to the following formula:

[0017]

[0018] Where h represents the detection depth, v represents the propagation speed of electromagnetic waves, t represents the time required for the radar electromagnetic waves to travel from the transmitting antenna to the receiving antenna, and x represents the distance between the transmitting antenna and the receiving antenna.

[0019] Further, in step S2, extracting surface environmental factors affecting the structure of the surface karst zone from the remote sensing data of the above-ground environment includes:

[0020] Step S201. Select multiple environmental factors that affect the structure of the surface karst zone to construct an environmental factor dataset. The environmental factors include any combination of topographic factors, soil factors, rock factors, and vegetation factors.

[0021] Step S202. Screen each environmental factor in the environmental factor dataset to select the final controlling factor and use it as the surface environmental factor.

[0022] Further, in step S202, the step of filtering each environmental factor in the environmental factor dataset includes:

[0023] Step S221. Preprocess the collected environmental factor dataset to obtain the preprocessed independent variable matrix and dependent variable matrix, wherein the independent variables are environmental factors and the dependent variables are soil thickness and surface karst zone thickness.

[0024] Step S222. Extract principal components from the independent variable and the dependent variable;

[0025] Step S223. Calculate the VIP value of each principal component, and select the principal components with VIP values ​​greater than a preset threshold as the principal control factors. The formula for calculating the VIP value is as follows:

[0026]

[0027] In the formula, p is the number of independent variables, m is the number of principal components of the independent variables, and R0 is the number of principal components of the independent variables. yh It is the principal component t h For Y * Explanatory power, R y (cum) represents the relationship between each principal component and the dependent variable Y. * The cumulative explanatory power, W hj It is X * With principal component t h The correlation coefficient.

[0028] Furthermore, the topographic factors include any one or more of the following: elevation, aspect, slope, slope position, slope length factor, elevation variation coefficient, topographic relief, surface roughness, slope variability, topographic humidity index, and surface cutting depth; the soil factors include any one or more corresponding indicators of soil temperature, soil moisture, soil type, soil porosity, soil texture, soil mineral composition, and soil organic matter; the rock factors include rock exposure rate index and lithology index, wherein the rock exposure rate index is the percentage of the area of ​​exposed surface rock in the measured area to the total area of ​​the area; the vegetation factors include any one or more of the following: vegetation type, vegetation coverage, and vegetation index.

[0029] Furthermore, in step S3, the surface karst zone model based on the electromagnetic wave model is divided into two layers according to the spatial distribution of electromagnetic wave data, namely, the upper part is the surface karst zone and the lower part is the bedrock; the surface karst zone model based on the resistivity model adopts a two-layer model, namely, the upper part is soil and the lower part is bedrock, and the inflection point is determined when the increase or decrease in the average resistivity of the first layer and the second layer exceeds the preset threshold. The area above the determined inflection point is regarded as the soil layer and the area below the inflection point is regarded as the bedrock layer.

[0030] Further, step S4 includes:

[0031] Step S401. A first dataset is constructed from the apparent resistivity data, surface environmental factors and corresponding soil thickness, and a second dataset is constructed from the electromagnetic wave data, surface environmental factors and corresponding surface karst zone thickness.

[0032] Step S402. Train machine learning models using the first dataset and the second dataset respectively to obtain the first model and the second model. The first model takes apparent resistivity data and surface environmental factors as input and outputs the predicted soil thickness. The second model takes electromagnetic wave data and surface environmental factor data as input and outputs the predicted thickness of the surface karst zone.

[0033] Step S403. The first and second models obtained from the training are continuously adjusted and optimized until they meet the preset performance requirements, and finally the nonlinear mapping relationship between apparent resistivity and soil thickness, and between electromagnetic waves and the thickness of the surface karst zone is established.

[0034] A surface karst zone structure prediction device integrating multi-source data detection includes a processor and a memory, wherein the memory is used to store a computer program and the processor is used to execute the computer program to perform the method described above.

[0035] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0036] Compared with existing technologies, the advantages of this invention are as follows: This invention achieves surface karst zone structure prediction by combining physical detection methods with machine learning methods. First, it uses ground-penetrating radar and high-density electrical resistivity tomography (EDS) to detect the underground structure of the surface karst zone, acquiring apparent resistivity data of the underground soil layer and electromagnetic wave data of the underground karst zone. Simultaneously, it extracts surface environmental factors from remote sensing data of the surface environment of the karst region. By fusing apparent resistivity data, electromagnetic wave data, and surface environmental factors, a surface karst zone structure prediction model is trained based on a deep learning model. This model can fully explore the underground and surface environmental characteristics of the surface karst zone structure, achieving rapid and high-precision inversion of the surface karst zone structure. It also greatly improves the accuracy of data interpretation, significantly enhances prediction efficiency, and reduces the required cost. Attached Figure Description

[0037] Figure 1 This is a schematic diagram illustrating the implementation process of the surface karst zone structure prediction method that integrates multi-source data detection in this embodiment.

