Ground subsidence area permeable channel identification and cause analysis method based on high-density electrical method

Abstract

The invention discloses a land subsidence area permeable channel identification and cause analysis method based on a high-density electrical method, which overcomes the limitation of traditional single profile data interpretation by arranging a three-dimensional observation grid and fusing and collecting multi-source data. Through combination of finite element three-dimensional inversion and the improved convolutional neural network model, accurate description of the three-dimensional spatial morphology, the extension direction and the connectivity of the permeable channel is realized, and the accuracy and the spatial resolution of low-resistance abnormal region identification are improved. Compared with the prior art, the method has remarkable advantages in the aspects of permeable channel recognition accuracy, cause analysis comprehensiveness, result application practicability and the like, and effective support can be provided for long-acting prevention and control work of geological disasters in land subsidence areas.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, specifically to a method for identifying and analyzing the formation of water-permeable channels in subsidence areas based on high-density electrical resistivity tomography. Background Technology

[0002] Ground subsidence refers to the localized subsidence of the earth's surface caused by natural or human factors. Ground subsidence is the collapse or sinking of the ground, which is a localized reduction in elevation of the earth's crust caused by the movement of underground materials or the consolidation and compression of strata. Ground subsidence can lead to safety hazards such as damage to buildings and road collapses, and the problem is particularly prominent in densely populated areas such as the North China Plain and the Yangtze River Delta region.

[0003] Existing methods for identifying and analyzing the genesis of permeable channels in subsidence areas suffer from insufficient data resolution and low fusion of multi-source information. Furthermore, some methods rely solely on high-density electrical resistivity tomography (EDS) data from a single profile, making it difficult to accurately depict the three-dimensional spatial morphology and extension patterns of the permeable channels. Simultaneously, the causal analysis often neglects the synergistic effects of groundwater dynamics, stratigraphic lithology, and human engineering activities, leading to biased causal assessments. Moreover, traditional data processing algorithms are susceptible to interference noise in identifying low-resistivity anomalies, resulting in false positives or false negatives, reducing the accuracy of permeable channel identification and failing to meet the refined requirements of geological disaster early warning and prevention projects in subsidence areas. Therefore, this invention provides a method for identifying and analyzing the genesis of permeable channels in subsidence areas based on high-density EDS. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for identifying and analyzing the causes of permeable channels in subsidence areas based on high-density electrical resistivity tomography (EDT), thereby solving the aforementioned problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying and analyzing the formation of permeable channels in subsidence areas based on high-density electrical resistivity tomography, comprising the following steps: Step 1, Data Acquisition: A multi-line cross-layout method is adopted in the work area to form a three-dimensional observation grid covering the entire area, and to obtain the original resistivity data of different depth layers in the subsidence area; at the same time, the spatiotemporal distribution of geological borehole lithology data, groundwater level dynamic monitoring data and information on human engineering activities are collected. Step 2, Data Preprocessing: Wavelet transform combined with adaptive threshold filtering algorithm is applied to eliminate random noise and terrain interference in the original resistivity data; missing data points are filled in by Kriging interpolation to construct a complete two-dimensional resistivity dataset. Step 3, Three-dimensional resistivity inversion: A three-dimensional geoelectric model is constructed based on the finite element numerical simulation method. The preprocessed multi-line data is input into the regularized inversion algorithm. By iteratively optimizing the model parameters, the three-dimensional resistivity distribution characteristics of the target area are obtained, and low-resistivity anomaly areas are identified. Step 4, Data Fusion and Identification: Geological borehole lithology data, groundwater level dynamic monitoring data, and vector layers of human engineering activities are fused to construct a feature set including resistivity anomalies, lithological combinations, and hydraulic gradients; an improved convolutional neural network model is used to classify low-resistivity anomaly areas in the 3D inversion results to determine the 3D spatial morphology, extension direction, and connectivity of permeable channels. Step 5, Genetic Synergistic Analysis: Establish a generative analysis index system that includes stratigraphic lithology, groundwater dynamic conditions, and human engineering activities; quantify the weight of each index through the analytic hierarchy process (AHP); combine numerical simulation technology to analyze the coupling effect between the groundwater seepage field and the stratigraphic stress field, and clarify the dominant factors and multi-factor synergistic mechanisms in the formation of permeable channels. Step 6, Model Validation and Accuracy Assessment: The identification results are cross-validated by combining borehole core test results and groundwater tracer test data. The accuracy and recall indices are calculated to assess the identification accuracy and ensure that the needs of refined early warning and prevention of subsidence area disasters are met. Step 7, Results Output: Based on the above verification results, optimize the inversion parameters and identification model, and finally output the three-dimensional distribution map of the permeable channels and the cause analysis report.

