Soil settlement monitoring method and system

Abstract

The invention provides a soil settlement monitoring method and system, and belongs to the field of soil monitoring, and the method comprises the steps: dividing a monitoring region into a plurality of non-uniform three-dimensional grid units, and determining the risk levels of the three-dimensional grid units through a geological parameter critical value and a spatial proximity effect, the risk levels being high, medium and low; sensor groups are set for the three-dimensional grid units divided into the risk levels, data of the sensor groups are collected in real time, and the sensor groups specifically comprise vibrating wire settlement meters, pore water pressure sensors, throw-in type water level meters and MEMS tilt angle sensors; and inputting the real-time feature vector obtained by each three-dimensional grid unit into a trained XGBoost-Light model, and carrying out early warning and response of different levels according to the output risk levels. According to the soil settlement monitoring method and system provided by the invention, the accuracy of soil settlement monitoring is improved.

Description

Technical Field

[0001] The present application belongs to the field of soil monitoring technology, and more specifically, relates to a soil settlement monitoring method and system. Background Art

[0002] Soil settlement monitoring is a core aspect of geological engineering and geotechnical engineering. Its accuracy and timeliness directly affect engineering safety and disaster prevention. Traditional monitoring methods have the following drawbacks: Existing monitoring methods mostly use uniform grids to divide the monitoring area, ignoring the spatial differences of geological parameters and the proximity effect of three-dimensional geological bodies, resulting in one-sided risk assessment; the early warning mechanism relies on a single sensor parameter threshold and lacks analysis of the linkage effect of multi-source data, which can easily lead to false alarms or missed reports; the number of sensor groups and sampling frequency are fixed and cannot be dynamically adjusted according to the risk level, resulting in a contradiction between delayed monitoring in high-risk areas and waste of resources in low-risk areas; data processing relies on manual analysis of multi-dimensional characteristics, lacks real-time and intelligence, and cannot meet the needs of engineering safety for accurate early warning. Summary of the Invention

[0003] The purpose of this application is to provide a soil settlement monitoring method and system to improve the accuracy of soil settlement monitoring.

[0004] A first aspect of an embodiment of the present application provides a soil settlement monitoring method, comprising: S101: Divide the monitoring area into a number of non-uniform three-dimensional grid cells, and determine the risk level of the three-dimensional grid cells based on the critical value of geological parameters and spatial proximity effect, wherein the risk level is high, medium, or low; S102: Setting a sensor group for each of the three-dimensional grid cells that have been classified into risk levels, and collecting data from the sensor group in real time. The sensor group specifically includes a vibrating wire sedimentation meter, a pore water pressure sensor, a submersible water level meter, and a MEMS inclination sensor. S103: The real-time feature vector obtained from each three-dimensional grid cell is input into the trained XGBoost-Light model, and different levels of warning and response are performed according to the output risk level.

[0005] A second aspect of the embodiments of the present application provides a soil settlement monitoring system, comprising: Grid division module, used to divide the monitoring area into several uneven three-dimensional grid cells; A risk judgment module is used to determine the risk level of a three-dimensional grid cell based on a geological parameter critical value and a spatial proximity effect, wherein the risk level is high, medium, or low; The sensor allocation module is used to set up sensor groups for the three-dimensional grid cells that have been divided into risk levels and collect data from the sensor groups in real time. The sensor groups specifically include vibrating wire sedimentation meters, pore water pressure sensors, immersion water level meters, and MEMS inclination sensors. The early warning response module is used to input the real-time feature vector obtained from each three-dimensional grid cell into the trained XGBoost-Light model, and perform different levels of early warning and response according to the output risk level.

