Water-rich stratum subway foundation pit precipitation monitoring and environment-friendly three-in-one intelligent management system

Abstract

The invention discloses a three-in-one intelligent management system for rainfall monitoring and environmental protection of a subway foundation pit in a water-rich stratum, and relates to the technical field of intelligent management of foundation pit engineering. The risk granulate knowledge base module is used for performing iterative clustering on the potential risk state point cloud database through an improved coarse-grained risk granulate generation algorithm to obtain a risk granulate knowledge base; the decision module is used for judging a risk pellet to which the current high-dimensional system state vector belongs and a corresponding risk consequence, and predicting an evolution trajectory of the high-dimensional system state vector in a risk pellet space; and the control module is used for remotely regulating and controlling the rainfall and recharge equipment. According to the method, a set of intelligent immune decision normal form based on the risk granular balls is constructed by deeply fusing the granular calculation theory and foundation pit engineering risk management, and the core innovation of the method is that a set of brand-new system capable of performing multi-dimensional, dynamic and active risk pre-control is established.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent management of foundation pit engineering, and in particular to an integrated intelligent management system for monitoring dewatering and environmental protection in subway foundation pits in water-rich strata. Background Technology

[0002] In recent years, urban rail transit construction has become an important symbol of modern urban development. Among these projects, subway foundation pit engineering, especially the excavation of deep and large foundation pits in water-rich strata, is a critical link with high technical difficulty and significant safety risks. Dewatering operations are essential to ensure construction safety and control deformation of the surrounding environment. Traditional foundation pit dewatering monitoring and management systems mostly rely on a passive alarm mode that sets static thresholds for individual monitoring points (such as settlement and water level).

[0003] Currently, Chinese invention application CN202410794180.9 discloses a construction method for integrated pumping and recharge dewatering adjacent to a subway station, including the following steps: S1, construction of dewatering wells and recharge wells; S2, pumping test; S3, recharge test; S4, installation of water treatment device; S5, deployment of integrated pumping and recharge system; S6, dewatering and recharge construction of the foundation pit; S7, monitoring of dewatering and recharge of the foundation pit. This invention reduces drainage pressure, saves water resources, and lowers costs by combining dewatering and recharge construction using an integrated pumping and recharge method. All or part of the pumped groundwater is used for recharge construction, effectively reducing groundwater discharge, lowering discharge pressure, and reducing pollution of surrounding water bodies. Simultaneously, recharge water can partially or completely utilize municipal water supply, saving water resources and reducing project costs. However, current technologies suffer from limited risk identification dimensions, only able to determine whether a single point exceeds limits. They fail to identify complex risk patterns resulting from the combined effects of multiple factors (such as settlement, water level, and stress), leading to insufficient perception of systemic risks. Decision-making response is delayed; alarms are typically triggered only after a risk has occurred or is nearing a critical state, representing a reactive measure lacking proactive control capabilities. False alarms and false negatives coexist; oversensitivity to minor disturbances in data from a single measuring point easily leads to false alarms, while slow-developing, multi-point co-evolving systemic risks are prone to being missed. These issues collectively result in traditional management models falling short in efficiency, accuracy, and proactiveness when addressing foundation pit risks under complex geological conditions. Summary of the Invention

[0004] The technical problem solved by this invention is that the existing foundation pit management system has a single risk identification dimension, a lagging decision response, false alarms and missed alarms, and project experience is difficult to quantify and pass on, making it impossible to proactively and systematically deal with complex risks caused by the coupling of multiple factors.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a three-in-one intelligent management system for monitoring and protecting the environmental protection of subway foundation pits in water-rich strata, comprising a sensing module, a risk particle knowledge base module, a decision-making module, and a control module: The sensing module is used to collect and fuse multi-source heterogeneous monitoring data of the foundation pit project in real time, and dynamically form a high-dimensional system state vector describing the overall operating status. The risk sphere knowledge base module is used to iteratively cluster the potential risk state point cloud database using an improved coarse-grained risk sphere generation algorithm to obtain a risk sphere knowledge base. The risk sphere knowledge base consists of risk spheres with engineering semantic labels, risk consequences, center vectors, and radii. The decision module is used to calculate in real time the distance between the high-dimensional system state vector and the center vector of each risk particle in the risk particle knowledge base, determine the risk particle to which the current high-dimensional system state vector belongs and the corresponding risk consequences, and predict the evolution trajectory of the high-dimensional system state vector in the risk particle space. The control module is used to automatically execute the control strategy bound to the risk pellet based on the decision result of the decision module, and to remotely regulate the precipitation and recharge equipment.

