Mining area geological settlement risk assessment method and system

Abstract

The embodiment of the invention provides a mining area geological settlement risk assessment method and system, and belongs to the technical field of numerical simulation risk assessment. The method comprises the following steps: acquiring observation data and structural information of each monitoring point in a mining area, and constructing a point-level capsule for expressing settlement behavior characteristics of each monitoring point based on the observation data; calculating a structure coupling factor between the monitoring points based on the structure information, and taking the structure coupling factor as a weight adjustment parameter in a dynamic routing process in the capsule network; jointly inputting the point-level capsules and the structure coupling factors into a preset structure coupling capsule network, and guiding a dynamic routing mechanism to generate structure unit capsules for representing settlement response characteristics of different structure units according to the weight adjustment parameters; and based on the structural unit capsules, generating settlement risk levels and spatial distribution of the monitoring points. According to the scheme, the precision and the real-time performance of mining area settlement risk assessment are remarkably improved, and more reliable decision support is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of numerical simulation risk assessment, and in particular to a method and a system for assessing geological subsidence risk in mining areas. Background Art

[0002] Mine subsidence is a common and significant geological issue during mining operations. The resulting ground subsidence, cracks, and infrastructure damage often have significant impacts on mine safety and operations. To effectively manage these risks, accurate prediction of mine subsidence is essential. Currently, mine subsidence risk assessment methods rely heavily on traditional physical models, such as the finite element method (FEA) and geostatistical methods. While these methods can provide some level of subsidence prediction, they still have significant limitations when faced with complex mining geological environments.

[0003] The geological complexity of mining areas is one of the main problems that existing methods cannot effectively solve. In actual mining areas, the geological environment presents significant heterogeneity and complexity, such as fault zones, goafs, and uneven distribution of lithologic layers, which have a significant impact on the propagation and expansion of subsidence. Traditional methods usually assume that the subsidence process is spatially continuous and uniform, or rely only on local physical parameters for modeling, which makes them unable to accurately reflect the subsidence propagation pattern controlled by discontinuous geological structures such as fault zones or goafs. This spatially discontinuous subsidence propagation pattern causes existing subsidence assessment methods to often ignore or underestimate the prediction of high-risk areas, and cannot fully identify the potential subsidence risks in mining areas.

[0004] Another significant problem is the inability of existing assessment methods to handle multi-source data and spatial distribution. Subsidence monitoring data in mining areas typically include multiple data types such as subsidence rate, displacement changes, groundwater levels, lithology classification, and mining disturbances. These data are often unevenly distributed in space and have complex intercorrelations. However, most traditional assessment methods can only process a single data source or cannot fully utilize the spatial relationship between data. Therefore, when generating subsidence risk prediction results, these methods often ignore the coupling and spatial variability between data within the mining area, resulting in an inability to accurately reflect the spatial distribution of subsidence and risk hotspots. More importantly, existing methods find it difficult to generate spatial heat maps of subsidence risk, making it impossible for mine managers to quickly identify high-risk areas during monitoring, assessment, and emergency plan deployment.

[0005] Therefore, existing technologies have significant deficiencies in modeling the propagation of heterogeneous subsidence in complex mining areas, as well as in fusion of multi-source data and spatial distribution. This directly impacts the accuracy and reliability of subsidence risk assessments, and thus affects safety management and decision-making efficiency. To address these issues, a new subsidence risk assessment method is urgently needed that can overcome the limitations of existing methods, accurately simulate subsidence risks in complex mining environments, and provide more effective decision-making support for mining management. Summary of the Invention

[0006] The purpose of the embodiments of the present invention is to provide a method and system for geological subsidence risk assessment in mining areas, so as to at least solve the problem that the existing technology cannot accurately handle the non-uniform subsidence propagation caused by the complex geological structure of the mining area and the problem of insufficient multi-source data fusion and spatial distribution representation.

[0007] In order to achieve the above-mentioned objectives, the first aspect of the present invention provides a method for assessing geological subsidence risks in mining areas, the method comprising: obtaining observation data and structural information of each monitoring point in the mining area, and constructing a point-level capsule based on the observation data to express the settlement behavior characteristics of each monitoring point; calculating the structural coupling factor between the monitoring points based on the structural information, as a weight adjustment parameter for the dynamic routing process in the capsule network; inputting the point-level capsule and the structural coupling factor into a preset structural coupling capsule network, and guiding the dynamic routing mechanism to generate structural unit capsules for characterizing the settlement response characteristics of different structural units based on the weight adjustment parameter; and generating the settlement risk level and spatial distribution of each monitoring point based on the structural unit capsule.

[0008] Optionally, the observation data include: any one or more of the settlement rate, settlement acceleration, groundwater level disturbance amplitude, lithology coding information and mining disturbance parameters of the monitoring point; the structural information includes any one or more of the spatial distance between monitoring points, fracture structure topological relationship, hydrological channel connectivity and lithology similarity; the observation data is used to construct the initial state vector of the point-level capsule, and the structural information is used to determine the structural coupling factor to participate in the weight adjustment in the dynamic routing process.