[0006] Preferably, in step 1, the multiple survey lines are laid out in an orthogonal or oblique grid manner. The spacing between adjacent survey lines is set to 5 to 20 meters according to the target area size and resolution requirements. The length of a single survey line covers the main subsidence hazard points in the work area. The electrode spacing is 2 to 5 meters. During the data acquisition process, the GPS coordinates of the survey line position are recorded synchronously, and the control error is within ±0.5 meters.

[0007] Preferably, the spatiotemporal distribution of human engineering activity information in step 1 is constructed into a vector layer using Geographic Information System (GIS) technology, which includes elements such as underground pipelines, foundation pit excavation, and groundwater extraction wells, with data accuracy reaching the meter level.

[0008] Preferably, in step 2, the wavelet transform uses the db4 wavelet basis function for 3-level decomposition, and adaptive threshold filtering is applied to the high-frequency coefficients. The threshold calculation is based on the Birge-Massart strategy. When performing Kriging interpolation, a spherical variogram model is selected, and the interpolation error is controlled within 5% of the standard deviation of the original data, effectively ensuring the integrity and accuracy of the two-dimensional resistivity dataset.

[0009] Preferably, the three-dimensional geoelectric model in step 3 is divided using a tetrahedral unstructured mesh, with the mesh in the low-resistivity anomaly region locally refined to 1 / 3 of the original mesh size, and the total number of mesh cells ≥ 10. 5The regularized inversion algorithm adopts Tikhonov regularization, and the regularization parameters are determined by the L-curve method. The iteration termination condition is that the root mean square error of the model parameters in two consecutive iterations is less than 0.5%. The spatial resolution of the final output three-dimensional resistivity distribution results meets the requirements of refined disaster prevention and control.

[0010] Preferably, the improved convolutional neural network model in step 4 introduces a channel attention module on the basis of the traditional CNN architecture, focusing on strengthening the extraction of edge and connectivity features in low-resistivity anomaly regions; the model training uses the Adam optimizer, with an initial learning rate of 0.001, and improves convergence stability by decaying the learning rate by 10% every 10 rounds; the ratio of training set to validation set samples is 7:3, and the model accuracy must reach more than 90% before proceeding to the next step.

[0011] Preferably, the causal analysis index system in step 5 includes a target layer, a criterion layer, and an index layer. The analytic hierarchy process (AHP) is based on the hierarchical structure of the target layer, criterion layer, and index layer. It determines the weight of each index by calculating the maximum eigenvalue and its corresponding eigenvector, and performs a consistency check, requiring a consistency ratio CR < 0.1 to ensure reasonable weight allocation. The numerical simulation technology uses the finite difference method to construct a seepage-stress coupling model. The model boundary conditions are set as the hydrogeological boundary and engineering load boundary of the actual site. The time step is 1 day, and iterative calculation is performed until the seepage field and stress field reach dynamic equilibrium. Finally, the distribution characteristics of the coupling field and the sensitivity analysis results of key indicators are output to clarify the contribution degree and synergistic effect mode of each factor to the formation of the permeable channel.

[0012] Preferably, the cross-validation in step 6 adopts the 5-fold cross-validation method, and the dataset is divided into training set and validation set in an 8:2 ratio; the accuracy and recall are calculated based on the confusion matrix, and the target recognition accuracy is ≥90% and the recall is ≥85%; if the target is not met, return to step 3 to adjust the inversion parameters or step 4 to optimize the model structure.

[0013] Preferably, the output of step 7 includes a three-dimensional rendering of the permeable channels, a heat map of the formation analysis, and a technical report. The three-dimensional rendering of the permeable channels is used to visually display the spatial distribution, burial depth, orientation, and connectivity characteristics of the permeable channels. The heat map of the formation analysis uses color gradients to visually reflect the contribution of the dominant factors in the formation of permeable channels in different areas. The technical report systematically elaborates on the technical details of the entire process of model construction, inversion identification, verification and evaluation, and proposes hierarchical and zoned governance suggestions based on the formation mechanism, providing a scientific basis for the construction of a disaster early warning system and long-term prevention and control in subsidence areas.

[0014] Beneficial effects Compared with the prior art, the present invention has the following advantages: 1. This invention overcomes the limitations of traditional single-profile data interpretation by deploying a three-dimensional observation grid and fusing multi-source data. Combining finite element three-dimensional inversion with an improved convolutional neural network model, it achieves precise characterization of the three-dimensional spatial morphology, extension direction, and connectivity of permeable channels, improving the accuracy and spatial resolution of low-resistivity anomaly identification.

