Urban groundwater inrush risk assessment method based on three-dimensional grid model and XGBoost algorithm

Abstract

The invention provides an urban groundwater inrush risk assessment method based on a three-dimensional grid model and an XGBoost algorithm, and belongs to the technical field of geological disaster early warning. The method comprises the following steps: acquiring geological, environmental and historical disaster data in real time through a multi-source heterogeneous data interface; constructing a heterogeneous grid unit based on three-dimensional geological modeling, and dynamically binding parameters such as underground water level, fault activity and lithologic permeability coefficient; an XGBoost algorithm is combined with second-order Taylor expansion to optimize gradient calculation, and a nonlinear risk assessment model is constructed through Bayesian hyper-parameter tuning and SHAP value feature interpretive analysis; identifying contiguous dangerous areas based on spatial autocorrelation analysis and an adaptive threshold segmentation algorithm; and a risk thermodynamic diagram, geological section superposition and real-time early warning are realized through a 3D GIS platform. According to the method, the problems that a traditional method depends on a linear model, data is single and local risk description is insufficient are solved, and the accuracy, efficiency and engineering practicability of risk assessment are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of urban groundwater gushing risk assessment, and in particular to a method for assessing urban groundwater gushing risk based on a three-dimensional grid model and the XGBoost algorithm. Background Art

[0002] Urban groundwater gushing is a common geological disaster, which poses a serious threat to urban infrastructure, the lives and property of residents. At present, the following deficiencies mainly exist in the assessment methods for urban groundwater gushing:

[0003] Data limitations: Traditional methods rely on empirical formulas and a small amount of static data (such as borehole data), and it is difficult to reflect the spatio-temporal dynamic changes of complex geological conditions, resulting in one-sided risk assessment.

[0004] Defects of model linearization: Existing technologies mostly adopt linear regression or simple statistical models, which cannot effectively handle the non-linear relationship between geological factors and gushing risk, and the prediction error is relatively large.

[0005] Neglect of local features: Most methods evaluate the area as a whole, ignoring local geological heterogeneities such as faults and fractures, resulting in missed or misjudged high-risk areas.

[0006] Insufficient visualization and decision support: The assessment results are mostly presented in two-dimensional charts, lacking an interactive three-dimensional spatial analysis function, and it is difficult to provide intuitive guidance for engineering disaster prevention.

[0007] In view of the above problems, the present invention proposes a method that combines three-dimensional grid modeling and the XGBoost algorithm. By depicting geological heterogeneity with high-resolution grids, combining machine learning to process complex non-linear relationships, and integrating spatial statistics and dynamic visualization technologies, accurate risk assessment and real-time decision support are realized. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a method for assessing urban groundwater gushing risk based on a three-dimensional grid model and the XGBoost algorithm to solve at least the above problems.

[0009] The technical solution adopted by the present invention is as follows:

[0010] A method for assessing urban groundwater gushing risk based on a three-dimensional grid model and the XGBoost algorithm, the method comprising the following steps:

[0011] S1. Through a multi-source heterogeneous data interface, real-time obtain geological data, environmental data and historical disaster data of the research area, wherein the geological data includes the three-dimensional spatial attributes of formation lithology, fault zone distribution, and karst caves, and the environmental data includes the dynamic changes of the groundwater level, the temporal characteristics of rainfall intensity, and human activity inducing factors;

[0012] S2. Based on the geographic coordinate system, divide the study area into three-dimensional grid cells of equal size. Each cell is dynamically bound with geological attribute data, and a heterogeneous geological structure model including rock layer dip angle, shear strength, and fracture density is generated through three-dimensional geological modeling software;

[0013] S3. Extract and fuse features of the three-dimensional grid cells, extract the interaction features of the main control factors and the inducing factors, where the main control factors include the groundwater level gradient, fault activity, and lithological permeability coefficient, and the inducing factors include the cumulative rainfall effect, construction disturbance intensity, and seismic wave propagation attenuation coefficient, and optimize the input feature set using a dynamic weighting method based on feature importance;

[0014] S4. Construct a groundwater gushing risk assessment model based on the XGBoost algorithm:

[0015] S5. Input the real-time data into the trained model, output the gushing probability values of each grid cell, use the adaptive threshold segmentation algorithm, remove isolated high-risk cells according to the spatial continuity constraint, and conduct spatial autocorrelation analysis based on the Moran index to divide the contiguous dangerous areas;

[0016] S6. Interactive decision support: Through the three-dimensional geographic information system 3D GIS platform, dynamically render the risk heat map and the superposition view of the geological section, and integrate the real-time early warning module. When the volume of the dangerous area exceeds the preset safety threshold, trigger a multi-level response mechanism.