[0009] Optionally, a point-level capsule for expressing the settlement behavior characteristics of each monitoring point is constructed based on the observation data, including: performing unified scale normalization processing on the observation data of each monitoring point to obtain corresponding multidimensional features; encoding the normalized multidimensional features into a composite structure including a state vector and a posture matrix based on capsule encoding rules to obtain a point-level capsule for expressing the settlement behavior characteristics of each monitoring point; wherein the state vector is used to characterize the settlement behavior intensity and change trend of the monitoring point; and the posture matrix is ​​used to identify the response pattern of the settlement behavior in different structural directions.

[0010] Optionally, the structural coupling factor between monitoring points is calculated based on the structural information, including: constructing a candidate connection graph between monitoring points based on the structural information, screening out pairs of monitoring points that are directly connected or physically associated on the structural path as a point pair set; taking each pair of point pairs as a unit, extracting the spatial coupling feature vector between the structural paths through a graph embedding method, and constructing a learnable weighted fusion function based on the spatial coupling feature vector combined with the historical settlement collaborative change behavior; adaptively adjusting the fusion coefficients of various structural sub-factors of the weighted fusion function during the network training process to obtain the structural coupling factor for dynamic routing weight adjustment.

[0011] Optionally, the structurally coupled capsule network includes: a point-level capsule encoding layer of the input layer, a structurally coupled expression layer, and a structural unit response layer; the structurally coupled capsule network transmits information between layers in a non-fixed connection manner through a dynamic routing mechanism; the structurally coupled capsule network introduces a structural coupling factor as a control condition to replace the routing weight calculation based on the similarity score in the capsule network.

[0012] Optionally, the point-level capsule and the structural coupling factor are jointly input into a preset structural coupling capsule network, and a dynamic routing mechanism is guided according to the weight adjustment parameter to generate structural unit capsules for characterizing the settlement response characteristics of different structural units, including: performing a transformation operation on the point-level capsule to generate a set of prediction vectors for all candidate structural unit capsules; performing weighted aggregation on the prediction vectors between each point-level capsule and all candidate structural unit capsules according to the weight adjustment matrix constructed by the structural coupling factor to generate a pre-activation input of the structural unit layer; performing a normalization operation on the pre-activation input based on a nonlinear compression function to obtain a final state vector of the structural unit capsule; wherein the final state vector is composed of a plurality of sub-feature vectors, which respectively represent the settlement amplitude distribution, the settlement change rate trend and the coupling response intensity to the surrounding structural disturbance of the target structural unit.

[0013] Optionally, based on the structural unit capsule, the settlement risk level and its spatial distribution of each monitoring point are generated, including: comparing and mapping the final state vector of the structural unit capsule with a preset risk level grading rule to obtain the risk interval corresponding to each eigenvalue in the structural response area; reflecting the risk interval to each monitoring point according to the distribution position and information contribution ratio of each monitoring point in the corresponding structural unit, and generating a risk heat map in two-dimensional space based on the coordinates of the monitoring points to represent the settlement risk level and its spatial distribution of each monitoring point.

[0014] Optionally, after obtaining the settlement risk level of each monitoring point and its spatial distribution, the method further includes: based on the response consistency judgment between the structural unit capsule and the monitoring points it contains, identifying the monitoring points whose response deviation exceeds a preset threshold, and marking them as abnormal response points; identifying the distribution characteristics of each abnormal response point in the structural boundary area to construct a risk drift trend map or generate a list of recommended intervention areas; wherein the response deviation is determined by the difference between the actual risk level of the monitoring point and the risk value predicted by the structural unit capsule state vector.

[0015] The second aspect of the present invention provides a geological subsidence risk assessment system for mining areas, the system comprising: an acquisition unit for acquiring observation data and structural information of each monitoring point in the mining area, and constructing a point-level capsule based on the observation data to express the settlement behavior characteristics of each monitoring point; a processing unit for calculating the structural coupling factor between the monitoring points based on the structural information, as a weight adjustment parameter for the dynamic routing process in the capsule network; a model construction unit for inputting the point-level capsule and the structural coupling factor into a preset structural coupling capsule network, and guiding the dynamic routing mechanism to generate structural unit capsules for characterizing the settlement response characteristics of different structural units based on the weight adjustment parameter; and an evaluation unit for generating the settlement risk level and spatial distribution of each monitoring point based on the structural unit capsule.

[0016] On the other hand, the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, enables the computer to execute the above-mentioned method for assessing geological subsidence risk in mining areas.

[0017] Through the above-mentioned technical solution, the present invention constructs a mining area geological subsidence risk assessment method based on a structurally coupled capsule network, effectively addressing the existing problem of inaccurately reflecting the coupling and propagation of subsidence caused by the complex geological structure of mining areas. By acquiring observational data and structural information from monitoring points in the mining area and constructing point-level capsules based on the observational data, the settlement behavior characteristics of each monitoring point can be accurately expressed. Furthermore, the structural information is used to calculate the structural coupling factor between monitoring points, which serves as a weight adjustment parameter in the dynamic routing process, thereby enhancing the model's adaptability to subsidence propagation within the heterogeneous and complex structure of the mining area. By inputting both the point-level capsules and the structural coupling factor into the structurally coupled capsule network, the dynamic routing mechanism accurately generates the settlement response characteristics of different structural units, and then generates the settlement risk level and spatial distribution of the monitoring points based on the structural unit capsules. This method not only captures the spatial coupling relationship of subsidence but also generates a precise spatial distribution of subsidence risk, significantly improving the accuracy and real-time performance of mining area subsidence risk assessment and providing more reliable decision support.