[0006] Preferably, the sensing module includes an underground monitoring unit, a surface monitoring unit, a macroscopic monitoring unit, and a dynamic inspection unit: The underground monitoring unit is used to acquire soil layer data, geological parameters, retaining structure displacement, structural stress, pore water pressure and groundwater level through an automated hydrostatic level, pore water pressure gauge, inclinometer, support axial force gauge, intelligent water level gauge and flow meter. The soil layer data includes soil layer classification, soil layer position, soil layer thickness, soil layer distribution and interrelationship between soil layers. The soil layer classification includes miscellaneous fill, silty clay, sand and gravel layer and bedrock. The geological parameters include permeability coefficient and elastic modulus. The ground monitoring unit is used to acquire surface subsidence, building tilt angle, pipeline strain, and hydrological conditions through a high-precision GPS receiver, building tilt sensor, crack gauge, and pipeline strain sensor. The hydrological conditions include rainfall intensity and external water supply.

[0007] Preferably, the risk particle knowledge base module includes a database unit and a clustering unit: The database unit includes: Based on the data collected by the sensing module, a three-dimensional numerical model of fluid-structure interaction is constructed using geotechnical engineering software. A high-dimensional system state vector output by the three-dimensional numerical model of fluid-structure interaction with Monte Carlo simulation of disturbance is used to generate a potential risk state point cloud database. The fluid-structure interaction three-dimensional numerical model includes a geometric model, multiphysics coupling control equations, a material constitutive model and parameters, and boundary and initial conditions: The three-dimensional spatial geometric model integrates the geological body and the engineering entity in a unified coordinate system by performing point-to-point coordinate correspondence. The three-dimensional spatial geometric model provides an accurate solution domain for physical field calculations. The geological body includes stratigraphic structure, hydrological boundaries, and topography; The engineering entity includes the foundation pit and support structure, the three-dimensional model structure of dewatering and recharge, and the BIM model of the surrounding environment; The multiphysics field coupling control equations include the solid mechanical field and the porous medium seepage field. The control equation of the solid mechanical field is the static equilibrium equation of the soil and structure solved based on the elastoplastic mechanics theory. The control equation of the porous medium seepage field is the motion equation of groundwater in the porous medium solved based on the non-Darcy seepage model for high-speed flow. The constitutive model and parameters of the materials include soil materials, structural materials and interface elements. The constitutive model of the soil materials is a hardened soil model. The parameters of the soil materials include mechanical parameters and hydraulic parameters. The mechanical parameters include effective cohesion, effective internal friction angle, elastic modulus and Poisson's ratio. The hydraulic parameters include saturated permeability coefficient, porosity and specific yield. The constitutive model of structural materials is a linear elastic model, and the parameters of structural materials include elastic modulus, Poisson's ratio, and density. The constitutive model of the interface element is used to simulate the relative slip and friction effects at the interface between the soil material and the structural material. The parameters of the interface element include the interface friction angle, cohesion, normal and tangential stiffness. Boundary conditions and initial conditions include: Boundary conditions include mechanical boundaries and hydraulic boundaries. Mechanical boundaries include: Fixed constraints are zero displacement in all directions. The vertical boundaries around the fluid-structure interaction 3D numerical model are set as normal constraints. Normal constraints allow vertical displacement and restrict horizontal displacement. The top surface of the fluid-structure interaction 3D numerical model is a free boundary. The hydraulic boundary includes: The bottom and far-field vertical boundaries of the model are set as impermeable boundaries, the top of the aquifer at the far-field boundary is set as a constant head boundary, and the positions of the dewatering wells and recharge wells are defined as flow boundaries and time-varying head boundaries. Initial conditions include geostress equilibrium conditions and initial water field conditions. Geostress equilibrium conditions include geostress equilibrium calculations performed before any excavation and dewatering. Initial water field conditions include the groundwater level and pore water pressure distribution under natural conditions.