[0017] Furthermore, in step S2, the size of the three-dimensional grid cells is adaptively adjusted according to the geological structure complexity. Among them, a 1-meter-level high-resolution grid is used in the fault zone and karst area, and a 5-10-meter-level grid is used in the homogeneous rock layer area, and the grid division is dynamically coupled with the simulation results of the groundwater flow field.

[0018] Furthermore, in step S3, the dynamic weighting method is specifically as follows:

[0019] Calculate the contribution degree of each feature to the gushing risk prediction through the feature splitting gain value Gain generated during the training process of the XGBoost model, where the gain value is defined as the total gain when the feature is used as a splitting node in all trees;

[0020] The contribution degree is the percentage of the gain value of each feature in the total gain value, and the calculation formula is:

[0021]

[0022] where F is the total number of features, assign a weight of 1.2 - 1.5 times to the features with a contribution degree higher than the preset threshold, and randomly discard the redundant features by Dropout.

[0023] Furthermore, constructing a groundwater inrush risk assessment model based on the XGBoost algorithm specifically includes the following steps:

[0024] S41. Define the loss function as the binary logistic loss function:

[0025] And introduce the second-order Taylor expansion to approximate the gradient of the loss function to accelerate gradient calculation;

[0026] S42. Adopt the tree structure regularization term:

[0027]

[0028] where γ is the minimum gain threshold for leaf node splitting, with a value range of 0.1 to 1.0, λ is the L2 regularization coefficient, with a value range of 0.01 to 0.5; T is the total number of leaf nodes of a single tree, and w j is the weight value of the jth leaf node. Combine early stopping and Bayesian hyperparameter tuning to control the model complexity;

[0029] S43. Quantify the contribution degree of each feature to the inrush risk based on SHAP values and dynamically adjust the feature weights.

[0030] Furthermore, in step S4, a spatio-temporal cross-validation strategy is adopted during model training: the historical data is divided into a training set and a validation set according to the time series, and it is ensured that the data of the same spatial grid unit is strictly isolated during the training and validation phases to avoid overfitting caused by data leakage.

[0031] Furthermore, in the said step S5, the adaptive threshold segmentation algorithm is as follows:

[0032] Identify high-risk aggregation areas based on local spatial statistics, and cluster the grid cells that meet

[0033]

[0034] where x i is the risk value of the ith grid cell, is the mean of the risk values of all grid cells in the study area, s 2 is the sample variance of the risk values, Z a is the significance level threshold, and w ij is the spatial weight matrix.

[0035] Furthermore, the 3D GIS platform supports multi-dimensional comparative analysis functions, including:

[0036] Overlay and compare the real-time risk prediction results with the spatial distribution of historical inrush events;

[0037] Based on the virtual drilling function, display the correlation between the geological properties and risk values of any cross-section;

[0038] Generate an animation of the risk evolution time series to simulate the sudden outburst diffusion path and influence range.

[0039] Furthermore, it also includes the model iteration and optimization steps:

[0040] Through the online learning framework, real-time receive sensor monitoring data, and use the Focal Loss function to weight the high-risk events in the new samples, where:

[0041] The Focal Loss function is defined as FL(p t )=-a t (1-p t ) γ log(p t ), where p t is the predicted probability of the model for the sample, a t is the class weight coefficient, and γ is the focusing parameter;

[0042] The determination criterion for high-risk events is: the grid cells within 50 meters around historical sudden outburst events, or the cells with a groundwater level change rate exceeding 10 cm / h in real-time monitoring data.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1. A method for urban groundwater sudden outburst risk assessment based on a three-dimensional grid model and the XGBoost algorithm proposed by the present invention solves the problem of single data in traditional methods by integrating remote sensing, geological exploration, meteorological monitoring and human activity data and updating through a real-time interface.

[0045] 2. A method for urban groundwater sudden outburst risk assessment based on a three-dimensional grid model and the XGBoost algorithm proposed by the present invention can finely depict the local characteristics of high-risk areas such as faults and karst areas by using 1-10 meter-level dynamic grid division, thereby improving the accuracy of dangerous area identification.