[0006] The beneficial effects of the soil settlement monitoring method and system provided by the embodiments of the present application are: (1) A three-dimensional grid risk assessment model is constructed by combining the critical values ​​of geological parameters with spatial proximity effects. A dense grid is used for high-risk geological areas, and a sparse grid is used for medium- and low-risk areas. Combining the number of risk points with the distance attenuation correction mechanism, this method breaks through the extensive monitoring mode of the traditional uniform grid. This method can accurately locate high-risk geological bodies and their impact ranges, dynamically match the monitoring density with the actual risk level of the soil body, and resolve the contradiction between the lack of monitoring focus and the waste of resources in the traditional method. (2) Differentiate the number of sensor groups and sampling frequency according to the risk level: high-risk units improve data reliability through cross-validation of multiple sensor groups, and high-frequency sampling can capture risk mutation signals in a short time scale; medium-risk units achieve abnormality mutual detection through basic redundant configuration; low-risk units use lightweight single-group configuration to reduce costs and improve the overall operation efficiency of the system; (3) A real-time vector containing multi-dimensional features such as settlement-pressure correlation and water level-inclination linkage is constructed. The nonlinear modeling capability of the XGBoost-Light model is used to automatically analyze the physical coupling laws between parameters. This model improves the accuracy of risk identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0008] Figure 1 A schematic diagram of a flow chart of a soil settlement monitoring method provided in one embodiment of the present application; Figure 2 This is a structural block diagram of a soil settlement monitoring system provided in one embodiment of the present application. DETAILED DESCRIPTION

[0009] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0010] In order to make the purpose, technical solutions and advantages of this application clearer, specific embodiments will be described below with reference to the accompanying drawings.

[0011] Please refer to Figure 1 , Figure 1 A flow chart of a soil settlement monitoring method provided in one embodiment of the present application includes: S101: Divide the monitoring area into a number of non-uniform three-dimensional grid cells, and determine the risk level of the three-dimensional grid cells by using the critical value of geological parameters and the spatial proximity effect. The risk level is divided into three parts: high, medium and low.