[0015] 2. In the data preprocessing stage, the db4 wavelet basis function decomposition and adaptive threshold filtering algorithm are adopted to effectively suppress false positives or false negatives caused by random noise and terrain interference, thus ensuring the reliability of the data.

[0016] 3. In the causal analysis stage, the weights of multiple factors were quantified by the Analytic Hierarchy Process (AHP) and combined with seepage-stress coupling numerical simulation, which systematically revealed the synergistic mechanism between stratigraphic lithology, groundwater dynamic conditions and human engineering activities, thus avoiding the one-sidedness that may exist in traditional causal analysis methods.

[0017] 4. The model validation and accuracy assessment process ensures that the accuracy of the identification results meets the refined requirements of disaster early warning and prevention in subsidence areas through cross-validation and index calculation. The final output of the three-dimensional distribution map and cause analysis report can provide a scientific basis for the hierarchical and zoned governance of subsidence areas and the construction of a disaster early warning system.

[0018] 5. Compared with existing technologies, this invention has significant advantages in terms of the accuracy of permeable channel identification, the comprehensiveness of causal analysis, and the practicality of results application, and can provide effective support for the long-term prevention and control of geological disasters in subsidence areas. Attached Figure Description

[0019] Figure 1 This is a flowchart of the process of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 A method for identifying and analyzing the formation of permeable channels in subsidence areas based on high-density electrical resistivity tomography includes the following steps: Step 1: Data Acquisition. Multiple survey lines are laid out across the work area to form a three-dimensional observation grid covering the entire area. This grid enables comprehensive and high-precision acquisition of electrical parameters of strata at different depths within the work area, effectively capturing the distribution characteristics of permeable channels in three-dimensional space. This avoids information omissions due to limited observation range or insufficient resolution, providing continuous and reliable raw data support for subsequent inversion identification and causal analysis of permeable channels. Resistivity raw data at different depths in the subsidence area are obtained. This raw resistivity data directly reflects the electrical differences in the formation medium. Since permeable channel areas are filled with loose soil or water with high water content, their resistivity is usually significantly lower than that of the surrounding intact bedrock or dense strata. Based on this, areas with resistivity anomalies can be preliminarily screened. Subsequently, data preprocessing steps (such as removing random noise, etc.) are performed. (Correcting topographic influences and electrode coupling interference, etc.) to further improve data quality, providing high-quality input for the construction of subsequent three-dimensional resistivity inversion models, and helping to accurately restore the geometric shape and spatial distribution of permeable channels in three-dimensional space; simultaneously collecting geological borehole lithology data, groundwater level dynamic monitoring data, and spatiotemporal distribution information of human engineering activities, and then being able to perform multi-source data fusion analysis of the obtained resistivity structure with actual stratigraphic lithology, groundwater dynamic conditions, and human activity influences, accurately distinguishing the types of permeable channels with natural causes (such as karst development, fault fracture zones) and human causes (such as underground engineering disturbance, pipeline leakage), and correcting the uncertainties in the inversion model through cross-validation, ultimately providing a comprehensive and solid scientific basis for the spatial positioning, scale quantification, and causal tracing of permeable channels in subsidence areas, supporting the optimized design of subsequent prevention and control engineering schemes.

[0022] Step 2, Data Preprocessing: Wavelet transform combined with an adaptive threshold filtering algorithm is used to remove random noise and topographic interference from the original resistivity data. This allows for more accurate preservation of the effective signals of electrical differences in the formation medium, and lays a good foundation for subsequent electrode coupling interference correction. The output is a preprocessed resistivity dataset with improved signal-to-noise ratio, providing high-quality input support for the accurate construction of the 3D resistivity inversion model in Step 3. This ensures that the inversion results more realistically and meticulously reflect the actual electrical distribution characteristics of the subsidence area strata, reducing inversion bias caused by data noise or interference. Kriging interpolation is used to complete missing data points, constructing a complete 2D resistivity dataset, which provides continuous and uniform basic grid support for the generation of the 3D data volume. Based on this, a topographic correction factor is introduced to perform elevation normalization on the 2D dataset, eliminating resistivity value deviations caused by topographic undulations at observation points, ensuring that electrical data from different locations have a unified reference benchmark. Subsequently, through coordinate transformation and spatial mapping of the electrode arrangement, the two-dimensional profile data is expanded into a discrete sampling point cloud in three-dimensional space. Combined with prior geological information of the stratigraphic structure, the sampling point cloud is spatially gridded and attribute-assigned, ultimately forming a structured three-dimensional resistivity initial dataset. This provides a coherent and reliable spatial data framework for the accurate solution of the subsequent three-dimensional inversion model, effectively improving the ability to capture the three-dimensional morphology of the permeable channel during the inversion process.