[0038] Figure 2 This is a schematic diagram illustrating the principle of multi-source data detection of the surface karst zone structure in this embodiment.

[0039] Figure 3This is a schematic diagram illustrating the arrangement principle of the Wenner device when using high-density electrical resistivity tomography in this embodiment.

[0040] Figure 4 This is a schematic diagram illustrating the implementation process of filtering the detection radar data in this embodiment. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0042] Both high-density electrical resistivity tomography (EDS) and ground-penetrating radar (GPR) can detect underground structures. EDS, through electrode arrays and current injection, measures the distribution of the underground electric field, interprets it using mathematical models, and provides high-resolution resistivity data, thus revealing details of the underground structure. GPR, by emitting high-frequency electromagnetic waves and receiving signals reflected back from the underground medium, acquires radar image data of the underground medium, revealing its distribution. This invention comprehensively utilizes the physical detection methods of GPR and EDS to detect the underground structure of the surface karst zone, acquiring apparent resistivity data of the underground soil layer and electromagnetic wave data of the underground karst zone, thereby obtaining corresponding resistivity distribution data of the soil layer and electromagnetic wave distribution data of the underground karst zone.

[0043] Meanwhile, considering that the structural state of the surface karst zone is influenced by the surface environment—such as topography, parent material, climate, and vegetation growth—it is closely related to the development of the surface karst zone. Therefore, by analyzing remote sensing data of the surface environment, environmental factors related to the structure of the surface karst zone can be identified. Furthermore, by combining apparent resistivity and electromagnetic wave data obtained from high-density electrical resistivity and ground-penetrating radar (GPR), the underground and above-ground environmental characteristics of the surface karst zone structure can be comprehensively obtained. However, the aforementioned data is often massive and exceptionally complex. Traditional data analysis methods are often time-consuming and labor-intensive to process, and are easily affected by human factors, leading to subjectivity and uncertainty in the interpretation results. This invention applies machine learning methods to process and analyze the aforementioned geophysical data, combining physical detection methods with machine learning to predict the structure of the surface karst zone. This fully explores the underground and above-ground environmental characteristics of the surface karst zone structure, achieving rapid and high-precision inversion of the surface karst zone structure, greatly improving the accuracy of data interpretation, and significantly increasing prediction efficiency while reducing costs.

[0044] like Figure 1 As shown, the steps of the surface karst zone structure prediction method integrating multi-source data detection in this embodiment include:

[0045] Step S1. Multi-source detection data acquisition: Acquire apparent resistivity data of the underground soil layer in the karst area under test by high-density electrical resistivity tomography, acquire electromagnetic wave data of the underground karst zone in the karst area under test by ground penetrating radar, and acquire remote sensing data of the surface environment of the karst area under test. Based on the apparent resistivity data and electromagnetic wave data, obtain the corresponding spatial distribution data to obtain the spatial distribution data of apparent resistivity and the spatial distribution data of electromagnetic waves.

[0046] like Figure 2 As shown, after planning the field measurement area in the surface karst zone, high-density electrical resistivity method is used to obtain resistivity data of shallow soil layers. Ground penetrating radar is used to obtain electromagnetic wave data of deeper underground karst zones by emitting high-frequency electromagnetic waves and receiving signals reflected back from the underground medium. At the same time, remote sensing data collected from the surface layer is also obtained. This allows for comprehensive detection of underground soil layers, underground karst zones, and the surface environment.

[0047] In specific application examples, when using high-density electrical resistivity tomography (ED tomography), a comprehensive preliminary survey of the study area can be conducted based on remote sensing imagery and historical data to determine the actual measurement area. A high-density resistivity instrument is then used to deploy an electrode array within the measurement area. Through automatic point placement, automatic electrode rotation, automatic power supply, and automatic observation, the apparent resistivity distribution data of the subsurface is acquired. When laying out the survey lines, the differences in horizontal and vertical scales can be comprehensively considered to fully account for the survey line resolution. After the high-density EDM measurement is completed, a subsurface apparent resistivity distribution map is obtained. The coordinate information of each sampling point is recorded using the measuring equipment. The collected high-density EDM data undergoes preprocessing operations such as noise reduction and filtering to improve data quality.