[0012] In this embodiment, according to the geological assessment results, a non-uniform grid is divided on the plane projection map of the monitoring area: For high-risk geological areas, such as soft soil distribution areas and fault zones, the grid side length can be set to 5~10m; For medium- and low-risk areas, such as dense hard soil layers, the grid side length can be set to 15-50m; In this embodiment, the plane coordinates (X, Y) are used to uniquely identify the grid, such as X01Y05, where X01 indicates that the grid is numbered 01 in the X-axis direction, i.e., the grid is located in the first partition in the X-axis direction; and Y05 indicates that the grid is numbered 05 in the Y-axis direction, i.e., the grid is located in the fifth partition in the Y-axis direction. The geological assessment results must be completed before construction. Risk zoning results are generated through processes such as current situation investigation, data analysis, and model calculation. The assessment results are usually presented in the form of thematic maps, such as geological hazard zoning maps and risk control area distribution maps. High-risk areas and medium- and low-risk areas are clearly marked to provide a direct basis for plane grid division. Initial assessments of geological risk are usually based on surface surveys and planar monitoring data. Surface surveys include geological drilling, geophysical exploration, and topographic mapping, while planar monitoring data include GPS surface displacements and groundwater level distribution. These data are more easily integrated on a planar scale to form preliminary risk zoning. In this embodiment, the depth layers are divided according to the soil layer interfaces in the geological survey report, such as the fill and clay interfaces. The sensitive layer is divided into thin layers of 2 to 5 meters, and the stable layer is divided into thick layers of 10 to 20 meters. In this embodiment, the sensitive layer includes highly compressible soft soil layers, weak interlayers, loose sand layers, liquefied soil layers, and organic soil layers, and the stable layer includes dense hard soil layers, dense hard soil layers, and low compressibility clay layers; In this embodiment, each layer is identified by the depth layer number Z. For example, Z1 is the surface layer and Z2 is the soft soil layer; The planar assessment is only based on the comprehensive geological conditions within a certain average depth range below the ground surface, without refining the risk differences of soil layers at different depths. The three-dimensional grid needs to further divide each planar grid into multiple layers according to depth. For example, layer Z01 is 0 - 5 meters, layer Z02 is 5 - 10 meters, and layer Z03 is 10 - 20 meters. It is necessary to separately evaluate the risks according to the geological parameters of each depth layer; Each of the above planar grids can serve as the "base" of the three-dimensional grid. Subsequently, different depth units are divided vertically according to geological stratification (such as shallow soft soil, middle - layer clay, deep - layer bedrock), forming a three - dimensional structure of planar grid + vertical stratification, such as X01Y05Z1. This structure can not only retain the differences in planar risks but also flexibly incorporate depth data; In this embodiment, the specific steps for determining the risk level of three-dimensional grid units through geological parameter critical values and spatial proximity effects are as follows: Set binary critical values for each geological parameter. The threshold values can be determined by referring to standards such as the Code for Design of Building Foundation. The specific triggering conditions are as follows: void ratio e > eth, groundwater depth h < hth, compression modulus E < Eth, where eth, hth, and Eth are all threshold values; In this embodiment, obtain the void ratio, groundwater depth, and compression modulus of the 8 boundary corner points of the three - dimensional grid unit, and select the mean value of the void ratios of the 8 boundary corner points, the minimum groundwater depth, and the minimum compression modulus as the initial selection values for the above triggering; The void ratio reflects the degree of compaction of the soil mass. In three - dimensional space, it usually shows continuous and gradual change characteristics. The influence of local single - point anomalies on the whole is relatively small. Taking the mean value of the 8 corner points can weaken accidental errors and reflect the overall average state of the void ratio within the three - dimensional grid unit, meeting the evaluation requirements for the comprehensive characteristics of the soil mass in engineering; The lower the groundwater depth, the higher the risks of seepage failure and soil softening, etc. Taking the minimum value can capture the most unfavorable working conditions within the grid, meeting the conservative principle of engineering risk assessment; The compression modulus is the ability of the soil mass to resist deformation. The smaller its value, the higher the risk of foundation settlement. Taking the minimum value can lock the soil characteristics with the strongest compressibility within the unit, ensuring the safety of foundation design; In this embodiment, assign fixed risk points to the parameters that trigger risks. If triggered by the void ratio, add 15 points; if triggered by the groundwater depth, add 10 points; if triggered by the compression modulus, add 12 points; if not triggered, add 0 points; In this embodiment, for each three-dimensional grid unit, the risk points of all trigger parameters are accumulated to form an initial risk point number Rinit, where Rinit=sum(trigger risk point number); For example, if the triggered porosity ratio of a certain three-dimensional grid cell is 15 points, the triggered groundwater level depth is 10 points, and the untriggered compression modulus is 0 points, then Rinit is 25 points; At the same time, considering that the geological bodies near the 3D grid unit will affect the stability of the 3D grid unit, the initial number of risk points can be modified according to the risk level of the adjacent geological bodies.