[0023] Step 3: Three-dimensional resistivity inversion: A three-dimensional geoelectric model is constructed based on the finite element method. Tetrahedral or hexahedral elements are used to finely discretize the strata in the study area, accurately characterizing the boundary features of irregular permeable channels such as karst cavities and fault fracture zones, while balancing computational efficiency and model accuracy. During model construction, the geological borehole lithology data and prior stratigraphic structure information collected in Step 1 are introduced as constraints to limit the initial range of stratigraphic electrical parameters within a reasonable range, avoiding the inversion process from getting trapped in local optima. After inversion convergence, a three-dimensional resistivity distribution model is output. Through isosurface extraction and visualization analysis of the model, the spatial location, orientation, depth, and scale of low resistivity anomaly areas are identified. Combined with multi-source data collected in Step 1 (such as groundwater level dynamics and human engineering activity information), the formation type (natural or anthropogenic) of permeable channels is further clarified. The preprocessed multi-line data is input into a regularized inversion algorithm to iteratively optimize the model parameters, obtaining the three-dimensional resistivity of the target area. By analyzing the resistivity distribution characteristics, low-resistivity anomaly zones can be identified, enabling precise reconstruction of the three-dimensional spatial morphology of permeable channels. Depth matching with geological borehole lithology data verifies the actual stratigraphic medium type corresponding to the low-resistivity anomaly zones. Simultaneously, combined with dynamic groundwater level monitoring data, the seepage path and dynamic characteristics of groundwater within the permeable channels are analyzed, revealing their spatiotemporal correlation mechanism with subsidence. Furthermore, the obtained three-dimensional resistivity structure is overlaid with spatiotemporal distribution data of human engineering activities to clarify the scope and extent of the impact of human factors such as underground engineering excavation and pipeline leakage on the formation and expansion of permeable channels, providing direct quantitative evidence for distinguishing between naturally and anthropogenically formed permeable channels. The final three-dimensional inversion results not only achieve precise spatial positioning and scale quantification of permeable channels but also provide high-resolution geophysical evidence for subsequent causal analysis and target selection for prevention and control projects, effectively improving the scientific rigor and reliability of permeable channel identification and causal diagnosis in subsidence areas, and facilitating the precise implementation of subsequent prevention and control measures.

[0024] Step 4, Data Fusion and Identification: Geological borehole lithology data, groundwater level dynamic monitoring data, and vector layers of human engineering activities are fused. These vector layers are spatially overlaid and attribute-associated with the 3D resistivity inversion model output in Step 3 to construct a multi-source data fusion geospatial information system (GIS) platform. Based on the fused multi-source data, an ensemble learning algorithm combining random forest and gradient boosting tree is adopted to construct a feature set containing resistivity anomalies, lithological combinations, and hydraulic gradients. The feature set is input into the ensemble learning model for training and optimization. The model hyperparameters (such as decision tree depth, subsample ratio, learning rate, etc.) are adjusted through 10-fold cross-validation to improve the model's classification accuracy and generalization ability for permeable channel types. The trained model is used to label the low-resistivity anomaly areas in the 3D resistivity inversion results unit by unit, and a spatial distribution vector layer of permeable channels is output to clarify their geometric shape, burial depth range, and size. Subsequently, the labeled results and geological borehole lithological profiles were overlaid on the GIS platform, and a 3D visualization model of the permeable channel was generated through spatial interpolation. This model intuitively displays the spatial relationship between the permeable channel and the surrounding strata, groundwater system, and human engineering activities. Simultaneously, time series analysis of groundwater level dynamic monitoring data was combined to identify the dynamic changes in groundwater seepage within the permeable channel, further verifying the reliability of the model's identification results. Finally, a comprehensive identification report was output, including the spatial location of the permeable channel, its formation type (natural / anthropogenic), scale quantification, and risk level. This provides intuitive and operable technical support for the precise management of permeable channels in subsidence areas. An improved convolutional neural network model was used to classify low-resistivity anomaly areas in the 3D inversion results, determining the 3D spatial morphology, extension direction, and connectivity of the permeable channel. This allows for more precise identification of the key control sections of the permeable channel, providing targeted positioning basis for subsequent engineering management measures such as grouting and curtain seepage interception.