[0018] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings: Figure 1 This is a flowchart of the steps of a method for assessing geological subsidence risk in mining areas provided by one embodiment of the present invention; Figure 2 This is a system structure diagram of a mining area geological subsidence risk assessment system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following describes the specific embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention.

[0021] Figure 1 This is a flowchart of the steps of a method for assessing geological subsidence risk in mining areas provided by one embodiment of the present invention. The method includes: Step S10: Obtain observation data and structural information of each monitoring point in the mining area, and construct a point-level capsule based on the observation data to express the settlement behavior characteristics of each monitoring point.

[0022] Specifically, the observation data includes: any one or more of the settlement rate, settlement acceleration, groundwater level disturbance amplitude, lithologic coding information, and mining disturbance parameters of the monitoring point; the structural information includes any one or more of the spatial distance between monitoring points, fracture structure topological relationship, hydrological channel connectivity, and lithologic similarity; the observation data is used to construct the initial state vector of the point-level capsule, and the structural information is used to determine the structural coupling factor to participate in the weight adjustment in the dynamic routing process.

[0023] Furthermore, based on the observation data, point-level capsules are constructed to express the settlement behavior characteristics of each monitoring point, including: performing unified scale normalization processing on the observation data of each monitoring point to obtain corresponding multidimensional features; encoding the normalized multidimensional features into a composite structure including a state vector and a posture matrix based on capsule encoding rules to obtain point-level capsules for expressing the settlement behavior characteristics of each monitoring point; wherein, the state vector is used to characterize the settlement behavior intensity and change trend of the monitoring point; and the posture matrix is ​​used to identify the response pattern of the settlement behavior in different structural directions.

[0024] In the embodiments of the present application, various types of observation data related to subsidence are collected at the multi-source monitoring points arranged in the mining area, and the observation data at least include: the subsidence rate of the monitoring point (i.e., the rate of change of the ground elevation per unit time), the subsidence acceleration (the rate of change of the subsidence rate), the groundwater level disturbance amplitude (reflecting the degree of influence of local hydrological conditions on the mechanics of the stratum), lithology coding information (such as categories of mudstone, sandstone, limestone, etc. and their corresponding numerical representation), and mining disturbance parameters related to the area (such as the time of generating the goaf, the depth of the mining layer, the time sequence of mining, and the disturbance radius range, etc.). The above data can be used independently or in combination, depending on the geological characteristics of the area where the monitoring point is located and the availability of data.

[0025] After obtaining the above-mentioned original observation data, in order to ensure the consistency of the dimensions and the uniformity of the numerical scales in the subsequent modeling process, uniform scale normalization processing needs to be performed on all observation data. Preferably, Z-score standardization or Min-Max normalization method is used to make each dimension feature fall into a uniform numerical interval or standard normal distribution, so as to avoid the unbalanced influence on network training caused by the large dimension of a certain dimension data. The normalized observation data forms a multi-dimensional feature vector, which is used as the basic input for constructing point-level capsules.

[0026] Further, based on the coding rules of the capsule network, the above-mentioned normalized feature vector is converted into a capsule unit with structural expression capability. Preferably, the point-level capsule is modeled by using a composite structure of a state vector plus a pose matrix, wherein: 1) The state vector (activation vector) is used to represent the subsidence behavior intensity and change trend of the monitoring point in the current time window, and its dimension can be selected according to business needs, such as a real number vector with a length of 8 or 16. The vector length can express the subsidence degree, and the direction can guide the dynamic allocation of the subsequent routing process.

[0027] 2) The pose matrix (pose matrix) is used to describe the response mode of the subsidence behavior of the monitoring point in different geological structure directions. Preferably, a small dimension matrix of 2x2 or 4x4 is used to encode the spatial projection characteristics and the sensitivity of each direction of the subsidence characteristics. This matrix reflects the directional trend of the subsidence change of a monitoring point under the control of the geological structure (such as the fault tendency, the rock layer bedding direction, the hydrological flow direction, etc.).

[0028] Based on the above structure, the point-level capsule not only has the ability to represent the subsidence intensity, but also has strong spatial response description capability, providing a high expression degree of atomic unit for the subsequent dynamic information routing based on structural coupling.

[0029] Furthermore, to improve the consistency and stability of capsule encoding, regularization strategies can be introduced during the construction process. For example, constraining the modulus of the state vector to not exceed a preset threshold, or applying an L2 norm regularization term to the pose matrix, can help prevent numerical explosion or gradient instability. Furthermore, based on the actual observed data dimensions, small feedforward networks or shallow convolutional layers can be used to extract and encode features from the observed data to improve the robustness of the capsule representation.

[0030] In addition to observational data, structural information related to the geological relationships between monitoring points must also be acquired simultaneously. This structural information preferably includes: the spatial Euclidean distance between any two monitoring points; the topological relationship of the fracture structure between them (such as whether they cross a fault, the distance between faults, and the direction of fault extension); the connectivity of underground hydrological pathways (which can be obtained through numerical simulation or seepage modeling); and a lithologic similarity index (obtained by comparing the lithologic codes of the rock formations at each monitoring point and calculating a similarity score). This structural information can be used to measure the strength of the geological coupling between two monitoring points, forming the basis for the subsequent calculation of the structural coupling factor.