[0046] 3. A method for urban groundwater sudden outburst risk assessment based on a three-dimensional grid model and the XGBoost algorithm proposed by the present invention can accelerate the gradient calculation by combining the XGBoost algorithm with the second-order Taylor expansion to improve the model training efficiency; introduce SHAP value feature interpretability analysis, and scientifically allocate the weights of key factors (such as groundwater level gradient), further improving the prediction accuracy. Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only the preferred embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0048] Figure 1 is a schematic diagram of the overall process of a method for assessing the risk of urban groundwater outburst based on a three-dimensional grid model and the XGBoost algorithm proposed in the embodiments of the present invention. Specific implementation manners

[0049] The following describes the principles and features of the present invention in conjunction with the drawings. The listed embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0050] Refer to Figure 1 , the present invention provides a method for assessing the risk of urban groundwater outburst based on a three-dimensional grid model and the XGBoost algorithm. The method includes the following steps:

[0051] S1. Through a multi-source heterogeneous data interface, geological data, environmental data, and historical disaster data of the study area are obtained in real time. The geological data includes the three-dimensional spatial attributes of formation lithology, fault zone distribution, and karst caves. The environmental data includes the dynamic changes of groundwater levels, the temporal characteristics of rainfall intensity, and human activity-induced factors;

[0052] S2. Based on the geographic coordinate system, the study area is divided into three-dimensional grid cells of equal size. Each cell is dynamically bound with geological attribute data, and a heterogeneous geological structure model including rock layer dip angle, shear strength, and fracture density is generated through three-dimensional geological modeling software;

[0053] S3. Feature extraction and fusion are performed on the three-dimensional grid cells to extract the interaction features of the main control factors and induced factors. The main control factors include the groundwater level gradient, fault activity, and lithology permeability coefficient, and the induced factors include the rainfall cumulative effect, construction disturbance intensity, and seismic wave propagation attenuation coefficient. A dynamic weighting method based on feature importance is used to optimize the input feature set;

[0054] S4. Build a groundwater outburst risk assessment model based on the XGBoost algorithm:

[0055] S5. Input the real-time data into the trained model, output the outburst probability values of each grid cell, use the adaptive threshold segmentation algorithm to eliminate isolated high-risk cells according to the spatial continuity constraint, and perform spatial autocorrelation analysis based on the Moran index to divide the contiguous hazardous areas;

[0056] S6. Interactive decision support: Through the three-dimensional geographic information system 3D GIS platform, the risk heat map and geological profile overlay view are dynamically rendered, and a real-time early warning module is integrated. When the volume of the dangerous area exceeds the preset safety threshold, a multi-level response mechanism is triggered.

[0057] Exemplarily, the method mainly includes six steps: data acquisition, three-dimensional grid division and geological modeling, feature extraction and fusion, construction of risk assessment model, model application and risk division and interactive decision support. For S1 data collection, geological data (such as stratum lithology, fault zone distribution), environmental data (such as groundwater level changes, rainfall intensity) and historical disaster data (such as past groundwater surge events) of a certain urban study area can be obtained in real time through multi-source heterogeneous data interfaces; for S2 three-dimensional grid division and geological modeling, the study area can be divided into multiple three-dimensional grid units based on the geographic coordinate system. In the fault zone and karst area, due to the complex geological structure, a 1-meter-level high-resolution grid is used; in the homogeneous rock layer area, a 5-10-meter-level grid is used. These grid units are dynamically bound to geological attribute data, and a heterogeneous geological structure model containing information such as rock layer inclination and shear strength is generated through three-dimensional geological modeling software; for S3 feature extraction and fusion, the main controlling factors (such as groundwater level gradient, fault activity) and inducing factors (such as rainfall cumulative effect, construction disturbance intensity) can be extracted from the three-dimensional grid units. The contribution of each feature to the prediction of the sudden surge risk is calculated through the feature splitting gain value generated during the training process of the XGBoost model, and higher weights are given to features with high contribution, while redundant features are discarded. For S4: Constructing a risk assessment model, a groundwater sudden surge risk assessment model can be constructed based on the XGBoost algorithm. During the model training process, the gradient of the second-order Taylor expansion approximation loss function is introduced to accelerate the calculation, and the tree structure regularization term is used to control the model complexity. In addition, the contribution of each feature to the sudden surge risk is quantified based on the SHAP value, and the feature weight is dynamically adjusted. For S5 model application and risk division, real-time data can be input into the trained model to output the sudden surge probability value of each grid unit. Adaptive threshold segmentation algorithm is used to eliminate isolated high-risk units according to spatial continuity constraints, and spatial autocorrelation analysis is performed based on the Moran index to divide the contiguous dangerous areas. For S6 interactive decision support, the risk heat map and geological profile overlay view can be dynamically rendered through the three-dimensional geographic information system (3D GIS) platform. It supports multi-dimensional comparative analysis functions, such as superimposing and comparing real-time risk prediction results with the spatial distribution of historical surge events, displaying the correlation between geological attributes and risk values ​​of any section, and generating time-series animations of risk evolution.