[0025] Step 5: Genetic Synergistic Analysis: Establish a genetic analysis index system encompassing stratigraphic lithology, groundwater dynamic conditions, and human engineering activities. This system quantifies the contribution of each factor to the formation and evolution of permeable channels. Specifically, stratigraphic lithology indicators can be combined with low-resistivity anomaly distributions in the three-dimensional resistivity inversion model, correlated with karst development strata and fault fracture zones revealed by geological boreholes, to assess the inherent vulnerability of the natural medium. Groundwater dynamic conditions indicators integrate dynamic water level monitoring data with seepage path characteristics in resistivity structures to analyze the impact of groundwater runoff intensity and hydraulic gradient on channels. The expansion of permeable channels is driven by human activities. Human engineering activity indicators are based on spatiotemporal distribution information, matching the spatial overlap between disturbed areas such as underground excavation and pipeline leakage and permeable channels to assess the induced effects of human disturbance. The weight of each indicator is quantified through the analytic hierarchy process (AHP), and the coupling effect between the groundwater seepage field and the stratum stress field is analyzed by combining numerical simulation technology. This clarifies the dominant factors and multi-factor synergistic mechanisms in the formation of permeable channels, thereby enabling precise differentiation of the evolution paths of permeable channels of different origins. This provides precise guidance for subsequent targeted prevention and control measures, curbs the further expansion of permeable channels, and ensures the engineering safety and geological environment stability of subsidence areas.

[0026] Step 6, Model Validation and Accuracy Assessment: The identification results are cross-validated by combining borehole core test results and groundwater tracer test data. This method allows for point-by-point comparison of the actual permeable layers and lithological characteristics of the borehole core with the resistivity anomaly areas at the corresponding depths in the 3D inversion model, verifying the accuracy of the spatial positioning of the permeable channels. The migration path and concentration distribution curve of the tracer in the groundwater tracer test are used to cross-validate whether the connectivity and seepage direction of the identified permeable channels are consistent with reality. If there are deviations, the filtering parameters of the data preprocessing or the regularization factor of the 3D inversion are adjusted retrospectively to further optimize the model. Accuracy and recall are calculated to assess the identification precision, ensuring that the refined requirements for disaster early warning and prevention in subsidence areas are met. Using these metrics to evaluate accuracy quantifies the model's precision in identifying permeable channels. Accuracy reflects the proportion of real permeable channel units in the model's identification results, effectively measuring the probability of misidentifying non-permeable areas as permeable channels. Recall reflects the model's ability to capture actual permeable channel units, accurately assessing the risk of missing permeable channels. In other words, the combination of accuracy and recall allows for a comprehensive evaluation of the model's overall identification performance, avoiding evaluation bias caused by a single metric.

[0027] Step 7, Results Output: Based on the above verification results, optimize the inversion parameters and identification model, and finally output a three-dimensional distribution map of permeable channels and a causal analysis report; that is, it can intuitively present the three-dimensional spatial morphology, scale boundary and causal type of permeable channels, and provide visual support for the accurate positioning and causal tracing of permeable channels in subsidence areas.

[0028] Specifically, in step 1, multiple survey lines are laid out using orthogonal or oblique grids to ensure accurate alignment between the spatial position of the survey lines and the geological borehole and topographic contour data, laying a unified coordinate benchmark for subsequent spatial overlay analysis of multi-source data. The spacing between adjacent survey lines is set to 5–20 meters according to the target area size and resolution requirements. The length of a single survey line covers the main subsidence hazard points within the work area, and the electrode spacing is 2–5 meters. By setting the survey line spacing and electrode spacing, high-density coverage of all subsidence hazard areas within the work area can be achieved, avoiding information omissions under a single spacing. Furthermore, the GPS coordinates of the survey line positions are recorded synchronously during data acquisition, with the error controlled within ±0.5 meters. Controlling the GPS coordinate error within ±0.5 meters effectively controls the spatial offset of the three-dimensional observation grid, ensuring the positional accuracy of each sampling point in the subsequent three-dimensional resistivity inversion model. This provides reliable spatial positioning support for accurately reconstructing the geometry, burial depth distribution, and spatial orientation of the permeable channel, further enhancing the scientific rigor and accuracy of the entire identification method.

[0029] Specifically, the spatiotemporal distribution information of human engineering activities in step 1 is used to construct a vector layer using GIS technology. This layer includes elements such as underground pipelines (pipe diameter, burial depth, laying time), foundation pit excavation (range, depth, construction period), and groundwater extraction wells (location, monthly variation in extraction volume). The data accuracy reaches the meter level. The spatial location of underground pipeline leakage points can directly correspond to the dense distribution zone of low-resistivity anomaly areas in the three-dimensional resistivity inversion model; the depth and range of foundation pit excavation can explain the lateral expansion direction and longitudinal extension depth of shallow permeable channels; and the monthly variation curve of groundwater extraction volume can be correlated with the periodic fluctuation characteristics of groundwater dynamic conditions within the permeable channels.