[0031] The constructed point-level capsules are used as the underlying expression units and fed into the network's routing mechanism along with structural information. This structural information is used to calculate the structural coupling factor between monitoring points and to weight the paths assigned to point-level capsules during dynamic routing. This approach not only retains the ability to accurately express single-point subsidence characteristics but also introduces a global coupling mechanism based on geological structural relationships, laying the foundation for multi-point collaborative modeling of subsidence risk in mining areas.

[0032] Step S20: Calculate the structural coupling factor between monitoring points based on the structural information as a weight adjustment parameter for the dynamic routing process in the capsule network.

[0033] Specifically, a candidate connection graph between monitoring points is constructed based on structural information, and pairs of monitoring points that are directly connected or physically associated on the structural path are screened out as a point pair set; taking each pair of points as a unit, the spatial coupling feature vector between the structural paths is extracted through graph embedding, and a learnable weighted fusion function is constructed based on the spatial coupling feature vector combined with the historical settlement coordinated change behavior; during the network training process, the fusion coefficients of various structural sub-factors of the weighted fusion function are adaptively adjusted to obtain the structural coupling factor for dynamic routing weight adjustment.

[0034] In an embodiment of the present invention, a candidate connection graph between monitoring points is constructed based on the acquired structural information. The graph is used to describe the potential coupling relationship between each monitoring point in the mining area at the geological structure level. In the graph, nodes represent monitoring points and edges represent the possibility of structural connection. The judgment of structural connection is not limited to physical distance proximity, but is based on one or more of the following structural factors: fracture structure topology (for example, whether two monitoring points are located in the same fault block or whether they cross the fault), hydrological channel connectivity (for example, seepage simulation determines the hydraulic connectivity between two points), lithologic similarity (for example, similarity is calculated using vector distance of lithologic encoding), and consistency of tectonic stress direction. This structural information can be obtained from geological modeling results, seismic interpretation maps, and hydrological analysis models.

[0035] Based on the candidate connectivity graph, we screen all possible monitoring point pairs for those with direct connectivity or physical coupling along the structural path to form a valid point pair set. Criteria for this selection include, but are not limited to, fracture spacing less than a threshold, hydrological coupling flux exceeding background values, and lithologic coding similarity exceeding a set threshold. This point pair set is used in subsequent coupling factor construction to ensure that modeling focuses on actual relationships supported by physical mechanisms.

[0036] For each valid point pair, the spatial coupling characteristics between their structural paths need to be extracted. To achieve this goal, the structural graph is preferably embedded and modeled based on graph embedding technology, compressing the original graph topology and the structural attributes of each monitoring point into a representation in vector space. During the graph embedding process, the attribute vectors of the nodes include fault zone distribution, goaf boundaries, lithology categories, etc.; adjacency aggregation embedding methods such as multi-layer graph convolution or random walk-based node encoding strategies can be used to obtain the structural embedding feature vector between each pair of monitoring points.

[0037] Furthermore, to incorporate temporal synergy beyond structural paths, a synergistic behavior vector can be constructed between pairs of points based on historical settlement observation data. This synergistic behavior describes the synchronization and correlation of settlement changes between two monitoring points within a historical time window. For example, this synergistic behavior can be calculated by calculating the Pearson correlation coefficient, maximum dynamic time warping (DTW) similarity, or mutual information of settlement rate changes over multiple time periods. These metrics can capture dynamic correlation patterns that are non-structural but exhibit coupled behavior.

[0038] Then, based on the aforementioned spatial coupling eigenvectors and synergistic behavior vectors, a learnable weighted fusion function is constructed to integrate the contributions of various structural sub-factors (such as fractures, lithology, and hydrology) and output a final structural coupling factor. This weighted function can be implemented using a shallow neural network. Its input is the concatenation of the structural embedding vectors and synergistic eigenvectors of point pairs, and its output is a normalized structural coupling factor value, typically limited to a range between 0 and 1. To improve the interpretability and stability of training, a regularization term is introduced in the fusion coefficient during network training to control the dominance of different structural factors.

[0039] Throughout the neural network training phase, this fusion function is automatically optimized as the capsule network loss function backpropagates. The goal is to ensure that pairs of monitoring points with high coupling relationships have stronger routing weights in the final risk assessment results. For example, if a pair of monitoring points have historically correlated settlement and are located within the same lithologic block, their structural coupling factor will be assigned a higher value. Conversely, if there is a significant difference in structural connectivity, hydrological barriers, or lithologic properties between the two points, the coupling factor will be automatically suppressed.

[0040] The resulting structural coupling factor is then incorporated into the dynamic routing process to weight the fit scores between point-level capsules and structural unit capsules. Specifically, the routing probability in the original capsule network is weighted with the structural coupling factor, making pairs of monitoring points with strong structural coupling more likely to be assigned to the same structural unit during routing, thereby enabling physical coupling modeling between non-adjacent points across space.

[0041] Through the aforementioned technical approach, this implementation achieves a deep integration of the complex structural relationships of mining areas with a data-driven routing mechanism. This improves the model's ability to identify risks in structurally discontinuous areas, such as fault crossings, goaf boundaries, and groundwater disturbances. Furthermore, it makes the modeling of subsidence risk propagation more consistent with the actual geological structure and historical subsidence evolution trends, providing an accurate foundation for subsequent detailed assessments of subsidence risk levels and identification of regional hotspots.