[0048] Taking the high-density electrical resistivity method using a SYSCAL resistivity meter as an example, the high-density electrical resistivity parameters are initialized during field measurement, including the size of the measuring point spacing and the type of electrode device. The spacing of the high-density electrical resistivity measuring lines can be set from 0 to 2m. Considering that the smaller the spacing, the more detailed the underground structure is detected, in order to obtain more underground structure information and considering the amount of data, this embodiment sets the measuring line spacing to 0.5m, which can detect an underground depth of 6m.

[0049] There are many types of electrode devices for high-density electrical resistivity tomography (EDT). Based on the spatial relationship between the power supply electrode and the measuring electrode, they can be divided into Wenner, dipole-dipole, monopole-dipole, and electrode-electrode devices. In this embodiment, considering the underground structure of the surface karst zone, a Wenner device is used for detection. Figure 3As shown, the electrodes A, M, N, and B of the Winner device are arranged in a straight line with equal spacing. A and B are power supply electrodes, and M and N are measurement electrodes. By controlling the power supply status of each electrode, the distance between the electrodes is made such that AM = MN = NB = n*x = a during each data acquisition, where AM, MN, and NB are the distances between the electrodes, n is the isolation coefficient, and a is the electrode spacing. When using the Winner device for detection, AM = MN = NB = x is set. A, M, N, and B are moved point by point along the measurement line to obtain the next profile. Then, the distance between two adjacent electrodes is increased by one unit electrode spacing, and A, M, N, and B are moved point by point along the measurement line again to obtain the next profile. The above steps are repeated until an inverted trapezoidal cross-section is finally obtained.

[0050] In this embodiment, the apparent resistivity data of the underground structure in the karst area under test can be obtained from the data obtained by high-density electrical resistivity tomography (EDT) according to the following formula:

[0051]

[0052] Where ρ is the apparent resistivity, K is the device coefficient, ΔV is the potential difference, I is the supply current, and AN, BM, and BN are the distances between the electrodes.

[0053] After obtaining the apparent resistivity data of the subsurface soil layer in the karst area using high-density electrical resistivity tomography (EDT), a ground-penetrating radar (GPR) system is used along the existing EDT line. This system transmits high-frequency electromagnetic waves and receives the signals reflected back from the subsurface medium to acquire electromagnetic wave data. For example, the MALA ProEx 3D ground-penetrating radar instrument can be used. The GPR parameters are set as follows: a 50MHz unshielded antenna is selected, point-based data acquisition is used, and to match the EDT data, the sampling interval is set to 0.5m, the sampling frequency is ~500~800MHz, the sampling window is 1000ns, the number of data stacks per channel is 128, and the detection depth is 40m. After the GPR detection is completed, the acquired electromagnetic wave data is filtered. Figure 4 As shown, using Ground Vision data processing software, one-dimensional filtering (to remove DC drift) is used to remove zero-point drift in the data. Then, static correction (movement start time) and gain (energy attenuation) are selected to amplify the deep signal. Two-dimensional filtering (to extract average channels) is used to remove the horizontal portion of the image. Next, one-dimensional filtering / Butterworth bandpass filtering is performed according to the frequency selected based on the antenna used. Finally, two-dimensional filtering / moving average is used to suppress noise, smooth the image, and obtain the electromagnetic wave distribution map of underground radar through layer tracking.

[0054] In this embodiment, the electromagnetic wave data of the underground structure of the karst area under test can be obtained from the ground-penetrating radar detection data according to the following formula:

[0055]

[0056] Where h represents the detection depth, v represents the propagation speed of electromagnetic waves, t represents the time required for the radar electromagnetic waves to travel from the transmitting antenna to the receiving antenna, and x represents the distance between the transmitting antenna and the receiving antenna.

[0057] In specific application embodiments, GPS instruments can also be used to record the location information of the survey line and the location information corresponding to each measuring point, so as to better locate the target survey line. When selecting the survey line, the longer the survey line, the better, based on the land use conditions being investigated, to ensure that more underground structure information can be obtained. In order to obtain relatively stable resistivity and electromagnetic wave values, the survey line can also be laid out under long periods of sunny weather, depending on the weather conditions, to avoid misjudgment of soil resistivity due to increased soil moisture content after precipitation.

[0058] In a specific application embodiment, the preprocessed apparent resistivity data and electromagnetic wave data are analyzed using Res2dinv and Reflexw software respectively, and underground apparent resistivity distribution maps and underground radar images are obtained.

[0059] Step S2. Comprehensive Feature Extraction: Based on the apparent resistivity data and electromagnetic wave data, obtain the corresponding spatial distribution data to obtain the spatial distribution data of apparent resistivity and electromagnetic wave. Extract the surface environmental factors that affect the structure of the surface karst zone from the remote sensing data of the ground environment.