[0008] Preferably, the fluid-structure interaction three-dimensional numerical model closely links the solid mechanical field and the seepage field based on the effective stress principle, forming a two-way influence: The two-way influence includes the effect of seepage on deformation and the effect of deformation on seepage. The effect of seepage on deformation includes: Rainfall causes a decrease in pore water pressure. According to Terzaghi's effective stress principle, the effective stress of the soil skeleton increases accordingly. The increase in effective stress leads to soil compression, resulting in consolidation settlement and deformation of the retaining structure. The effects of deformation on seepage include: Deformation caused by excavation unloading and soil consolidation changes the void ratio of the soil, which in turn affects the permeability coefficient. The change in the permeability coefficient, in turn, changes the distribution of the seepage field. The fluid-structure interaction three-dimensional numerical model transforms the static geometric model into a dynamic construction process simulation through step-by-step construction. The transformation process includes: The initial ground stress and water pressure field are calculated. The retaining structure is identified in the fluid-structure interaction three-dimensional numerical model. The dewatering process is simulated by manually setting the flow rate parameters of the dewatering wells. After the hydraulic field reaches a new equilibrium or a specified time, excavation is simulated. After excavation to a specified depth, the support of the soil layer corresponding to the specified depth is activated and prestress is applied. The above steps are repeated until the foundation pit is excavated to the bottom. A spatiotemporal four-dimensional data field set is output, which includes stress cloud map, displacement vector or settlement cloud map, pore water pressure isosurface vector map, seepage velocity vector map, and structural internal force distribution map. The structural internal forces include bending moment, axial force, and shear force. The spatiotemporal four-dimensional data field set is subjected to Monte Carlo simulation perturbation to obtain the high-dimensional system state vector.

[0009] Preferably, generating a point cloud database of potential risk states through Monte Carlo simulation includes: The spatiotemporal four-dimensional data field set is sliced ​​in one dimension according to the time axis to obtain the high-dimensional system state vector corresponding to each time slice. Each high-dimensional system state vector represents the state characteristics of the foundation pit in a certain scenario and at a certain moment. At the same time, the state characteristics are marked with risk consequences, including safety, danger and warning. Obtain the probability distribution of the high-dimensional system state vector corresponding to each time slice, and perform cyclic random simulation of the state based on the probability distribution to generate a time series set; The risk consequences are calculated for each time series set, and each time series is assigned a risk consequence label. The risk consequence calculation logic includes: When the maximum surface settlement is greater than 60 mm, or the axial force of the support is greater than 1.2 times the conventional design value of the axial force of the support, or the maximum displacement of the retaining structure is greater than 100 mm, the risk consequence of the current time series is labeled as dangerous. When the maximum surface settlement is greater than 40 mm and less than 60 mm, or the axial force of the support is greater than 1.05 times the conventional design value of the axial force of the support, or the maximum displacement of the retaining structure is greater than 70 mm and less than 100 mm, the risk consequence label assigned to the current time series is warning. When the maximum surface settlement is less than 40 mm, the axial force of the support is less than 1.05 times the conventional design value of the axial force of the support, and the maximum displacement of the retaining structure is less than 70 mm, the risk consequence of the current time series is labeled as safe. All high-dimensional system state vectors with risk consequences labels generated in the simulation are stored in the potential risk state point cloud database. The risk consequences are categorized into three levels: Danger is the highest level, Alert is the next highest level, and Safety is the lowest level.

[0010] Preferably, the clustering unit comprises: An improved coarse-grained risk sphere generation algorithm is used to iteratively cluster the potential risk state point cloud database to obtain a risk sphere knowledge base. The risk sphere knowledge base consists of risk spheres with engineering semantic labels, risk consequences, center vectors, and radii. The improved coarse-grained risk particle generation algorithm includes: Step S1: Take out a risk particle from the head of the risk particle column to be processed, count the number of sample points belonging to each risk consequence in the risk particle, and obtain the risk consequence corresponding to the highest number of sample points as the dominant risk consequence. The initial state of the risk particle array to be processed is the potential risk state point cloud database. Splitting the risk particle array is equivalent to classifying the potential risk state point cloud database, and the sample points are the state vectors of each high-dimensional system. Step S2: Calculate the ratio of the number of sample points of the dominant risk consequence to the total number of sample points in the risk sphere to obtain the purity of the risk sphere. When the purity of the risk sphere is greater than or equal to a preset purity threshold, the risk sphere is stored in the risk sphere knowledge base. When the purity of the risk sphere is less than a preset purity threshold, a risk sphere splitting operation is performed. Step S3: Remove the split risk particles from the list of risk particles to be processed, add all the new sub-risk particles generated after splitting to the end of the list of risk particles to be processed, return and repeat steps S1 and S2 until the list of risk particles to be processed is empty, and output the risk particle knowledge base. The engineering semantic tags for risk particles include safe operation risk particles, initial uniform settlement risk particles, over-extraction-induced regional settlement risk particles, improper reinjection leading to local heave risk particles, confined water seepage risk particles, support system stress concentration risk particles, and piping burst precursor risk particles.