[0042] Step S30: inputting the point-level capsule and the structural coupling factor into a preset structural coupling capsule network, and guiding a dynamic routing mechanism to generate structural unit capsules for characterizing the settlement response characteristics of different structural units according to the weight adjustment parameter.

[0043] Specifically, the structurally coupled capsule network includes: a point-level capsule encoding layer of the input layer, a structurally coupled expression layer, and a structural unit response layer; the structurally coupled capsule network transmits information between layers in a non-fixed connection manner through a dynamic routing mechanism; the structurally coupled capsule network introduces a structural coupling factor as a control condition to replace the routing weight calculation based on similarity scores in the capsule network.

[0044] In an embodiment of the present invention, in the point-level capsule coding layer, the input is the point-level capsule of each monitoring point constructed through pre-processing. Each point-level capsule consists of a state vector and a posture matrix, which is used to characterize the settlement behavior characteristics of the current monitoring point. Among them, the state vector can describe the overall level of settlement intensity or change trend, and the posture matrix is ​​used to reflect the response pattern of the point in the direction of the geological structure, such as the distribution difference of settlement response in different inclination directions. In the point-level capsule coding layer, each capsule is not directly connected to all capsule units in the next layer, but the transmission path and degree of attribution of the information are determined by the routing mechanism.

[0045] In traditional capsule networks, dynamic routing mechanisms typically determine information transmission weights based on the similarity score between the prediction vector and the target capsule activation vector. However, in the context of mining subsidence risk modeling, the physical coupling of structural relationships often plays a more dominant role than pure feature space similarity. To this end, this implementation introduces a structural coupling factor as a new control variable, replacing the traditional similarity score, to modify the original coupling weight distribution during the dynamic routing process.

[0046] In the structural coupling representation layer, activation information from point-level capsules is first transformed into a corresponding predicted capsule representation through an affine transformation. This predicted representation is not directly used to calculate similarity with the activation vector in the upper layer, but is combined with the structural coupling factor to participate in the weight update process. Specifically, for each connection between a point-level capsule and a candidate structural unit capsule, the structural coupling factor in the structure graph is used as a weighted correction factor to update the initial fitness score in dynamic routing. The correction rule is preferably: the original fitness score is multiplied by the structural coupling factor to obtain a new attribution weight coefficient. In this way, if a pair of capsules has a high structural coupling strength, its information transfer weight is significantly increased; otherwise, its attribution probability is reduced, limiting the propagation path of information interference.

[0047] In practice, this structural coupling factor can be understood as a structure-aware attention mechanism, enabling the network to prioritize physically meaningful transmission paths among multiple candidate connections, thereby mitigating biases often driven by feature similarity alone. This mechanism is particularly applicable to subsidence risk assessment and modeling in complex geological scenarios within mining areas, such as fracture-controlled areas, goaf boundaries, and lithologic abrupt zones.

[0048] The structural unit response layer is responsible for converging the activation information of multiple point-level capsules adjusted by the structural coupling expression layer to generate a structural unit capsule representing the regional subsidence response. The structural unit capsule is essentially an aggregated representation of multiple coupled point capsules, and its output state vector is used to depict the subsidence risk intensity of the target region, and the posture matrix represents the subsidence response mode of the region in each direction. For example, if the coupled points of a certain structural unit in the east-west direction all show significant subsidence trends, the corresponding dimension weight of the posture matrix will be pulled up, so that the risk deduction process has directional guidance significance.

[0049] It should be noted that, in the routing iteration process, in order to further improve the network stability and expression ability, the present embodiment introduces a regularization constraint term in the updating process of the structural coupling weight to prevent the problem of excessive concentration of weight or gradient disappearance. In addition, in order to adapt to the actual needs of the mining area subsidence evaluation, the maximum number of routing iterations can be set and the routing saturation can be dynamically adjusted to match the information fusion depth required by different structural unit capsules.

[0050] In a possible implementation, taking a typical mining area as an example, 20 monitoring points are set up, and the subsidence rate, subsidence acceleration, water level disturbance value, lithology type (encoded), and disturbance level corresponding to underground operation of each monitoring point are collected. The above data constitute the observation data vector of each monitoring point. At the same time, the fracture zone distribution map, hydrological channel model and lithology distribution model are extracted from the geological model to construct the structural information matrix between the monitoring points.

[0051] First, the point-level capsule input is constructed based on the observation data of each monitoring point. For each monitoring point i, the observation data vector is normalized to obtain the input vector u i . Then, the prediction vector is generated by the affine mapping matrix :

[0052] The vector represents the activation prediction of monitoring point i on candidate structural unit capsule j.

[0053] Subsequently, the structural coupling factor matrix is constructed based on the fracture topology, hydrological flux and lithology similarity, where each value represents the comprehensive structural coupling strength of the monitoring point to . The construction method is as follows: 1) For the monitoring point pair , it is calculated whether it is in the same fault block, whether it is hydrologically connected, and whether the lithology code distance is lower than the threshold.

[0054] 2) Each feature is converted into a sub-factor in the [0, 1] interval.

[0055] 3) Set the weighting coefficient , define the comprehensive coupling factor:

[0056] in, Represent the fracture, hydrological, and lithological coupling functions, respectively. satisfy .