[0060] Topography, parent material, climate, and vegetation growth in karst regions are all closely related to the development of surface karst zones. This study selects environmental factors such as topography, soil, rock, and vegetation based on remote sensing image data to explore their impact on the structure of surface karst zones. By combining multi-source remote sensing data such as DEM, vegetation index, and land use status, environmental factors that may affect the structure of surface karst zones, including geology, topography, and vegetation, are collected to construct an environmental factor database. Data fusion methods are then used to extract the multi-dimensional spatial distribution characteristics of above-ground and underground factors affecting the structure of surface karst zones.

[0061] In this embodiment, the following steps can be used to extract surface environmental factors affecting the structure of the surface karst zone from remote sensing data of the ground environment:

[0062] Step S201. Select multiple environmental factors that affect the structure of the surface karst zone to construct an environmental factor dataset. The environmental factors include topographic factors, soil factors, rock factors, vegetation factors, etc.

[0063] Topographic factors can be extracted from remote sensing data using ArcGIS 10.7 software and the reverse extraction method. These factors include elevation, aspect, slope, slope position, slope length factor, elevation variation coefficient, topographic relief, surface roughness, slope variability, topographic moisture index, and surface incision depth. Soil factors can be obtained through long-term station monitoring and indoor experiments. Station monitoring indicators can include soil temperature and soil moisture, while indoor sampling and analysis indicators can include soil type, soil porosity, soil texture, soil mineral composition, and soil organic matter. Rock indicators mainly include lithology and rock exposure rate. The rock exposure rate is the percentage of surface rock exposed in a region relative to the total area. This value is the most direct manifestation of rocky desertification and is an important factor leading to its formation and change. In other words, the rock exposure rate is also closely related to the structure of the surface karst zone.

[0064] The rock exposure rate can be calculated using a pixel-based binary model based on the Normalized Difference Rock Index (NDRI). For example, the formula for calculating NDRI can be expressed as:

[0065] NDRI=(TM7-TM4) / (TM7+TM4) (4)

[0066] f r =(NDRI-NDRI0) / (NDRI) r -NDRI0) (5)

[0067] Where TM7 is the reflectance value in the shortwave infrared band, TM4 is the reflectance value in the near-infrared band, and f r NDRI is the rock exposure rate, which is information acquired by the sensor. r NDRI is the NDRI value when the rock is entirely composed of bare rock, while NDRI0 is the NDRI value when there is no bare rock at all.

[0068] Vegetation factors include vegetation type and vegetation cover. Vegetation cover can be estimated using a pixel-based binary model based on the Normalized Difference Vegetation Index (NDVI). NDVI is derived from Landsat 8TM remote sensing imagery. The annual average NDVI is calculated using the annual remote sensing impact over a specified time period, and then normalized. The formulas for calculating vegetation index and vegetation cover can be expressed as follows:

[0069]

[0070] fc c =(NDVI-NDVI) soil) / (NDVI veg -NDVI soil (7)

[0071] Where NIR is the reflectance value in the near-infrared band, and R is the reflectance value in the red band; NDVI soil NDVI is the NDVI value of the soil portion. veg f represents the NDVI value of the vegetation portion. c This refers to vegetation coverage.

[0072] Step S202. Screen each environmental factor in the environmental factor dataset to select the final controlling factor and use it as the surface environmental factor.

[0073] In this embodiment, VIP (Variable Importance in Projection) analysis can be used to screen the environmental factors in the environmental factor dataset. VIP analysis is a variable screening method based on partial least squares. The VIP value can effectively identify the predictor variable that best explains the dependent variable among all independent variables. For example, environmental factors with high VIP values ​​are considered to be the main control factors, and their impact on the surface environment is the most significant.

[0074] In this embodiment, the steps of using VIP importance analysis to screen environmental factors in the environmental factor dataset include:

[0075] Step S221. Data preprocessing: The collected environmental factor dataset is preprocessed to obtain the preprocessed independent variable matrix and dependent variable matrix, where the independent variables are environmental factors and the dependent variables are soil thickness and surface karst zone thickness.

[0076] Specifically, preprocessing includes data cleaning (removing duplicates, invalid or outliers), missing value handling (such as interpolation or deleting severely missing samples), and standardization (such as normalization or standardization to eliminate the influence of different units and numerical ranges on the model). This is achieved by processing the independent variable (environmental factor) matrix X = (x... ij The matrix Y = (n×k) and dependent variables (soil thickness, surface karst zone thickness) are given by: i The n×1 matrix is ​​standardized to obtain the standardized variable matrix X. * and Y * .