[0013] Specifically, a 3D geological body model can be constructed, its spatial range can be defined, and the number of spatial proximity effect correction points can be added to the 3D grid cells adjacent to the geological body according to the distance attenuation rule: When the shortest distance between the two is ∈(0,dmin), if the geological body belongs to a high-risk geological body, 10 points are added; if the geological body belongs to a medium-risk geological body, 5 points are added; if the geological body belongs to a low-risk geological body, 3 points are added; if the geological body belongs to a special geological body, 8 points are added; When the shortest distance between the two is ∈[dmin,dmax), if the geological body belongs to a high-risk geological body, 5 points are added; if the geological body belongs to a medium-risk geological body, 3 points are added; if the geological body belongs to a low-risk geological body, 1 point is added; if the geological body belongs to a special geological body, 4 points are added; When the shortest distance between the two is ∈[dmax,∞), if the geological body belongs to a high-risk geological body, then add 1 point; if the geological body belongs to a medium-risk geological body, then add 1 point; if the geological body belongs to a low-risk geological body, then add 0 points; if the geological body belongs to a special geological body, then add 2 points. Among them, dmin and dmax are distance thresholds, which need to be set according to the specific environment; High-risk geological bodies include but are not limited to active faults, liquefiable sand layers, and thick soft soil areas. Most of them are highly active or physically unstable geological bodies. They directly threaten engineering safety at close range. The risks are sudden and highly destructive, and increase with distance. Although they are beyond the main impact range, the residual risks under extreme working conditions must be considered. Therefore, the number of correction points is set to a high starting point and a steep decay. Medium-risk geological bodies include but are not limited to weak interlayers, karst development areas, and areas with strong groundwater permeability. They are characterized by progressive risks and have local effects at close range, requiring targeted engineering measures. As the distance increases, the risk gradually decreases to the background level, and the number of correction points is linearly attenuated to adapt to conventional design requirements; Low-risk geological bodies include but are not limited to medium-dense sand layers, intact bedrock outcrops, and slightly collapsible loess. Their inherent mechanical properties are stable, and only minor local defects require attention at close range. Beyond the dmax, they are considered to have no significant risk, and the correction points are only retained to provide minimum warnings to avoid excessive defense. Special geological bodies include but are not limited to artificial backfill, karst caves, and mined-out areas. They are concealed and complex. Short-range risks are difficult to completely eliminate due to the uncertainty of the spatial morphology of the geological bodies. The intensity of physical effects decreases with increasing distance. However, due to the influence of the correlation between the regional geological environment, a certain correction value must still be retained to indicate the necessity of comprehensive exploration. The shortest distance between the two is the minimum value of the three-dimensional Euclidean distance between the center point coordinates of the three-dimensional grid unit and the surface of the geological model; In this embodiment, the coordinates of the center point of the three-dimensional grid unit are the average of the coordinates of the eight boundary corner points; The geological body model is a product of digital modeling of the spatial form, boundary range and physical properties of structures with specific geological attributes in the underground three-dimensional space, such as soft soil lenses, fault zones, karst areas, mined-out areas, etc. In this embodiment, if the corrected number of points Rnew∈[Rmax,∞), the three-dimensional grid cell is classified as high risk; if Rnew∈[Rmin,Rmax), the three-dimensional grid cell is classified as medium risk; if Rnew∈[0,Rmin), the three-dimensional grid cell is classified as low risk, where Rmin and Rmax are point number thresholds; S102: Setting a sensor group for the three-dimensional grid cells divided into risk levels and collecting data from the sensor group in real time. The sensor group specifically includes a vibrating wire sedimentation meter, a pore water pressure sensor, a submersible water level meter, and a MEMS inclination sensor.