[0057] In the dynamic routing process, the structural coupling factor is introduced to modify the weight score of the traditional similarity drive. The coupling score is 0, which is recorded as In each iteration, the routing weight is adjusted according to the following update formula:

[0058] Among them, α is the learning step size; is the current output vector of the structural unit capsule j; Denotes the inner product of the prediction and the actual output as the adaptation score.

[0059] The normalized routing weight is calculated by the softmax function:

[0060] Then, the input aggregation vector of the structure unit capsule j is:

[0061] The final output vector is obtained by the squash function:

[0062] The squash function compresses the length of the output vector, retains the direction and normalizes the modulus to (0,1), which represents the capsule activation strength. The output of the structural unit capsule v j This is further decoded into the subsidence risk level for each monitoring point. The decoder uses a fully connected neural network to output a five-level risk classification label (very low, low, medium, high, and very high), and uses a softmax function to give the confidence level for each risk category.

[0063] Specifically, the point-level capsule and the structural coupling factor are jointly input into a preset structural coupling capsule network, and the dynamic routing mechanism is guided by the weight adjustment parameter to generate structural unit capsules for characterizing the settlement response characteristics of different structural units, including: performing a transformation operation on the point-level capsule to generate a set of prediction vectors for all candidate structural unit capsules; performing weighted aggregation on the prediction vectors between each point-level capsule and all candidate structural unit capsules according to the weight adjustment matrix constructed by the structural coupling factor to generate a pre-activation input of the structural unit layer; performing a normalization operation on the pre-activation input based on a nonlinear compression function to obtain the final state vector of the structural unit capsule; wherein the final state vector is composed of a plurality of sub-feature vectors, which respectively represent the settlement amplitude distribution, settlement change rate trend and coupling response intensity to surrounding structural disturbances of the target structural unit.

[0064] In an embodiment of the present invention, an affine transformation is performed on the point-level capsule constructed at each monitoring point to obtain multiple sets of prediction vectors in the directions of different candidate structural units. The prediction vectors are used to represent the estimation of the response pattern of each monitoring point to different structural units. Subsequently, a weight adjustment matrix for controlling routing preferences is generated based on the pre-constructed structural coupling factor. This matrix is ​​derived from structural information in multiple dimensions such as fault structure, hydrological channels, and lithologic similarity. It reflects the physical linkage strength between different monitoring points through weighted fusion of coupling strength. This adjustment matrix does not rely on spatial proximity, but reflects the correlation in a structural sense, so that some non-adjacent monitoring points can establish connections in the network.

[0065] During dynamic routing, the prediction vectors between the point-level capsule and all candidate building block capsules are weighted and aggregated according to the aforementioned adjustment matrix to form the pre-activation input for the building block layer. This aggregation not only considers the similarity between the predicted content and the current output but also incorporates the modulating effect of the structural coupling factor, giving point pairs with stronger structural physical coupling greater weight in information routing. Nonlinear compression is then performed on the aggregated pre-activation input to obtain the final building block capsule state vector.

[0066] It's worth noting that the final output of the structural unit capsule state vector is composed of multiple subvectors, each representing the response characteristics of the structural unit in different dimensions. Specifically, these include: a subsidence amplitude subvector, used to characterize the overall subsidence intensity and distribution characteristics of the monitoring point set; a rate of change subvector, used to reflect the direction and speed of subsidence trends over time; and a coupled response intensity subvector, used to describe the structural unit's sensitivity to surrounding disturbances, such as mining activities or external inputs such as water level fluctuations. This multi-subfeature splicing method gives the structural unit capsule a richer expressive power, facilitating more detailed discrimination and grading in subsequent risk assessments.

[0067] Step S40: Based on the structural unit capsule, generate the settlement risk level and spatial distribution of each monitoring point.

[0068] Specifically, the final state vector of the structural unit capsule is compared and mapped with the preset risk level classification rules to obtain the risk interval corresponding to each eigenvalue in the structural response area; according to the distribution position and information contribution ratio of each monitoring point in the corresponding structural unit, the risk interval is reflected to each monitoring point, and a risk heat map is generated in two-dimensional space based on the coordinates of the monitoring points to represent the settlement risk level of each monitoring point and its spatial distribution.

[0069] In an embodiment of the present invention, the output state vector of the structural unit capsule is composed of multiple sub-vectors, which respectively represent the multi-dimensional response characteristics of the structural unit in terms of settlement amplitude, change rate and structural coupling response. In order to convert these high-dimensional semantic information into feasible risk level labels, it is necessary to first establish a set of risk level classification rules that match the mining engineering experience. The rules can be formulated based on historical settlement data, existing geological disaster level standards, regulatory requirements, etc., and usually divide risks into five or three levels, and set a threshold range or typical feature pattern for each type of interval. By comparing and mapping the various response characteristics output by the structural unit capsule with the above rules, the settlement risk interval in which the structural unit is located can be determined.

[0070] However, because capsule networks employ a many-to-many dynamic routing approach, a single monitoring point often contributes to the state of multiple structural units. Therefore, it's not possible to simply assign a structural unit's risk level to a specific monitoring point. Instead, it's necessary to further consider the monitoring point's participation in multiple structural units and perform a contribution-based risk mapping. To this end, the routing weight of each monitoring point in the structural unit it contributes to is calculated and used as the proportion of its risk information attributed. Ultimately, the subsidence risk level for that monitoring point is derived by weighted fusion of the risk intervals of multiple structural units, ensuring the spatial continuity and physical rationality of the assessment results.