[0030] Specifically, in step 2, the wavelet transform uses the db4 wavelet basis function for three-level decomposition, and adaptive threshold filtering is applied to the high-frequency coefficients. The threshold calculation is based on the Birge-Massart strategy, which can accurately separate the high-frequency noise components in the original resistivity data from the effective low-frequency signals reflecting the changes in formation electrical properties. This preserves the detailed features of the resistivity differences between the permeable channels and the surrounding formations to the greatest extent, avoiding the problem of blurred abnormal boundaries caused by excessive smoothing in traditional filtering methods. When performing Kriging interpolation, a spherical variogram model is used, and the interpolation error is controlled within 5% of the standard deviation of the original data, effectively ensuring the integrity and accuracy of the two-dimensional resistivity dataset. This allows for the generation of continuous and detailed two-dimensional resistivity profiles, clearly revealing the boundaries and distribution characteristics of resistivity anomalies in the formation. By spatially integrating the two-dimensional profile data of multiple survey lines, a high-resolution three-dimensional resistivity model can be constructed. The three-dimensional resistivity model can accurately depict the location, morphology, and extension pattern of permeable channels in three-dimensional space. It can be corroborated with the human engineering activity vector layer information constructed by GIS in step 1, effectively linking the distribution characteristics of permeable channels with the spatiotemporal impact of human engineering activities, and providing solid geophysical data support for subsequent in-depth analysis of the formation mechanism of permeable channels.

[0031] Specifically, in step 3, the three-dimensional geoelectric model uses a tetrahedral unstructured mesh, with the mesh in the low-resistivity anomaly region locally refined to 1 / 3 of the original mesh size, and the total number of mesh elements not less than 10. 5 The regularized inversion algorithm employs Tikhonov regularization, with regularization parameters determined using the L-curve method. The iteration termination condition is that the root mean square error of the model parameters in two consecutive iterations is less than 0.5%. The final output of the three-dimensional resistivity distribution results meets the spatial resolution requirements for disaster prevention and control. Furthermore, it can accurately distinguish the boundaries and spatial distribution relationships of low-resistivity anomalies of different depths and scales. Combining the human engineering activity vector layer constructed by GIS in step 1 (such as underground pipeline laying trajectories, excavation range of foundation pits, and distribution of groundwater extraction wells) with the low-resistivity anomaly characteristics of the two-dimensional resistivity profile in step 2, the spatial overlay degree and temporal evolution correlation between permeable channels and surrounding human engineering activity areas can be further quantified.

[0032] Specifically, the improved convolutional neural network model in step 4 adds a channel attention module to the traditional CNN architecture, focusing on strengthening the extraction of edge and connectivity features in low-resistivity anomaly areas. The model training uses the Adam optimizer with an initial learning rate of 0.001. The convergence stability is improved by a learning rate decay strategy (decreasing by 10% every 10 rounds). The ratio of training set to validation set samples is 7:3. The model accuracy must reach above 90% before proceeding to subsequent steps. This enables accurate identification of the spatial morphology and extension direction of permeable channels, automatic labeling of key nodes (such as the starting burial depth, lateral extension boundary, and intersection with underground pipelines), and combined with the three-dimensional resistivity distribution results in step 3 and the human engineering activity information from the GIS vector layer in step 1, to achieve intelligent correlation analysis between permeable channels and surrounding environmental elements, providing a highly reliable quantitative identification basis for subsequent tracing of the causes of permeable channels.

[0033] Specifically, the causal analysis index system in step 5 includes a target layer (cause of permeable channels), a criterion layer (stratum lithology, groundwater dynamics, and human engineering activities), and an index layer (12 specific indicators such as stratum permeability, groundwater flow velocity, and excavation depth of foundation pits). The analytic hierarchy process (AHP) is based on the hierarchical structure of the target layer, criterion layer, and index layer. By calculating the maximum eigenvalue and its corresponding eigenvector, the weights of each index are obtained. At the same time, a consistency check is performed, requiring a consistency ratio CR < 0.1 to ensure the rationality of the weight allocation. The numerical simulation technology uses the finite difference method to construct a seepage-stress coupling model. The model boundary conditions are set as the hydrogeological boundary and engineering load boundary of the actual site. The time step is 1 day. Iterative calculations are performed until the seepage field and stress field reach dynamic equilibrium. Finally, the distribution characteristics of the coupling field and the sensitivity analysis results of key indicators are output to clarify the contribution degree and synergistic effect mode of each factor to the formation of permeable channels.