[0011] Preferably, the risk granule splitting operation includes the following sub-steps: Step S21: Inside the risk sphere to be split, form a heterogeneous sample set by combining all sample points that do not belong to the dominant risk consequences, calculate the arithmetic mean of all sample points in the heterogeneous sample set, and obtain the virtual heterogeneous mean point. Step S22: Form a sample set of all sample points belonging to the dominant risk consequences, calculate the high-dimensional Euclidean distance from each sample point in the sample set to the virtual outlier mean point, obtain a list of high-dimensional Euclidean distances, sort the list of high-dimensional Euclidean distances in ascending order, obtain the median distance of the list of high-dimensional Euclidean distances, obtain the sample point in the sample set that is closest to the median distance based on the list of high-dimensional Euclidean distances, and use the sample point that is closest to the median distance as the initial stable cluster center of the new risk sphere representing the dominant risk consequences that will be split off. Step S23: Based on the high-dimensional Euclidean distance list, obtain the sample points in the heterogeneous sample set that are closest to the initial stable cluster center, and obtain the initial auxiliary cluster center. Use the initial stable cluster center and the initial auxiliary cluster center as the split center. Create an empty list of sub-risk particles, the list of sub-risk particles including a list of first cluster core sub-risk particles and a list of first cluster core sub-risk particles; Traverse each sample point in the original risk sphere to be split, and calculate the high-dimensional Euclidean distance between each sample point and the initial stable cluster center and the initial auxiliary cluster center to obtain the first cluster center distance and the second cluster center distance. When the distance between the first cluster centers is less than the distance between the second cluster centers, the corresponding sample point is assigned to the list of risk particles of the first cluster center. When the distance between the first cluster centers is greater than or equal to the distance between the second cluster centers, the corresponding sample point is assigned to the list of risk particles for the second cluster centers.

[0012] Preferably, the decision-making module includes a risk pattern matching unit and a state trajectory prediction unit: The risk pattern matching unit is used to calculate the Euclidean distance between the current high-dimensional system state vector and the center vector of each risk particle in the risk particle knowledge base in real time in the high-dimensional feature space, and to determine the risk particle to which the current high-dimensional system state vector belongs and the corresponding risk consequences based on the nearest neighbor principle. The state trajectory prediction unit is used to continuously track the movement trajectory of a high-dimensional system state vector in the risk particle space based on a recurrent neural network algorithm. The movement trajectory includes the movement speed and direction, and predicts the probability that the high-dimensional system state vector will enter a higher-level risk consequence risk particle within a preset time period in the future.

[0013] Preferably, the control module includes a state trigger mode and a trajectory trigger mode; The state triggering mode includes: when the risk pattern matching unit determines that the current high-dimensional system state vector has entered the interior of a certain risk particle, the response strategy bound to the risk particle is immediately and automatically triggered and executed based on the risk particle-strategy mapping library; The trajectory triggering mode includes: when the state trajectory prediction unit predicts that the current high-dimensional system state vector will enter the interior of a certain risk particle within a preset time, the response strategy bound to the risk particle will be executed in advance based on the risk particle-strategy mapping library; The risk particle-policy mapping library includes: When the state vector of the high-dimensional system is determined to enter the subsidence risk particle of the area caused by over-pumping, the recharge compensation strategy is triggered. The recharge compensation strategy includes: automatically increasing the recharge flow of the recharge wells near the subsidence over-limit area, and at the same time reducing the pumping power of the dewatering wells within the specified range. When the state vector of the high-dimensional system is determined to enter the stress concentration risk particle of the support system, the stress release and support strengthening strategy is triggered. The stress release and support strengthening strategy includes: sending a level 2 alarm to the project management personnel, suggesting local earthwork backfilling and adding temporary supports in the corresponding area of ​​the stress concentration risk particle of the support system, and increasing the collection frequency of relevant axial force monitoring points. When the state vector of the high-dimensional system is determined to enter the risk sphere of the confined water surge precursor, an emergency locking and alarm strategy is triggered. The alarm strategy includes: immediately and automatically shutting down all dewatering wells in the risk area, starting all backup reinjection wells at full power, sending the highest level alarm to all senior management personnel of the project through sound and light, SMS and application software, and automatically locking the access control at the pit entrance.