[0014] In this embodiment, the surface settlement and settlement rate are obtained by a vibrating wire settlement meter, which can directly reflect the vertical deformation of the soil and is the core indicator of settlement monitoring; The pore water pressure and pressure change rate are obtained through the pore water pressure sensor, which can quantify the soil pore water pressure fluctuation and determine the effective stress change and seepage stability; The depth and amplitude of groundwater level are obtained through the immersion water level meter, which can more intuitively monitor the groundwater dynamics and indirectly reflect the soil saturation and load changes; The MEMS inclination sensor can obtain the inclination angle and inclination change rate, capture the horizontal displacement or tilt trend of the soil, and identify signs of sliding at an early stage; In this embodiment, it is also necessary to construct the sedimentation-pressure correlation coefficient and the water level-inclination regression slope; The settlement-pressure correlation coefficient quantifies the linear correlation between settlement and pore water pressure, revealing intrinsic physical mechanisms such as soil consolidation and stress transfer, and verifying whether the monitoring data conforms to basic theories such as the effective stress principle. Its core role is to screen abnormal data through parameter linkage analysis, identify sudden changes in the non-physical laws of sensors, avoid misjudgment of single parameters, and provide multi-dimensional collaborative verification for risk warnings, enhancing the ability to identify risks such as soil compression and structural deformation. The water level-inclination regression slope quantifies the impact of water-soil interaction on structural stability by constructing a dynamic response relationship between water level changes and structural inclination. This parameter can effectively identify early signals of progressive damage such as soil softening and foundation deformation. Slope anomalies can be used to determine whether the monitored object has entered the plastic deformation stage or has local damage. The above-mentioned real-time surface settlement amount, settlement rate, pore water pressure, pressure change rate, groundwater level depth, water level amplitude, inclination angle, inclination change rate, settlement-pressure correlation coefficient, and water level-inclination regression slope are combined to obtain a real-time feature vector; In this embodiment, for high-risk three-dimensional grid cells, the sensor group is set to 3 groups, and data is collected every 10 minutes. For medium-risk three-dimensional grid cells, the sensor group is set to 2 groups, and data is collected every 30 minutes. For low-risk three-dimensional grid cells, the sensor group is set to 1 group, and data is collected every 60 minutes. High-risk three-dimensional grid cells have a high probability of subsidence and potentially serious consequences. Cross-validation of multiple sensor groups is required to reduce the risk of single-point failures. If a sensor group outputs abnormal data due to environmental interference, the consistency of the data from the other two groups can be used to eliminate the misjudgment. Disaster precursors within high-risk three-dimensional grid cells are often accompanied by short-term mutations. High-frequency sampling can capture minute-level trends and create a time window for emergency response. For medium-risk 3D grid cells, a balance between monitoring accuracy and cost is required. Two sets of sensors can identify obvious anomalies through pairwise mutual inspection, while reducing hardware and maintenance costs. Changes in medium-risk areas are usually gradual, and a 30-minute interval can capture daily and weekly trend changes while reducing data transmission and storage pressure. Low-risk 3D grid cells focus on trend monitoring. A single set of sensors can meet basic data collection needs, significantly reducing hardware procurement, installation, and maintenance costs. They are also suitable for background data collection in long-term stable areas, providing a comparison benchmark for high- and medium-risk 3D grid cells. Low-risk areas are mainly monitored based on long-term trends, and the rate of change is usually lower than the sensor noise level. 60-minute sampling can meet the needs of interannual change analysis while minimizing energy consumption and costs.

[0015] S103: The real-time feature vector obtained from each three-dimensional grid cell is input into the trained XGBoost-Light model, and different levels of warning and response are performed according to the output risk probability.