[0034] Specifically, the cross-validation in step 6 adopts the 5-fold cross-validation method, and the dataset is divided into training set and validation set in an 8:2 ratio; accuracy and recall are calculated based on the confusion matrix, and the target recognition accuracy is ≥90% and the recall is ≥85%. If the target is not met, return to step 3 to adjust the inversion parameters or step 4 to optimize the model structure.

[0035] Specifically, the output of step 7 includes a three-dimensional rendering of the permeable channels, a heat map of the formation analysis, and a technical report. The three-dimensional rendering of the permeable channels is used to visually display the spatial distribution, burial depth, orientation, and connectivity characteristics of the permeable channels. The heat map of the formation analysis uses color gradients to visually reflect the degree of contribution of the dominant factors in the formation of permeable channels in different areas. The technical report systematically summarizes the technical details of the entire process of model construction, inversion identification, verification and evaluation, and proposes hierarchical and zoning governance suggestions based on the formation mechanism, providing a scientific basis for the construction of a disaster early warning system and long-term prevention and control in subsidence areas.

[0036] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0037] First, high-density electrical resistivity tomography lines are laid out using orthogonal or oblique grids, and geological borehole lithology, groundwater level dynamic monitoring and human engineering activities GIS vector layers are integrated simultaneously to establish a unified spatial coordinate benchmark. Secondly, the electrical resistivity data preprocessing was completed by using the db4 wavelet basis function three-level decomposition and adaptive threshold filtering. The continuous two-dimensional resistivity profile was generated by combining the spherical variogram Kriging interpolation. Then, a high-resolution three-dimensional resistivity model was constructed by tetrahedral unstructured mesh generation and Tikhonov regularization inversion. Then, a geoscience information system platform is constructed by integrating multi-source data. The random forest-gradient boosting tree ensemble learning algorithm and the improved convolutional neural network with added channel attention module are used to classify and label the low-resistivity anomaly areas in the three-dimensional inversion results, and output a vector layer of permeable channel spatial distribution and a three-dimensional visualization model. Furthermore, by using a three-level indicator system that includes stratigraphic lithology, groundwater dynamics, and human engineering activities, combined with the analytic hierarchy process (AHP) to quantify weights and seepage-stress coupling numerical simulation, the contribution and synergistic mechanism of each factor to the formation of permeable channels are analyzed. Finally, through cross-validation of borehole core testing and groundwater tracer experiments, and evaluation of accuracy and recall, the inversion parameters and model structure were optimized, and a three-dimensional rendering of the permeable channel, a heat map of the formation analysis, and a technical report were output.

[0038] The above process is based on the deep fusion of multi-source data and supported by intelligent algorithms and numerical simulation, forming a complete technical chain from data collection to governance decision-making. This ensures the accuracy of permeable channel identification and the scientific nature of cause diagnosis, providing targeted technical support for disaster prevention and control in subsidence areas.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying and analyzing the cause of water permeation channels in a subsidence area based on high-density electrical method, characterized in that, Comprise the following steps: Step 1, data acquisition: in the work area, multiple measuring lines are cross-laid to form a three-dimensional observation grid covering the whole area, and the resistivity raw data of different depth layers in the subsidence area are obtained; the spatial and temporal distribution of the lithology data of geological drilling, the dynamic monitoring data of underground water level and the information of human engineering activities are synchronously collected; Step 2, data preprocessing: the random noise and terrain interference in the original resistivity data are eliminated by applying wavelet transform combined with adaptive threshold filtering algorithm; The missing data points are completed by Kriging interpolation method to construct a complete two-dimensional resistivity data set; Step 3, three-dimensional resistivity inversion: based on the finite element numerical simulation method, a three-dimensional geoelectric model is constructed, the preprocessed multi-line data is input into the regularization inversion algorithm, the model parameters are iteratively optimized, the three-dimensional resistivity distribution characteristics of the target area are obtained, and the low-resistance abnormal area is identified; Step 4, data fusion and identification: the lithology data of geological drilling, the dynamic monitoring data of underground water level and the vector layer of human engineering activities are fused to construct a feature set containing resistivity anomaly, lithology combination and hydraulic gradient; the improved convolutional neural network model is used to classify the low-resistance abnormal area in the three-dimensional inversion result to determine the three-dimensional spatial form, extension direction and connectivity of the water permeable channel; Step 5, cause analysis: a cause analysis index system containing stratum lithology, underground water dynamic condition and human engineering activity is established; the weights of each index are quantified by AHP method, the coupling effect of underground water seepage field and stratum stress field is analyzed by numerical simulation technology, and the dominant factor and multi-factor synergistic mechanism of the formation of the water permeable channel are determined; Step 6, model verification and precision evaluation: the identification results are cross-verified by combining the drilling core test results and underground water tracing test data, the accuracy and recall rate indexes are calculated to evaluate the identification precision, and the needs of fine early warning and prevention of the subsidence area disaster are ensured; Step 7, result output: the inversion parameters and identification model are optimized according to the above verification results, and finally the three-dimensional distribution map of the water permeable channel and the cause analysis report are output.