[0071] At the spatial level, to visualize risk levels, the risk level of each monitoring point needs to be mapped back to its original coordinate position in two-dimensional geographic space, thereby generating a spatial heat map of subsidence risk. This map uses a color gradient to express changes in risk intensity, which can not only visually demonstrate the concentrated distribution of high-risk areas but also help identify abnormal risk in boundary transition zones. To further enhance the continuity of spatial distribution, interpolation algorithms such as the inverse distance weighted method or the Gaussian kernel function can be introduced based on the monitoring point data to perform continuous surface interpolation of risk levels, generating a smoother risk heat zone distribution map.

[0072] Preferably, after obtaining the settlement risk level of each monitoring point and its spatial distribution, the method further includes: based on the response consistency judgment between the structural unit capsule and the monitoring points it contains, identifying the monitoring points whose response deviation exceeds a preset threshold, and marking them as abnormal response points; identifying the distribution characteristics of each abnormal response point in the structural boundary area to construct a risk drift trend map or generate a list of intervention area recommendations; wherein the response deviation is determined by the difference between the actual risk level of the monitoring point and the risk value predicted by the structural unit capsule state vector.

[0073] In an embodiment of the present invention, each structural unit capsule has integrated its associated point-level capsule information through the aforementioned dynamic routing process to form a high-dimensional state vector, which represents the composite response pattern of its overall settlement behavior. At the same time, each monitoring point also has its own independently generated settlement risk level value, which is derived from its own observation data and feature contribution in the capsule network. In order to determine whether there is an abnormal deviation between the two, it is necessary to map the state vector output by the structural unit capsule to the risk level space to obtain the predicted risk value of the structural unit in the direction of the corresponding monitoring point. The difference between this predicted value and the actual risk level of the monitoring point is calculated to obtain a numerical deviation index.

[0074] If the response deviation of a monitoring point exceeds a preset threshold (this threshold can be based on historical data distribution or experience), the point is determined to have behavioral anomalies within the current structural unit. Such points are marked as "abnormal response points," indicating that their subsidence trend is inconsistent with the overall pattern of the structural unit to which they belong, and may be subject to additional influences such as local lithologic changes, sudden water level disturbances, or mining activities.

[0075] Furthermore, to reveal the spatial distribution characteristics of outliers, this method performs spatial clustering and boundary area analysis on all outlier response points, focusing on identifying whether they are clustered along the edges of structural units or across structural interfaces. If most outliers are found to be located at the edges of the structure, there may be a "risk drift" phenomenon, that is, the original risk center area has shifted spatially. Based on this phenomenon, a risk drift trend map can be further generated to express the potential migration path of risk hotspots. At the same time, based on the density and spatial connectivity of outliers, high-priority intervention areas can be automatically screened out, and a list of recommendations can be output as a basis for decision-making on intensified engineering monitoring, increased warning levels, or construction avoidance.

[0076] Figure 2 This is a system structure diagram of a mining area geological subsidence risk assessment system provided by an embodiment of the present invention. Figure 2As shown, an embodiment of the present invention provides a geological subsidence risk assessment system for mining areas, the system comprising: an acquisition unit for acquiring observation data and structural information of each monitoring point in the mining area, and constructing a point-level capsule for expressing the settlement behavior characteristics of each monitoring point based on the observation data; a processing unit for calculating the structural coupling factor between the monitoring points based on the structural information, as a weight adjustment parameter for the dynamic routing process in the capsule network; a model construction unit for inputting the point-level capsule and the structural coupling factor into a preset structural coupling capsule network, and guiding the dynamic routing mechanism to generate structural unit capsules for characterizing the settlement response characteristics of different structural units according to the weight adjustment parameter; and an evaluation unit for generating the settlement risk level and spatial distribution of each monitoring point based on the structural unit capsule.

[0077] An embodiment of the present invention further provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, enables the computer to execute the above-mentioned method for assessing geological subsidence risk in mining areas.

[0078] Those skilled in the art will appreciate that all or part of the steps in the methods described in the aforementioned embodiments can be performed by instructing the relevant hardware through a program. The program, stored in a storage medium, includes instructions for causing a microcontroller, chip, or processor to execute all or part of the steps in the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0079] The above describes in detail the optional embodiments of the present invention in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, a variety of simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the scope of protection of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner unless there is any contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention will no longer describe the various possible combinations separately.

[0080] In addition, the various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A method for assessing geological subsidence risk in mining areas, characterized in that: The method comprises: Obtaining observation data and structural information of each monitoring point in the mining area, and constructing a point-level capsule based on the observation data to express the settlement behavior characteristics of each monitoring point; Calculating a structural coupling factor between monitoring points based on the structural information as a weight adjustment parameter for a dynamic routing process in a capsule network; Inputting the point-level capsule and the structural coupling factor into a preset structural coupling capsule network, and guiding a dynamic routing mechanism to generate structural unit capsules for characterizing the settlement response characteristics of different structural units according to the weight adjustment parameter; Based on the structural unit capsule, the settlement risk level and spatial distribution of each monitoring point are generated.