[0014] Preferably, the real high-dimensional system state vector data generated throughout the construction process is continuously added to the potential risk state point cloud database; The improved coarse-grained risk particle generation algorithm was re-run on the supplemented database to obtain the supplemented and corrected risk particle knowledge base.

[0015] The beneficial effects of this invention are as follows: In precipitation management, it no longer involves blindly pumping water to reach the target water level. It can identify specific risk spheres caused by over- or under-pumping and automatically trigger the optimal pumping-irrigation combination strategy associated with that sphere, achieving precise, dynamic, and environmentally friendly precipitation that meets the requirements of pit operations while minimizing environmental disturbance. In monitoring and early warning, it completely changes the problems of false alarms caused by data disturbances and missed alarms caused by slow perception of systemic risks in traditional methods. Through pattern matching in high-dimensional space, it can accurately identify complex risk precursors composed of multiple factors and multiple measurement points. More importantly, through the state trajectory prediction function, the system can not only react when a risk occurs (state triggering) but also predict the evolution trend of the risk and intervene in advance (trajectory triggering). This proactive immunity capability eliminates a large number of potential risks in their infancy, realizing a fundamental shift in safety management from post-event remediation to pre-event prevention. In terms of environmental control, the system uses the deformation of the surrounding environment (such as ground subsidence and building tilt) as a key component of the high-dimensional system state vector, and generates risk spheres directly related to environmental protection, such as local uplift caused by improper recharge. Once the system state shows a tendency to enter such spheres, the control module will proactively adjust the recharge strategy to ensure the integrated achievement of environmental protection and engineering safety goals. This greatly enhances the initiative and safety of the integrated management of dewatering, monitoring, and environmental protection in water-rich strata subway foundation pits. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the basic process of a three-in-one intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata, provided as an embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Reference Figure 1 As one embodiment of the present invention, a three-in-one intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata is provided, including a sensing module, a risk particle knowledge base module, a decision-making module, and a control module: The sensing module is used to collect and fuse multi-source heterogeneous monitoring data of the foundation pit project in real time, and dynamically construct a high-dimensional system state vector describing the overall operating status. The Risk Particle Knowledge Base module is used to iteratively cluster the potential risk state point cloud database using an improved coarse-grained risk particle generation algorithm to obtain the risk particle knowledge base. The risk particle knowledge base consists of risk particles with engineering semantic labels, risk consequences, center vectors, and radii. The decision module is used to calculate the distance between the state vector of the high-dimensional system and the center vector of each risk particle in the risk particle knowledge base in real time, determine the risk particle to which the current high-dimensional system state vector belongs and the corresponding risk consequences, and predict the evolution trajectory of the high-dimensional system state vector in the risk particle space. The control module is used to automatically execute the control strategy bound to the risk particle based on the decision results of the decision module, and to remotely regulate the precipitation and recharge equipment.

[0019] This invention deeply integrates particle computing theory with foundation pit engineering risk management to construct an intelligent immune decision-making paradigm based on risk particles. Its core innovation lies in establishing a brand-new system capable of multi-dimensional, dynamic, and proactive risk pre-control.

[0020] The sensing module includes an underground monitoring unit, a surface monitoring unit, a macroscopic monitoring unit, and a dynamic inspection unit. The underground monitoring unit is used to acquire soil layer data, geological parameters, retaining structure displacement, structural stress, pore water pressure and groundwater level through an automated hydrostatic level, pore water pressure gauge, inclinometer, support axial force gauge, intelligent water level gauge and flow meter. Soil layer data includes soil layer classification, soil layer position, soil layer thickness, soil layer distribution and interrelationship between soil layers. Soil layer classification includes miscellaneous fill, silty clay, sand and gravel layer and bedrock. Geological parameters include permeability coefficient and elastic modulus. The measurements from the underground monitoring unit provide a basis for setting the stratigraphic structure, hydrological boundaries, material constitutive models, and parameters in the fluid-structure interaction three-dimensional numerical model; The ground monitoring unit is used to acquire information on surface subsidence, building tilt angle, pipeline strain, and hydrological conditions, including rainfall intensity and external water supply, through a high-precision GPS receiver, building tilt sensor, crack gauge, and pipeline strain sensor.

[0021] The ground monitoring unit's measurements provide a basis for setting the surrounding environment BIM model, boundary conditions, and initial conditions in the fluid-structure interaction three-dimensional numerical model. In this embodiment, the collected measurements are used to provide initial construction data and calibration basis for the fluid-structure interaction three-dimensional numerical model. On the other hand, the core measurements serve as key components to dynamically constitute the high-dimensional system state vector.