[0016] In this example, XGBoost-Light can reduce the model parameter size and memory usage, meeting the computing power constraints of edge computing nodes. Its algorithm complexity is lower than that of traditional deep learning, and it can achieve fast training and inference on hardware with limited floating-point computing capabilities, which meets the monitoring system's demand for lightweight deployment. XGBoost-Light excels at processing nonlinear relationships and feature interactions for structured monitoring data such as settlement, pressure, water level, and inclination. Through its gradient boosting mechanism, the model can automatically capture the linkage patterns of multi-sensor data, avoiding the need to manually construct complex physical equations and improving the intelligence of risk assessment. In actual monitoring scenarios, data from high-risk three-dimensional grid cells may exhibit small sample sizes due to sporadic disasters. XGBoost-Light uses regularization to prevent overfitting and maintain generalization capabilities even when data is limited. Furthermore, the model supports incremental training, dynamically updating the model as monitoring data accumulates, adapting to the time-varying characteristics of the geological environment. The specific steps for training the XGBoost-Light model are: Collect 10-dimensional real-time feature vectors corresponding to different risk states in historical monitoring data to form a data set D = {(xi,yi)}, where xi is the real-time feature vector and yi is the risk label; Eliminate abnormal data caused by sensor failure, such as sudden changes in settlement to negative numbers and pressure values ​​exceeding physical limits; Geological experts or based on historical disaster events, mark yi as three integer levels: if yi is 0, it means low risk, the data shows no significant changes, and all parameters are stable; if yi is 1, it means medium risk, and the data shows an abnormal trend; if yi is 2, it means high risk, and the data corresponds to the state of disaster precursors or historical disasters. The data is divided into training set, test set, and validation set in a ratio of 7:2:1. The training set is used to learn the correspondence between features and risk levels; the test set is used to test the model during training and adjust parameters such as tree depth and learning speed to prevent overfitting; the validation set is used to ultimately evaluate the generalization ability of the model. Configure the model for a multi-classification task and set the initial parameters of the model: the tree depth is 5, the learning rate is 0.1, the number of trees is 100, the sub-sample is 0.8, and the regularization coefficient is 0.1; The training set is input to iteratively generate a decision tree. The test set is used to monitor the training effect and trigger the early stopping mechanism to prevent overfitting. After training is completed, the validation set is used to evaluate the classification accuracy, high-risk recall rate, and feature importance. Deploy the trained model to the edge computing node to input a 10-dimensional real-time feature vector to obtain a single risk level; In this embodiment, for a high-risk three-dimensional grid unit, it has three sensor groups and collects data every 10 minutes, that is, three sets of data are collected at the same time, and ((3×T) / 10) data within the collection time T are input into the trained XGBoost-Light model to obtain the risk level corresponding to each data. The ratio of the number of different risk levels to the total number is calculated, and the risk level with the largest proportion is selected as the real-time risk level of the three-dimensional grid unit at that moment; For a medium-risk 3D grid cell, it has two sets of sensors and collects data every 30 minutes. That is, two sets of data are collected at the same time. ((2×T) / 30) data within the collection time T are input into the trained XGBoost-Light model to obtain the risk level corresponding to each data point. The ratio of the number of different risk levels to the total number is calculated, and the risk level with the largest proportion is selected as the real-time risk level of the 3D grid cell at that moment. For a low-risk 3D grid cell, which has one set of sensors and collects data every 60 minutes, (T / 60) pieces of data are collected within time T. These pieces of data are input into the trained XGBoost-Light model to obtain the risk level corresponding to each piece of data. The ratio of the number of different risk levels to the total number of risk levels is calculated, and the risk level with the largest proportion is selected as the real-time risk level of the 3D grid cell at that moment. Where T is the time threshold, which is a multiple of 60; In this embodiment, if the real-time risk level is high, a red alert is issued and the following measures must be taken: immediately halt all construction activities in the risk area and evacuate personnel and equipment; initiate emergency reinforcement projects, such as driving anti-slip piles and laying anchor lattices, and simultaneously implement dewatering or recharge to control groundwater; use Beidou displacement monitoring combined with drone aerial surveys to conduct millimeter-level real-time dynamic monitoring, updating settlement data every 5 minutes; The soil has entered a state of imminent sliding, and traditional monitoring methods can no longer meet real-time requirements. Emergency engineering measures can quickly increase the soil's anti-sliding strength and offset the driving force of settlement; If the real-time risk level is medium, a yellow alert is issued and the following measures must be taken: Initiate manual on-site verification, such as re-leveling and checking for cracks on the soil surface; implement preventive reinforcement in areas with potential settlement risk, such as shallow grouting to improve soil quality and optimize drainage systems; If the soil shows abnormal trends, it may enter the plastic deformation stage. At this time, manual verification can eliminate sensor errors and confirm the actual settlement risk; If the real-time risk level is low, a green alert is issued and the following measures need to be taken: Regularly analyze data trends and compare with historical baselines; The soil is in a stable state with no obvious settlement risk. By continuously accumulating data, a more accurate normal state model can be established, providing a benchmark for subsequent anomaly identification.

[0017] Corresponding to a soil settlement monitoring method of the above embodiment, Figure 2 This is a structural block diagram of a soil settlement monitoring system provided in one embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown. Figure 2 , specifically including: Grid division module, used to divide the monitoring area into several uneven three-dimensional grid cells; A risk judgment module is used to determine the risk level of a three-dimensional grid cell based on a geological parameter critical value and a spatial proximity effect, wherein the risk level is high, medium, or low; The sensor allocation module is used to set up sensor groups for the three-dimensional grid cells that have been divided into risk levels and collect data from the sensor groups in real time. The sensor groups specifically include vibrating wire sedimentation meters, pore water pressure sensors, immersion water level meters, and MEMS inclination sensors. The early warning response module is used to input the real-time feature vector obtained from each three-dimensional grid cell into the trained XGBoost-Light model, and perform different levels of early warning and response according to the output risk level.