[0077] Step S222. From the independent variable X * (Environmental factors) and dependent variable Y * The main component t was extracted from (soil thickness, surface karst zone thickness). h .

[0078] Specifically, principal components (t=1,...h) can be extracted from the independent and dependent variables using algorithms such as PLS (Partial Least Squares Regression). These principal components can carry the variation information in their respective datasets to the maximum extent, and the correlation between the principal components is maximized.

[0079] Step S223. Calculate the VIP value of each principal component, and select the principal components with VIP values ​​greater than a preset threshold as the principal control factors. The formula for calculating the VIP value is as follows:

[0080]

[0081] In the formula, p is the number of independent variables, m is the number of principal components of the independent variables, and R0 is the number of principal components of the independent variables. yh It is the principal component t h For Y * Explanatory power, R y (cum) represents the relationship between each principal component and the dependent variable Y. * The cumulative explanatory power, W hj It is X * With principal component t h The correlation coefficient.

[0082] In specific application examples, it can be assumed that when VIP>1, the independent variable is significantly correlated with the dependent variable, and the principal component with VIP>1 is taken as the principal control factor. When VIP is small, it means that the variable contributes very little to the model and can be considered for removal to improve the stability of the model.

[0083] This embodiment uses the factor importance calculated by the partial least squares model and the correlation matrix table of all explanatory variables to further screen environmental factors, which can effectively screen out the main surface environmental factors that can effectively characterize the structure of the surface karst zone.

[0084] Step S3. Preliminary structural identification: Based on apparent resistivity data, electromagnetic wave data and corresponding spatial distribution data, identify the layered interfaces of the soil and rock mass, the interface between the surface karst zone and the bedrock, and preliminarily determine the soil thickness data and the surface karst zone thickness data of the karst area being measured.

[0085] In specific application examples, data analysis tools such as MATLAB can be used to analyze the preprocessed apparent resistivity data and electromagnetic wave data. Combined with prior geological knowledge, the layering interface of the soil and rock mass, the interface between the surface karst zone and the bedrock can be preliminarily identified to determine the soil thickness range. Then, the thickness range of the surface karst zone can be determined by combining the radar image data of the underground medium.

[0086] In order to calculate the soil thickness distribution in the measured area, this embodiment defines the layer interface of the soil and rock mass. That is, the surface karst zone based on the resistivity model is only considered as a simple two-layer model. In this model, the upper part is soil and the lower part is bedrock. The first inflection point is determined only when the average resistivity of the first layer (soil) and the second layer (bedrock) increases or decreases significantly to a preset threshold. The area above the inflection point is regarded as the soil layer and the area below the inflection point is regarded as the bedrock layer.

[0087] Step S4. Prediction Model Construction: The apparent resistivity data, electromagnetic wave data, surface environmental factors, and corresponding soil thickness data and surface karst zone thickness data are used to construct a training dataset. The deep learning model is trained using the training dataset to obtain a surface karst zone structure prediction model, so as to establish a nonlinear mapping relationship between apparent resistivity and soil thickness, and between electromagnetic waves and surface karst zone thickness.

[0088] By using preprocessed apparent resistivity data, electromagnetic wave data, and extracted environmental factors as input, a surface karst zone structure prediction model is obtained through model training and validation. This model can quickly process complex geological data and establish nonlinear mapping relationships between apparent resistivity and soil thickness, and between electromagnetic waves and surface karst zone thickness.

[0089] In this embodiment, the construction of the surface karst zone structure prediction model can be carried out through the following steps:

[0090] Step S301. A first dataset is constructed from apparent resistivity data, surface environmental factors and corresponding soil thickness, and a second dataset is constructed from electromagnetic wave data, surface environmental factors and corresponding surface karst zone thickness.

[0091] Step S302. Train machine learning models using the first dataset and the second dataset respectively to obtain the first model and the second model. The first model takes apparent resistivity data and surface environmental factors as input and outputs the predicted soil thickness. The second model takes electromagnetic wave data and surface environmental factor data as input and outputs the predicted thickness of the surface karst zone.

[0092] Step S303. The first and second models obtained from the training are continuously adjusted and optimized until they meet the preset performance requirements, and finally the nonlinear mapping relationship between apparent resistivity and soil thickness, and between electromagnetic waves and the thickness of the surface karst zone is established.

[0093] Specifically, a first dataset can be constructed using apparent resistivity, aboveground environmental factors, and corresponding soil thickness, and a second dataset can be constructed using electromagnetic waves, aboveground environmental factors, and corresponding surface karst zone thickness. The constructed datasets are then used to train the machine learning model. By adjusting the model parameters, the prediction performance is optimized, and a nonlinear mapping relationship is established between apparent resistivity and soil thickness, and between electromagnetic waves and surface karst zone thickness.