[0022] This ensures that the data sources constituting the state vector of the high-dimensional system are comprehensive, multidimensional, and reliable. It not only provides a foundation for real-time risk diagnosis but also offers crucial information for the construction of the initial numerical model and subsequent calibration, guaranteeing the authenticity of the entire system's digital twin.

[0023] The risk particle knowledge base module includes database units and clustering units: Database units include: Based on the data collected by the sensing module, a three-dimensional numerical model of fluid-structure interaction is constructed using geotechnical engineering software. The high-dimensional system state vector output by the three-dimensional numerical model of fluid-structure interaction with Monte Carlo simulation of disturbance is used to generate a point cloud database of potential risk states. The three-dimensional numerical model of fluid-structure interaction includes the geometric model, multiphysics coupling governing equations, material constitutive model and parameters, and boundary and initial conditions: The three-dimensional spatial geometric model integrates the geological body and the engineering entity in a unified coordinate system to achieve point-to-point coordinate correspondence. The three-dimensional spatial geometric model provides an accurate solution domain for physical field calculations. Geological bodies include stratigraphic structure, hydrological boundaries, and topography; The stratigraphic structure is a three-dimensional stratigraphic model established based on the output data of the underground monitoring unit. The three-dimensional stratigraphic model includes the layer position, thickness, spatial distribution and interrelationship of each soil layer. The hydrological boundary is used to delineate the initial spatial position of the main aquifer, impermeable layer and groundwater level. The topography includes a topographic map of surface undulations and contour lines, a river distribution map and a road distribution map. The engineering entity includes the foundation pit and support structure, the three-dimensional model structure of dewatering and recharge, and the BIM model of the surrounding environment; The foundation pit and support structure include: modeling the excavation outline and depth of the foundation pit, as well as all support components, including the thickness, depth and joint type of the diaphragm wall and bored piles, and the cross-sectional dimensions, location and anchor length and angle of each support (concrete and steel supports). The three-dimensional model structure for precipitation and recharge includes: three-dimensional solid modeling of the location, diameter and filter pipe length of each precipitation well, observation well and recharge well; The surrounding environment BIM model includes important buildings, underground pipelines (water supply, drainage, gas, communication) and existing tunnels within the impact range of the foundation pit, in the form of a BIM model with attribute information, including foundation type, material and burial depth. This model is a complex digital complex integrating multiple information from geometry, physics, and materials. It serves as a digital twin and virtual testbed for the entire intelligent system, and its accuracy and comprehensiveness are the cornerstone for subsequent risk particle generation and intelligent decision-making. This is the skeleton of the model, constructed by integrating BIM (Building Information Modeling) and GIS (Geographic Information System) technologies to ensure a high degree of accuracy and consistency in geometric information.

[0024] The multiphysics field coupling control equations include the solid mechanical field and the porous medium seepage field. The control equation of the solid mechanical field is the static equilibrium equation of the soil and structure based on the elastoplastic mechanics theory. The control equation of the porous medium seepage field is the motion equation of groundwater in the porous medium based on the non-Darcy seepage model for high-speed flow. Solid mechanics fields are used to describe the stress, strain, and displacement distribution within a fluid-structure interaction three-dimensional numerical model under the action of external forces (self-weight, excavation unloading, support force, and water pressure). They are used to calculate soil deformation, surface settlement, retaining structure displacement, and internal force changes caused by dewatering and excavation. Porous media seepage fields are used to simulate groundwater level changes, the formation of drawdown cones, the distribution of pore water pressure, and the seepage velocity field caused by dewatering well pumping and recharge well injection.

[0025] The multiphysics coupling control equations are the physics engine of the model, defining the physical phenomena that occur in the model and their interactions.