2. The sinkhole area water-permeable passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, The multiple measuring lines in step 1 are cross-laid in orthogonal grid or oblique grid mode, the distance between adjacent measuring lines is set to 5-20 meters according to the scale and resolution requirement of the target area, the length of a single measuring line covers the main subsidence hidden danger points in the work area, the electrode spacing is 2-5 meters, and the GPS coordinates of the measuring line position are recorded synchronously during data collection, and the control error is within ±0.5 meters.

3. The sinkhole area water-permeable passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, The spatial and temporal distribution of human engineering activity information in step 1 is constructed into a vector layer by geographic information system (GIS) technology, including underground pipeline, foundation pit excavation, underground water exploitation well and other elements, and the data accuracy reaches meter level.

4. The sinkhole water passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, In step 2, the db4 wavelet basis function is used for 3-layer decomposition, the adaptive threshold filtering is applied to the high-frequency coefficients, and the threshold calculation is based on Birge-Massart strategy; when Kriging interpolation is used, the spherical variation function model is selected, the interpolation error is controlled within 5% of the standard deviation of the original data, and the integrity and accuracy of the two-dimensional resistivity data set are ensured.

5. The sinkhole water passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, The three-dimensional geoelectric model in step 3 is divided by using a tetrahedral unstructured grid, the grid in the low-resistance abnormal area is locally encrypted to 1 / 3 of the original grid size, and the total number of grid units is greater than or equal to 10 5 ; the regularization inversion algorithm adopts Tikhonov regularization, the regularization parameter is determined by the L curve method, and the iteration termination condition is that the root mean square error of adjacent two model parameters is less than 0.5%; the spatial resolution of the final output three-dimensional resistivity distribution result meets the requirement of fine prevention and control of disasters.

6. The sinkhole area water-permeable passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, The improved convolutional neural network model in step 4 introduces a channel attention module based on the traditional CNN architecture, focusing on enhancing the edge and connectivity feature extraction of low-resistance abnormal areas. The model training uses the Adam optimizer with an initial learning rate of 0.001, and the learning rate decay strategy of 10% decay every 10 rounds is used to improve the convergence stability. The training set and validation set sample ratio is 7:3, and the model accuracy needs to reach more than 90% before entering the subsequent steps.

7. The subsidence area water-permeable passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, The cause analysis index system in step 5 includes target layer, criterion layer and index layer. The analytic hierarchy process is based on the hierarchical structure of target layer, criterion layer and index layer, and the weight of each index is determined by calculating the maximum eigenvalue and its corresponding eigenvector, while the consistency test is carried out, requiring the consistency ratio CR<0.1 to ensure the reasonable weight distribution; The numerical simulation technology uses the finite difference method to construct the seepage-stress coupling model, the model boundary conditions are set as the actual hydrogeological boundary and engineering load boundary, the time step is 1 day, and the iterative calculation is carried out until the seepage field and stress field reach dynamic equilibrium; The final output of the coupling field distribution characteristics and the sensitivity analysis results of key indicators clearly shows the contribution degree and synergistic mode of each factor to the formation of the permeable channel.

8. The sinkhole water passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, The cross-validation in step 6 uses 5-fold cross-validation, and the data set is divided into training set and validation set in the ratio of 8:2; The accuracy and recall rate are calculated based on the confusion matrix, the target recognition accuracy is ≥90% and the recall rate is ≥85%; If not up to standard, return to step 3 to adjust the inversion parameters or step 4 to optimize the model structure.

9. The subsidence area water-permeable passage identification and cause analysis method based on high-density electrical method according to claim 1, characterized in that, The result output in step 7 includes the three-dimensional rendering of the permeable channel, the cause analysis heat map and the technical report; Among them, the three-dimensional rendering of the permeable channel is used to intuitively show the spatial distribution form, burial depth, trend and connectivity characteristics of the permeable channel; The cause analysis heat map directly reflects the contribution degree of the dominant factor of the formation of the permeable channel in different regions with color gradient; The technical report systematically describes the whole process technical details of model construction, inversion identification, verification evaluation, and proposes hierarchical zoning governance suggestions combined with the cause mechanism, providing scientific basis for the construction of disaster warning system and long-term prevention and control of subsidence area.

Images