[0094] Machine learning models can specifically employ Artificial Neural Networks (ANNs) or Support Vector Machines (SVMs). ANNs, which simulate the mechanisms of biological neural networks and incorporate their behavioral characteristics for information processing, possess highly high parallel processing capabilities, self-organization, nonlinear processing capabilities, and adaptability. This embodiment establishes a typical BPA algorithm-based three-layer feedforward neural network model to predict the structure of surface karst zones. The input vector consists of surface environmental factors selected by a factor selection model, apparent resistivity data, and electromagnetic wave data. The output layer uses one neuron to predict the surface karst zone structure. Furthermore, by determining the ratio of input and output data and selecting the input and output variables, the neural network structure can be further determined. The optimal number of hidden neurons (N) can be determined through trial and error, and then combined with model accuracy verification methods to select the number of hidden neurons, ultimately yielding the best-performing neural network model.

[0095] Support Vector Machines (SVMs) are based on the principles of linear separability and minimum structural risk. For linear cases, SVM function fitting first considers using a linear regression function f(x). When linearly inseparable, it can transform low-dimensional linearly inseparable models into high-dimensional linearly separable ones. Simultaneously, it seeks the optimal solution between model complexity and learning ability based on limited sample information, ensuring the final model has good generalization ability. The basic idea of ​​nonlinear SVM is to map the input vector to a high-dimensional feature space (Hilbert space) through a pre-determined nonlinear mapping, and then perform linear regression in this high-dimensional space, thus achieving the effect of nonlinear regression in the original space. The expression for the nonlinear fitting function is:

[0096]

[0097] In specific application embodiments, by selecting suitable machine learning algorithms for processing complex geological data, such as the aforementioned Support Vector Machine (SVM) and Artificial Neural Networks (ANNs), the first model takes preprocessed apparent resistivity data and above-ground environmental factor data as input and outputs the predicted soil thickness. The second model takes preprocessed electromagnetic wave data and above-ground environmental factor data as input and outputs the predicted thickness of the surface karst zone. The first and second models are combined to form a surface karst zone structure prediction model. Furthermore, the prediction results can be cross-validated with actual measurement results, and the model parameters can be continuously adjusted and optimized to ensure the applicability of the model in unknown areas and improve the model's generalization ability and prediction accuracy.

[0098] Step S4. Real-time structural prediction: Acquire real-time data on apparent resistivity, electromagnetic waves, and remote sensing data of the surface environment of the karst area under test, extract real-time surface environmental factors, and input the real-time acquired apparent resistivity data, electromagnetic waves, and surface environmental factors into the surface karst zone structure prediction model to obtain the predicted results of soil thickness and surface karst zone thickness in the karst area under test.

[0099] By inputting the real-time detected apparent resistivity data, electromagnetic wave data, and surface environmental factors of the karst area under test into the trained surface karst zone structure prediction model, the soil thickness and surface karst zone thickness of the karst area under test can be predicted, realizing rapid and high-precision inversion of the surface karst zone structure.

[0100] Furthermore, the predicted soil thickness data can be combined with the surface karst zone thickness data to obtain the surface karst zone structure distribution status.

[0101] In specific application examples, the trained surface karst zone structure prediction model is applied to high-density electrical resistivity tomography (EDT) data, ground-penetrating radar (GPR) data, and above-ground environmental factor data in the actual exploration area. The model inference process is executed to obtain the predicted results of soil thickness and surface karst zone thickness in the area. Based on the prediction results, the soil thickness and surface karst zone thickness in the karst area are accurately interpreted. The Surfer software is used to generate soil thickness distribution maps, surface karst zone thickness distribution maps, and surface karst zone structure profile maps and perform geological interpretation. This model can be flexibly applied to various application scenarios such as engineering construction and geological disaster early warning.