[0026] The constitutive model and parameters of the materials include soil materials, structural materials and interface elements. The constitutive model of the soil materials is a hardened soil model. The parameters of the soil materials include mechanical parameters and hydraulic parameters. The mechanical parameters include effective cohesion, effective internal friction angle, elastic modulus and Poisson's ratio. The hydraulic parameters include saturated permeability coefficient, porosity and specific yield. In this embodiment, the constitutive model of the soil material includes: for soft soil and sand in water-rich strata, a hardened soil model that can consider small strain stiffness characteristics is adopted, rather than a simple Mohr-Coulomb model, which is crucial for accurate settlement prediction; the mechanical parameters of the soil material include effective cohesion, effective internal friction angle, elastic modulus (such as E50, Eoed, Eur) under different stress paths, and Poisson's ratio; the hydraulic parameters of the soil material include saturated permeability coefficient (which may differ in the horizontal and vertical directions, i.e., anisotropy), porosity, and specific yield. The constitutive model of structural materials is a linear elastic model, and the parameters of structural materials include elastic modulus, Poisson's ratio, and density. In this embodiment, the constitutive model for the structural materials is a linear elastic model for both concrete and steel support structures. The constitutive model of the interface element is used to simulate the relative slip and friction effects at the interface between the soil material and the structural material. The parameters of the interface element include the interface friction angle, cohesion, normal and tangential stiffness. Material constitutive models and parameters are the lifeblood of fluid-structure interaction three-dimensional numerical models, giving different objects in the model realistic physical and mechanical behaviors.

[0027] Boundary conditions and initial conditions include: Boundary conditions include mechanical boundaries and hydraulic boundaries. Mechanical boundaries include: Fixed constraints are zero displacement in all directions. The vertical boundaries around the fluid-structure interaction 3D numerical model are set as normal constraints. Normal constraints allow vertical displacement and restrict horizontal displacement. The top surface of the fluid-structure interaction 3D numerical model is a free boundary. The hydraulic boundary includes: The bottom and far-field vertical boundaries of the model are set as impermeable boundaries, the top of the aquifer at the far-field boundary is set as a constant head boundary, and the positions of the dewatering wells and recharge wells are defined as flow boundaries and time-varying head boundaries. Initial conditions include geostress equilibrium conditions and initial water field conditions. Geostress equilibrium conditions include geostress equilibrium calculations performed before any excavation and dewatering. Initial water field conditions include the groundwater level and pore water pressure distribution under natural conditions.

[0028] In this embodiment, the hydraulic boundary is used to simulate the replenishment of water from a distant source and to accurately simulate the pumping and injection processes. The initial conditions are used to obtain the initial stress field generated by the soil's own weight, ensuring the initial stability of the model. The starting point and boundary constraints for solving the model are defined.

[0029] The technical path for constructing a high-precision digital twin is defined in detail. By integrating BIM and GIS, and adopting advanced constitutive models and clear boundary conditions, it is ensured that the potential risk state point cloud database generated by subsequent large-scale simulation can truly reflect the complex physical and mechanical behavior of the foundation pit project, laying a solid physical foundation for the quality of the knowledge base.

[0030] The fluid-structure interaction three-dimensional numerical model, based on the effective stress principle, closely links the solid mechanical field and the seepage field, forming a two-way influence: The two-way influence includes the effect of seepage on deformation and the effect of deformation on seepage. The effect of seepage on deformation includes: Rainfall causes a decrease in pore water pressure. According to Terzaghi's effective stress principle, the effective stress of the soil skeleton increases accordingly. The increase in effective stress leads to soil compression, resulting in consolidation settlement and deformation of the retaining structure. The effects of deformation on seepage include: Deformation caused by excavation unloading and soil consolidation changes the void ratio of the soil, which in turn affects the permeability coefficient. The change in the permeability coefficient, in turn, changes the distribution of the seepage field. In numerical software, this two-way coupling is achieved by solving a set of partial differential equations that include fluid pressure and solid displacement, ensuring the physical realism of the simulation results.

[0031] The fluid-structure interaction three-dimensional numerical model transforms the static geometric model into a dynamic construction process simulation through step-by-step construction. The transformation process includes: Calculate the initial ground stress and water pressure field, identify the retaining structure in the fluid-structure interaction three-dimensional numerical model, simulate the dewatering process by manually setting the flow rate parameters of the dewatering well, simulate excavation after the hydraulic field reaches a new equilibrium or reaches a specified time, activate the support of the soil layer corresponding to the specified depth after excavation to a specified depth, and apply prestress. Repeat the above steps until the foundation pit is excavated to the bottom of the pit. Output a spatiotemporal four-dimensional data field set, which includes stress cloud map, displacement vector or settlement cloud map, pore water pressure isosurface vector map, seepage velocity vector map and structural internal force distribution map. The structural internal forces include bending moment, axial force and shear force. Perform Monte Carlo simulation perturbation on the spatiotemporal four-dimensional data field set to obtain the high-dimensional system state vector.