[0058] In step S2, the three-dimensional grid cell size is adaptively adjusted according to the geological structure complexity. Among them, a high-resolution grid with a scale of 1 meter is used in the fault zone and karst area, and a grid with a scale of 5-10 meters is used in the homogeneous rock formation area. And the grid division is dynamically coupled with the simulation results of the groundwater flow field.

[0059] Exemplarily, when conducting a groundwater gushing risk assessment for a complex geological area in a certain city, if this area contains multiple fault zones and karst areas, and there are also large areas of homogeneous rock formation areas. For the grid division of the fault zone and karst area, in these areas with complex geological structures, a high-resolution grid with a scale of 1 meter can be used. The high-resolution grid can more accurately capture the geological characteristics of the fault zone and karst area, including the minute changes in the rock formation and the distribution of fissures. This is crucial for subsequent risk assessment because these areas are often high-risk areas for groundwater gushing. For the grid division of the homogeneous rock formation area: in the homogeneous rock formation area, a grid with a scale of 5-10 meters can be used. Compared with the fault zone and karst area, the geological structure of the homogeneous rock formation area is relatively simple, so a larger grid size can be adopted. This helps to reduce the computational amount and improve the assessment efficiency. For the dynamic coupling of grid division and groundwater flow field simulation: when conducting grid division, the results of groundwater flow field simulation are considered. By dynamically coupling grid division and groundwater flow field simulation, the impact of groundwater flow on the gushing risk can be more accurately evaluated. For example, in areas with a relatively fast groundwater flow velocity, a denser grid division may be adopted to more precisely capture the changes in the water flow.

[0060] In step S3, the specific dynamic weighting method is as follows:

[0061] Through the feature split gain value Gain generated during the training process of the XGBoost model, calculate the contribution degree of each feature to the gushing risk prediction. Among them, the gain value is defined as the total gain when the feature serves as a split node in all trees;

[0062] The contribution degree is the percentage of each feature gain value in the total gain value, and the calculation formula is:

[0063]

[0064] where F is the total number of features. Assign a weight of 1.2-1.5 times to the features with a contribution degree higher than the preset threshold, and randomly discard the redundant features by Dropout.

[0065] Exemplarily, by adopting a dynamic weighting method to calculate the contribution degree of each feature to the prediction of water inrush risk and adjusting the feature weights according to the contribution degree, it is assumed that a series of main control factors and induced factors are extracted from three-dimensional grid cells, including the groundwater level gradient, fault activity, rainfall cumulative effect, etc. During the training process of the XGBoost model, the total gain (i.e., the feature split gain value) of each feature as a split node in all trees is calculated. Then, according to these gain values, the contribution degree of each feature to the prediction of water inrush risk is calculated, and features with a contribution degree higher than the preset threshold are assigned higher weights (such as 1.2 - 1.5 times), while redundant features are discarded. So as to be able to capture key features more accurately and improve the accuracy of risk assessment.

[0066] Building a groundwater inrush risk assessment model based on the XGBoost algorithm specifically includes the following steps:

[0067] S41. Define the loss function as the binary logistic loss function:

[0068] And introduce the second-order Taylor expansion to approximate the gradient of the loss function to accelerate gradient calculation;

[0069] S42. Adopt the tree structure regularization term:

[0070]

[0071] Among them, γ is the minimum gain threshold for leaf node splitting, with a value range of 0.1 to 1.0, λ is the L2 regularization coefficient, with a value range of 0.01 to 0.5; T is the total number of leaf nodes of a single tree, w j is the weight value of the jth leaf node, and combined with the early stopping method and Bayesian hyperparameter tuning to control the model complexity;

[0072] S43. Quantify the contribution degree of each feature to the water inrush risk based on the SHAP value and dynamically adjust the feature weights.

[0073] Exemplarily, by defining the loss function as the binary logistic loss function, introducing the second-order Taylor expansion to approximate the gradient of the loss function to accelerate gradient calculation, and adopting the tree structure regularization term to control the model complexity, including setting parameters such as the minimum gain threshold for leaf node splitting and the L2 regularization coefficient. At the same time, combined with the early stopping method and Bayesian hyperparameter tuning to further optimize the model. Finally, quantify the contribution degree of each feature to the water inrush risk based on the SHAP value and dynamically adjust the feature weights according to these contribution degrees. This enables the model to continuously learn and optimize during the training process, improving the accuracy and stability of risk assessment.