2. The method for assessing geological subsidence risk in mining areas according to claim 1, characterized in that: The observation data include: Any one or more of the following: settlement rate, settlement acceleration, groundwater level disturbance amplitude, lithology coding information, and mining disturbance parameters at the monitoring point; The structural information includes any one or more of the spatial distance between monitoring points, topological relationship of fracture structure, connectivity of hydrological channels and lithologic similarity; The observation data is used to construct the initial state vector of the point-level capsule, and the structural information is used to determine the structural coupling factor to participate in the weight adjustment in the dynamic routing process.

3. The method for assessing geological subsidence risk in mining areas according to claim 1, characterized in that: Based on the observation data, a point-level capsule is constructed to express the settlement behavior characteristics of each monitoring point, including: The observation data of each monitoring point are normalized to a unified scale to obtain the corresponding multi-dimensional features; Based on the capsule encoding rule, the normalized multidimensional features are encoded into a composite structure containing a state vector and a posture matrix to obtain a point-level capsule for expressing the settlement behavior characteristics of each monitoring point; The state vector is used to characterize the settlement behavior intensity and change trend of the monitoring point; The attitude matrix is ​​used to identify the response patterns of settlement behavior in different structural directions.

4. The method for assessing geological subsidence risk in mining areas according to claim 1, characterized in that: Calculating a structural coupling factor between monitoring points based on the structural information includes: Based on the structural information, a candidate connection graph is constructed between monitoring points, and pairs of monitoring points that are directly connected or physically associated on the structural path are selected as point pair sets; Taking each pair of points as a unit, the spatial coupling feature vectors between structural paths are extracted through graph embedding. Based on the spatial coupling feature vectors and the historical settlement synergistic change behavior, a learnable weighted fusion function is constructed. During the network training process, the fusion coefficients of various structural sub-factors of the weighted fusion function are adaptively adjusted to obtain a structural coupling factor for dynamic routing weight adjustment.

5. The method for assessing geological subsidence risk in mining areas according to claim 1, characterized in that: The structure-coupled capsule network includes: The input layer includes a point-level capsule encoding layer, a structural coupling expression layer, and a structural unit response layer; The structurally coupled capsule network transmits information between layers in a non-fixed connection manner through a dynamic routing mechanism; The structural coupling capsule network introduces a structural coupling factor as a control condition to replace the routing weight calculation based on similarity score in the capsule network.

6. The method for assessing geological subsidence risk in mining areas according to claim 5, characterized in that: The point-level capsule and the structural coupling factor are input into a preset structural coupling capsule network, and a dynamic routing mechanism is guided according to the weight adjustment parameter to generate structural unit capsules for characterizing the settlement response characteristics of different structural units, including: Perform transformation operations on point-level capsules to generate a set of prediction vectors for all candidate structural unit capsules; According to the weight adjustment matrix constructed by the structural coupling factor, the prediction vectors between each point-level capsule and all candidate structural unit capsules are weightedly aggregated to generate the pre-activation input of the structural unit layer; The pre-activation input is normalized based on the nonlinear compression function to obtain the final state vector of the structural unit capsule; The final state vector is composed of a plurality of sub-eigenvectors, which respectively represent the settlement amplitude distribution of the target structural unit, the settlement change rate trend and the coupling response intensity to the surrounding structural disturbance.

7. The method for assessing geological subsidence risk in mining areas according to claim 6, characterized in that: Based on the structural unit capsule, the settlement risk level and spatial distribution of each monitoring point are generated, including: Compare and map the final state vector of the structural unit capsule with the preset risk level classification rules to obtain the risk interval corresponding to each eigenvalue in the structural response area; According to the distribution position and information contribution ratio of each monitoring point in the corresponding structural unit, the risk interval is reflected to each monitoring point, and a risk heat map is generated in two-dimensional space based on the coordinates of the monitoring points to represent the settlement risk level of each monitoring point and its spatial distribution.

8. The method for assessing geological subsidence risk in mining areas according to claim 7, characterized in that: After obtaining the settlement risk level and spatial distribution of each monitoring point, the method further includes: Based on the response consistency judgment between the structural unit capsule and the monitoring points it contains, the monitoring points whose response deviation exceeds the preset threshold are identified and marked as abnormal response points; Identify the distribution characteristics of each abnormal response point in the structural boundary area to construct a risk drift trend map or generate a list of recommended intervention areas; The response deviation is determined by the difference between the actual risk level of the monitoring point and the risk value predicted by the state vector of the structural unit capsule.

9. A geological subsidence risk assessment system for mining areas, characterized by: The system comprises: An acquisition unit, configured to obtain observation data and structural information of each monitoring point in the mining area, and construct a point-level capsule based on the observation data to express the settlement behavior characteristics of each monitoring point; A processing unit, configured to calculate a structural coupling factor between monitoring points based on the structural information as a weight adjustment parameter for a dynamic routing process in a capsule network; A model building unit is used to input the point-level capsule and the structural coupling factor into a preset structural coupling capsule network, and guide a dynamic routing mechanism to generate structural unit capsules for characterizing the settlement response characteristics of different structural units according to the weight adjustment parameter; The evaluation unit is used to generate the settlement risk level and spatial distribution of each monitoring point based on the structural unit capsule.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the mining area geological subsidence risk assessment method according to any one of claims 1 to 8.

Images