[0102] like Figure 2As shown, this embodiment first conducts geophysical data acquisition, selecting a test area from the study area, and using high-density electrical resistivity and ground-penetrating radar to detect the underground structure of the surface karst zone, obtaining apparent resistivity and electromagnetic wave data. Then, feature extraction is performed, environmental factors are selected, and a dataset of surface karst zone influencing factors is established through surface environmental surveys and remote sensing data processing. The influencing factors are then determined using a master control factor screening model. Next, surface karst zone structure identification is performed. Based on the preprocessed apparent resistivity and electromagnetic wave data and their spatial distribution, soil thickness data and surface karst zone thickness data are initially identified. Finally, the preprocessed apparent resistivity data, electromagnetic wave data, and surface environmental factors are used as inputs to train... By training and validating machine learning models to form predictive models, nonlinear mapping relationships are established between apparent resistivity and soil thickness, and between electromagnetic waves and the thickness of the surface karst zone. Finally, the trained machine learning model is applied to actual exploration data to generate spatial distribution maps of soil thickness and surface karst zone thickness in the target area. This approach integrates physical detection methods and machine learning methods to achieve rapid and high-precision inversion of the surface karst zone. It can enhance the depth and breadth of multi-source exploration data processing, fully integrate multi-source exploration data to fully explore the characteristics of the surface karst zone structure, thereby significantly improving the accuracy and efficiency of surface karst zone prediction. Ultimately, this enables geological research, engineering construction, and disaster early warning in karst areas.

[0103] This embodiment further provides a surface karst zone structure prediction device that integrates multi-source data detection, including a processor and a memory. The memory is used to store computer programs, and the processor is used to execute the computer programs to perform the methods described above.

[0104] It is understood that the method described in this embodiment can be executed by a single device, such as a computer or server, or it can be applied to a distributed scenario where multiple devices cooperate to complete the task. In a distributed scenario, one of the multiple devices may execute only one or more steps of the method described in this embodiment, and the multiple devices interact to complete the method. The processor can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the method described in this embodiment. The memory can be implemented using read-only memory (ROM), random access memory (RAM), static storage devices, and dynamic storage devices. The memory can store the operating system and other applications. When the method described in this embodiment is implemented through software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0105] This embodiment further provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0106] Those skilled in the art will understand that the above embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create an implementation for the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0107] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A method for predicting the structure of surface karst zones by integrating multi-source data detection, characterized in that the steps include... include: Step S1. Multi-source detection data acquisition: Acquire apparent resistivity data of the underground soil layer in the karst area under test by high-density electrical resistivity tomography, acquire electromagnetic wave data of the underground karst zone in the karst area under test by ground penetrating radar, and acquire remote sensing image data of the surface environment of the karst area under test. Obtain the corresponding spatial distribution data based on the apparent resistivity data and electromagnetic wave data respectively to obtain apparent resistivity spatial distribution data and electromagnetic wave spatial distribution data. Step S2. Comprehensive feature extraction: Extract surface environmental factors that affect the structure of the surface karst zone from the remote sensing image data of the above-ground environment; Step S3. Preliminary structural identification: Based on the apparent resistivity data, electromagnetic wave data and corresponding spatial distribution data, the layering interface of the soil and rock mass, the interface between the surface karst zone and the bedrock are identified, and the soil thickness data and surface karst zone thickness data of the karst area under test are preliminarily determined. Step S4. Prediction Model Construction: The apparent resistivity data, electromagnetic wave data, surface environmental factors, and corresponding soil thickness data and surface karst zone thickness data are used to construct a training dataset. The training dataset is used to train a deep learning model to obtain a surface karst zone structure prediction model, so as to establish a nonlinear mapping relationship between apparent resistivity and soil thickness, and between electromagnetic waves and surface karst zone thickness. Step S5. Real-time structural prediction: Acquire real-time data on apparent resistivity, electromagnetic waves, and remote sensing data of the surface environment of the karst area under test, and extract real-time surface environmental factors. Input the real-time acquired apparent resistivity data, electromagnetic waves, and surface environmental factors into the surface karst zone structure prediction model to obtain the predicted results of soil thickness and surface karst zone thickness in the karst area under test. Combine the predicted soil thickness data and surface karst zone thickness data to obtain the surface karst zone structure distribution status.

2. The method for predicting the structure of surface karst zones by integrating multi-source data detection according to claim 1, characterized in that, In step S1, the high-density electrical resistivity tomography (EDT) uses a Winner device for detection. The A, M, N, and B electrodes of the Winner device are arranged in a straight line with equal spacing. A and B are power supply electrodes, and M and N are measurement electrodes. By controlling the power supply state of each electrode, the distance between the electrodes is made such that AM=MN=NB=n*x=a during each data acquisition, where AM, MN, and NB are the distances between the electrodes, n is the isolation coefficient, and a is the electrode spacing. When using the Winner device for detection, AM=MN=NB=x is set, and A, M, N, and B move point by point along the measurement line to obtain the next profile. Then, the distance between two adjacent electrodes is increased by one unit electrode spacing, and A, M, N, and B move point by point along the measurement line again to obtain the next profile. This process is repeated until an inverted trapezoidal cross-section is finally obtained.