[0074] In step S4, a spatiotemporal cross-validation strategy is used during model training: historical data are divided into training sets and validation sets according to time series, and data from the same spatial grid unit are strictly isolated during the training and validation stages to avoid overfitting caused by data leakage.

[0075] For example, in order to ensure the generalization ability of the model and avoid overfitting, a spatiotemporal cross-validation strategy can be used for model training. For example, historical data is divided into training sets and validation sets according to time series, and data from the same spatial grid unit is strictly isolated during the training and validation stages. In this way, the model will not be exposed to the data in the validation set during training, thereby avoiding the overfitting problem caused by data leakage. Through the spatiotemporal cross-validation strategy, the performance of the model can be evaluated more accurately, and the optimal model parameters can be selected for subsequent risk assessment.

[0076] In step S5, the adaptive threshold segmentation algorithm is:

[0077] Identify high-risk clusters based on local spatial statistics and

[0078]

[0079] The grid cells are clustered, where x i is the risk value of the i-th grid unit, is the mean risk value of all grid cells in the study area, s 2 is the sample variance of risk value, Z a is the significance level threshold, w ij is the spatial weight matrix.

[0080] For example, after outputting the surge probability value of each grid unit, an adaptive threshold segmentation algorithm is used to remove isolated high-risk units and divide the contiguous dangerous areas. The algorithm identifies high-risk clusters based on local spatial statistics and clusters grid units that meet specific conditions, including the risk value of the grid unit, the mean and variance of the risk values ​​of all grid units in the study area, etc. The adaptive threshold segmentation algorithm can effectively remove isolated high-risk units and divide the contiguous dangerous areas, providing decision makers with a more intuitive risk distribution map.

[0081] The 3D GIS platform supports multi-dimensional comparative analysis functions, including:

[0082] Overlay and compare the real-time risk prediction results with the spatial distribution of historical surge events;

[0083] Based on the virtual drilling function, the correlation between the geological attributes and risk values ​​of any section is displayed;

[0084] Generate a time-series animation of risk evolution to simulate the sudden surge diffusion path and influence range.

[0085] Exemplarily, the 3D GIS platform used supports multi-dimensional comparative analysis functions, providing decision-makers with rich visualization tools and data analysis means. The real-time risk prediction results can be superimposed and compared with the spatial distribution of historical sudden surge events to display the correlation between geological attributes and risk values of any section, and generate a time-series animation of risk evolution, etc., so that decision-makers can more intuitively understand the risk distribution and change trends, and thus make more accurate decisions.

[0086] This embodiment further includes a model iteration and optimization step:

[0087] Through an online learning framework, sensor monitoring data is received in real time, and the Focal Loss function is used to weight high-risk events in new samples, where:

[0088] The Focal Loss function is defined as FL(p t )=-a t (1 - p t ) γ log(p t ), where p t is the predicted probability of the model for the sample, a t is the class weight coefficient, and γ is the focusing parameter;

[0089] The determination criterion for high-risk events is: grid cells within 50 meters around historical sudden surge events, or cells with a groundwater level change rate exceeding 10 cm / h in real-time monitoring data.

[0090] Exemplarily, in order to improve the accuracy and reliability of the model, an online learning framework can be used for model iteration and optimization. For example, through the online learning framework, sensor monitoring data can be received in real time, and the Focal Loss function can be used to weight high-risk events in new samples. High-risk events include grid cells within a certain range around historical sudden surge events or cells with a groundwater level change rate exceeding a set threshold in real-time monitoring data. By continuously updating and optimizing the model, the accuracy and real-time performance of risk assessment can be continuously improved, providing more powerful support for urban groundwater sudden surge risk warning and prevention.