[0018] The above are only specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A soil settlement monitoring method, characterized in that: Including: S101: Divide the monitoring area into several uneven three-dimensional grid cells, and determine the risk levels of the three-dimensional grid cells through geological parameter critical values and spatial proximity effects. The risk levels are high, medium, and low. S102: Set up sensor groups for the three-dimensional grid cells with determined risk levels, and collect data from the sensor groups in real time. The sensor groups specifically include vibrating wire settlement gauges, pore water pressure sensors, submersible water level gauges, and MEMS tilt sensors. S103: Input the real-time feature vectors obtained for each three-dimensional grid cell into the trained XGBoost-Light model, and issue warnings and responses of different levels according to the output risk levels.

2. A soil settlement monitoring method according to claim 1, characterized in that: When dividing the uneven three-dimensional grid cells, divide two-dimensional grids on the plane projection map of the monitoring area according to the geological assessment results: For high-risk geological areas, the grid side length can be set to 5 - 10 m. For medium- and low-risk areas, the grid side length can be set to 15 - 50 m. Use plane coordinates (X, Y) to uniquely identify the grids.

3. A soil settlement monitoring method according to claim 1, characterized in that: When dividing the uneven three-dimensional grid cells, divide depth layers according to the soil layer interfaces in the geological exploration report: Use thin layers of 2 - 5 m for sensitive layers. Use thick layers of 10 - 20 m for stable layers. Use depth layer numbers Z to identify each layer. The sensitive layers include highly compressible soft soil layers, weak interlayers, loose sand layers, liquefied soil layers, and organic soil layers, and the stable layers include dense hard soil layers, dense hard soil layers, and low compressibility clay layers.

4. A soil settlement monitoring method according to claim 1, characterized in that: The specific steps for determining the risk levels of three-dimensional grid cells through geological parameter critical values and spatial proximity effects are as follows: Set binary critical values for each geological parameter, and the triggering conditions are as follows: void ratio e > eth, groundwater depth h < hth, compression modulus E < Eth, where eth, hth, and Eth are all thresholds. Assign fixed risk points to the parameters that trigger risks. If triggered by the void ratio, add 15 points; if triggered by the groundwater depth, add 10 points; if triggered by the compression modulus, add 12 points; if not triggered, add points. For each three-dimensional grid cell, accumulate the risk points of all triggered parameters to form an initial risk point. Construct a three-dimensional geological body model, define its spatial range, and for the three-dimensional grid cells of adjacent geological bodies, increase the spatial proximity effect according to the distance attenuation rule to correct the initial risk points, and obtain the corrected points Rnew. If the corrected points Rnew ∈ [Rmax, ∞), then divide this three-dimensional grid cell into high risk; if Rnew ∈ [Rmin, Rmax), then divide this three-dimensional grid cell into medium risk; if Rnew ∈ [0, Rmin), then divide this three-dimensional grid cell into low risk, where Rmin and Rmax are point thresholds.

5. A soil settlement monitoring method according to claim 4, characterized in that: The specific steps for increasing the spatial proximity effect to correct points according to the distance attenuation rule are as follows: When the shortest distance between the two is ∈(0, dmin), if the geological body belongs to a high-risk geological body, 10 points are added; if the geological body belongs to a medium-risk geological body, 5 points are added; if the geological body belongs to a low-risk geological body, 3 points are added; if the geological body belongs to a special geological body, 8 points are added; wherein the shortest distance between the two is the minimum value of the three-dimensional Euclidean distance between the center point coordinates of the three-dimensional grid unit and the surface of the geological body model; When the shortest distance between the two is ∈[dmin,dmax), if the geological body belongs to a high-risk geological body, 5 points are added; if the geological body belongs to a medium-risk geological body, 3 points are added; if the geological body belongs to a low-risk geological body, 1 point is added; if the geological body belongs to a special geological body, 4 points are added; When the shortest distance between the two is ∈[dmax,∞), if the geological body belongs to a high-risk geological body, 1 point is added; if the geological body belongs to a medium-risk geological body, 1 point is added; if the geological body belongs to a low-risk geological body, 0 point is added; if the geological body belongs to a special geological body, 2 points are added; Among them, dmin and dmax are distance thresholds.