3. The method for predicting the structure of surface karst zones by integrating multi-source data detection according to claim 2, characterized in that, Based on the data obtained from high-density electrical resistivity resistivity measurements, the apparent resistivity data of the underground structure in the karst area was obtained using the following formula: ， ， Where ρ is the apparent resistivity, K is the device coefficient, ΔV is the potential difference, and I is the supply current. AN , BM , BN These are the distances between the electrodes; The electromagnetic wave data of the underground structure of the karst area being measured is obtained from the data obtained by ground penetrating radar according to the following formula: ， in, h Indicates the depth of detection. v Indicates the speed of propagation of electromagnetic waves. t This indicates the time required for radar electromagnetic waves to travel from the transmitted antenna to the receiving antenna. x This indicates the distance between the transmitting antenna and the receiving antenna.

4. The method for predicting the structure of surface karst zones based on multi-source data detection according to claim 1, characterized in that, Step S2 includes: Step S201. Select multiple environmental factors that affect the structure of the surface karst zone to construct an environmental factor dataset. The environmental factors include any combination of topographic factors, soil factors, rock factors, and vegetation factors. Step S202. Screen each environmental factor in the environmental factor dataset to select the final controlling factor and use it as the surface environmental factor.

5. The method for predicting the structure of surface karst zones by integrating multi-source data detection according to claim 4, characterized in that, Step S202, the step of filtering each environmental factor in the environmental factor dataset, includes: Step S221. Preprocess the collected environmental factor dataset to obtain the preprocessed independent variable matrix and dependent variable matrix, wherein the independent variables are environmental factors and the dependent variables are soil thickness and surface karst zone thickness. Step S222. Extract principal components from the independent variable and the dependent variable; Step S223. Calculate the VIP value of each principal component, and select the principal components with VIP values ​​greater than a preset threshold as the principal control factors. The formula for calculating the VIP value is as follows: ， In the formula, p It is the number of independent variables. m It is the number of principal components of the independent variable. R yh It is a principal component t h right Y * Explanatory power, R y (cum) The principal components are the relationship between the dependent variable and the principal components. Y * The cumulative amount of explanation, W hj It is the independent variable X * With principal components t h The correlation coefficient.

6. The method for predicting the structure of surface karst zones based on multi-source data detection according to claim 4, characterized in that, The topographic factors include any one or more of the following: elevation, aspect, slope, slope position, slope length factor, elevation variation coefficient, topographic relief, surface roughness, slope variability, topographic humidity index, and surface incision depth; the soil factors include any one or more corresponding indicators of soil temperature, soil moisture, soil type, soil porosity, soil texture, soil mineral composition, and soil organic matter; the rock factors include rock exposure rate and lithology indicators, wherein the rock exposure rate is the percentage of the area of ​​exposed rock in the measured area to the total area of ​​the area; the vegetation factors include any one or more of the following: vegetation type, vegetation coverage, and vegetation index.

7. The method for predicting the structure of surface karst zones by integrating multi-source data detection according to any one of claims 1 to 5, characterized in that, In step S3, the surface karst zone model based on the electromagnetic wave model is divided into two layers according to the spatial distribution of electromagnetic wave data, namely, the upper part is the surface karst zone and the lower part is the bedrock; the surface karst zone model based on the resistivity model adopts a two-layer model, namely, the upper part is soil and the lower part is bedrock. When the average resistivity of the first layer and the second layer increases or decreases by more than a preset threshold, the inflection point is determined. The area above the determined inflection point is regarded as the soil layer and the area below the inflection point is regarded as the bedrock layer.

8. The method for predicting the structure of surface karst zones by integrating multi-source data detection according to any one of claims 1 to 5, characterized in that, Step S4 includes: Step S401. A first dataset is constructed from the apparent resistivity data, surface environmental factors and corresponding soil thickness, and a second dataset is constructed from the electromagnetic wave data, surface environmental factors and corresponding surface karst zone thickness. Step S402. Train machine learning models using the first dataset and the second dataset respectively to obtain the first model and the second model. The first model takes apparent resistivity data and surface environmental factors as input and outputs the predicted soil thickness. The second model takes electromagnetic wave data and surface environmental factor data as input and outputs the predicted thickness of the surface karst zone. Step S403. The first and second models obtained from the training are continuously adjusted and optimized until they meet the preset performance requirements, and finally the nonlinear mapping relationship between apparent resistivity and soil thickness, and between electromagnetic waves and the thickness of the surface karst zone is established.

9. A device for predicting the structure of surface karst zones by integrating multi-source data detection, comprising a processor and a memory, wherein the memory is used to store computer programs, characterized in that, The processor is used to execute the computer program to perform the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images