[0091] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for assessing the risk of urban groundwater gushing based on a three-dimensional grid model and the XGBoost algorithm, characterized in that, The method includes the following steps: S1. Real-time obtain geological data, environmental data, and historical disaster data of the study area through a multi-source heterogeneous data interface, where the geological data includes formation lithology, fault zone distribution, and three-dimensional spatial attributes of karst caves, and the environmental data includes dynamic changes in groundwater levels, temporal characteristics of rainfall intensity, and human activity-induced factors; S2. Based on the geographic coordinate system, divide the study area into three-dimensional grid cells of equal size, dynamically bind geological attribute data to each cell, and generate a heterogeneous geological structure model including rock layer dip angle, shear strength, and fracture density through three-dimensional geological modeling software; S3. Extract and fuse features of the three-dimensional grid cells, extract interaction features of the main control factors and inducing factors, where the main control factors include groundwater level gradient, fault activity, and lithology permeability coefficient, and the inducing factors include rainfall cumulative effect, construction disturbance intensity, and seismic wave propagation attenuation coefficient, and optimize the input feature set using a dynamic weighting method based on feature importance; S4. Construct a groundwater inrush risk assessment model based on the XGBoost algorithm; S5. Input real-time data into the trained model, output the inrush probability values of each grid cell, use an adaptive threshold segmentation algorithm to remove isolated high-risk cells according to spatial continuity constraints, and perform spatial autocorrelation analysis based on the Moran index to divide contiguous hazardous areas; S6. Interactive decision support: Through a three-dimensional geographic information system (3D GIS) platform, dynamically render the risk heat map and the superimposed view of the geological section, and integrate a real-time warning module. When the volume of the hazardous area exceeds the preset safety threshold, trigger a multi-level response mechanism.

2. The method according to claim 1, wherein In step S2, the size of the three-dimensional grid cells is adaptively adjusted according to the geological structure complexity, where a 1-meter-level high-resolution grid is used for the fault zone and karst area, and a 5-10-meter-level grid is used for the homogeneous rock layer area, and the grid division is dynamically coupled with the simulation results of the groundwater flow field.

3. The method according to claim 1, wherein In step S3, the dynamic weighting method is specifically as follows: Calculate the contribution degree of each feature to the inrush risk prediction through the feature split gain value Gain generated during the XGBoost model training process, where the gain value is defined as the total gain when the feature is used as a split node in all trees; The contribution degree is the percentage of each feature gain value in the total gain value, and the calculation formula is: where F is the total number of features, assign a weight of 1.2 - 1.5 times to the features with a contribution degree higher than the preset threshold, and randomly discard redundant features through Dropout.

4. The method according to claim 1, wherein Constructing a groundwater inrush risk assessment model based on the XGBoost algorithm specifically includes the following steps: S41. Define the loss function as a binary logistic loss function: And introduce the gradient of the second-order Taylor expansion to approximate the loss function to accelerate gradient calculation; S42. Adopt a tree structure regularization term: Among them, γ is the minimum gain threshold for leaf node splitting, with a value range of 0.1 to 1.0, λ is the L2 regularization coefficient, with a value range of 0.01 to 0.5; T is the total number of leaf nodes of a single tree, and w j is the weight value of the j-th leaf node, combining early stopping and Bayesian hyperparameter tuning to control the model complexity; S43. Quantify the contribution degree of each feature to the inrush risk based on SHAP values and dynamically adjust the feature weights.

5. The method according to claim 4, wherein In step S4, a spatio-temporal cross-validation strategy is adopted during model training: Divide the historical data into a training set and a validation set according to the time series, and ensure that the data of the same spatial grid cell is strictly isolated during the training and validation phases to avoid overfitting caused by data leakage.

6. The method according to claim 1, characterized in that, In step S5, the adaptive threshold segmentation algorithm is: Identifying high-risk aggregation areas based on local spatial statistics, for satisfying Cluster the grid cells, where x i is the risk value of the i-th grid cell, is the mean of the risk values of all grid cells in the study area, s 2 is the sample variance of the risk values, Z a is the significance level threshold, w ij is the spatial weight matrix.

7. The method according to claim 1, characterized in that, In the said step S6, the 3D GIS platform supports multi-dimensional comparative analysis functions, including: Overlaying and comparing the real-time risk prediction results with the spatial distribution of historical outburst events; Based on the virtual borehole function, showing the correlation between the geological attributes and risk values of any section; Generating a time-series animation of risk evolution to simulate the outburst diffusion path and influence range.

8. The method according to claim 1, wherein It also includes the model iteration and optimization steps: Through an online learning framework, receiving sensor monitoring data in real time, and using the Focal Loss function to weight high-risk events in new samples, where: The Focal Loss function is defined as FL(p t ) = -a t (1 - p t ) γ log(p t ), where p t is the predicted probability of the model for the sample, a t is the class weight coefficient, and γ is the focusing parameter; The judgment criteria for high-risk events are: grid cells within 50 meters around historical outburst events, or cells with a groundwater level change rate exceeding 10 cm / h in real-time monitoring data.

Images