6. A soil settlement monitoring method according to claim 1, characterized in that: Obtain the void ratio, groundwater depth, and compression modulus of the eight boundary corner points of the three-dimensional grid unit, and select the average void ratio, minimum groundwater depth, and minimum compression modulus of the eight boundary corner points as the initial values ​​of the trigger condition. The coordinates of the center point of the three-dimensional grid unit are the average of the coordinates of the eight boundary corner points.

7. A soil settlement monitoring method according to claim 1, characterized in that: For high-risk three-dimensional grid cells, the sensor group is set to 3 groups, and data is collected every 10 minutes; for medium-risk three-dimensional grid cells, the sensor group is set to 2 groups, and data is collected every 30 minutes; for low-risk three-dimensional grid cells, the sensor group is set to 1 group, and data is collected every 60 minutes; the surface settlement and settlement rate are obtained by a vibrating wire sedimentation meter, the pore water pressure and pressure change rate are obtained by a pore water pressure sensor, the groundwater level depth and water level variation are obtained by a immersion water level meter, the inclination angle and inclination change rate are obtained by a MEMS inclination sensor, and the settlement-pressure correlation coefficient and water level-inclination regression slope are constructed. The above 10-dimensional features are combined to obtain a real-time feature vector.

8. A soil settlement monitoring method according to claim 1, characterized in that: The steps to obtain the real-time risk level of a 3D grid cell are: For high-risk 3D grid cells, ((3×T) / 10) pieces of data are collected within time T; For medium-risk 3D grid cells, ((2×T) / 30) pieces of data are collected within time T; For low-risk 3D grid cells, (T / 60) pieces of data are collected within time T; The data obtained from different risk 3D grid cells are input into the trained XGBoost-Light model to obtain the risk level corresponding to each data; Calculate the ratio of the number of different risk levels to the total number, and select the risk level with the largest proportion as the real-time risk level of the three-dimensional grid unit at that moment; Where T is the time threshold, which is a multiple of 60.

9. A soil settlement monitoring method according to claim 8, characterized in that: The specific steps to take corresponding measures based on the real-time risk level are: If the real-time risk level is high, a red alert will be issued: all construction activities in the risk area will be immediately stopped, and personnel and equipment will be evacuated; emergency reinforcement projects will be initiated, and precipitation or recharge will be implemented simultaneously to control groundwater; Beidou displacement monitoring combined with drone aerial surveys will be used for millimeter-level real-time dynamic monitoring, with settlement data updated every 5 minutes; If the real-time risk level is medium, a yellow warning is issued: manual on-site verification is initiated; Implement preventive reinforcement in areas with potential subsidence risk; If the real-time risk level is low, a green alert is issued: regularly analyze data trends and compare them with historical baselines.

10. A soil settlement monitoring system, characterized in that: include: Grid division module, used to divide the monitoring area into several uneven three-dimensional grid cells; A risk judgment module is used to determine the risk level of a three-dimensional grid cell based on a geological parameter critical value and a spatial proximity effect, wherein the risk level is high, medium, or low; The sensor allocation module is used to set up sensor groups for the three-dimensional grid cells that have been divided into risk levels and collect data from the sensor groups in real time. The sensor groups specifically include vibrating wire sedimentation meters, pore water pressure sensors, immersion water level meters, and MEMS inclination sensors. The early warning response module is used to input the real-time feature vector obtained from each three-dimensional grid cell into the trained XGBoost-Light model, and perform different levels of early warning and response according to the output risk level.

Images