[0032] In this embodiment, the construction process also includes the construction of the simulated main structure and the subsequent earthwork backfilling process; The specified depth is between 2 meters and 4 meters; The specified time refers to the time when the pore water pressure dissipates to a preset percentage (such as 95%), and the specified depth refers to the excavation depth corresponding to the installation elevation of the next support in the construction organization design plan. These rich outputs, after post-processing and data extraction, constitute a high-dimensional system state vector, which is ultimately incorporated into the potential risk state point cloud database, becoming the sole data source for the construction of the risk particle knowledge base, thereby driving the operation of the entire intelligent immune management system.

[0033] The fluid-structure interaction 3D numerical model is the digital heart and simulation engine of this system. It is not an isolated computational tool, but a core module deeply embedded in the system process. Its main functions include: Before construction, a point cloud of potential risk states covering various possibilities is generated through large-scale, multi-condition simulations, which serves as the original data source for building the risk particle knowledge base. Accurately simulate how precipitation (flow field change) induces soil deformation (stress field change), and how excavation unloading (stress field change) in turn affects the seepage path (flow field change), realistically reproducing the core physical process of fluid-solid bidirectional coupling; Testing and optimizing different precipitation, reinjection, and support schemes in a virtual environment, evaluating their effectiveness and environmental impact, and providing a scientific basis for developing a risk particle-strategy mapping library.

[0034] The core physics engine of the model was identified as fluid-structure interaction, and a dynamic, step-by-step construction simulation method was defined. This enables the system to realistically reproduce the complex interactive effects of dynamic processes such as dewatering, excavation, and support on the foundation pit and the environment, ensuring that the generated high-dimensional system state vector contains spatiotemporal evolution information, which is a fundamental prerequisite for subsequent risk pattern learning.

[0035] The point cloud database of potential risk states generated through Monte Carlo simulation includes: The spatiotemporal four-dimensional data field is sliced ​​in one dimension according to the time axis to obtain the high-dimensional system state vector corresponding to each time slice. Each high-dimensional system state vector represents the state characteristics of the foundation pit in a certain scenario and at a certain time. At the same time, the state characteristics are marked with risk consequences, including safety, danger and warning. Obtain the probability distribution of the high-dimensional system state vector corresponding to each time slice, and perform cyclic random simulation of the state based on the probability distribution to generate a time series set; Each independent random simulation operation includes: running a complete simulation calculation based on a set of parameters randomly sampled according to a preset probability distribution model. The simulation calculation generates a time series, which is a sequence of multiple high-dimensional system state vectors recorded in chronological order during the simulation. The risk consequences are calculated for each time series set, and each time series is assigned a risk consequence label. The risk consequence calculation logic includes: When the maximum surface settlement is greater than 60 mm, or the axial force of the support is greater than 1.2 times the conventional design value of the axial force of the support, or the maximum displacement of the retaining structure is greater than 100 mm, the risk consequence of the current time series is labeled as dangerous. When the maximum surface settlement is greater than 40 mm and less than 60 mm, or the axial force of the support is greater than 1.05 times the conventional design value of the axial force of the support, or the maximum displacement of the retaining structure is greater than 70 mm and less than 100 mm, the risk consequence label assigned to the current time series is warning. When the maximum surface settlement is less than 40 mm, the axial force of the support is less than 1.05 times the conventional design value of the axial force of the support, and the maximum displacement of the retaining structure is less than 70 mm, the risk consequence of the current time series is labeled as safe. All high-dimensional system state vectors with risk consequences labels generated in the simulation are stored in the potential risk state point cloud database. The risk consequences are categorized into three levels: Danger is the highest level, Alert is the next highest level, and Safety is the lowest level.

[0036] This database is a valuable resource for training the risk particle model, containing complete evolutionary path information on how foundation pit engineering can go from safe to dangerous under various possible conditions.

[0037] This step is fundamental to the simulation and aims to identify all uncertain variables that may affect the safety and deformation of the foundation pit. These variables will be used as random samples in the Monte Carlo simulation.

[0038] This process is the cornerstone of the system and is completed before construction.

[0039] In this embodiment, a highly realistic three-dimensional numerical model of fluid-structure interaction is established using the professional geotechnical engineering software FLAC3D and COMSOL. Large-scale Monte Carlo simulations are conducted to simulate tens of thousands of possible construction scenarios. Each simulation generates a complete time-series data chain. The state vectors in all these data chains are aggregated to form an extremely large potential risk state point cloud database. The potential risk state point cloud database covers a massive amount of high-dimensional system state vectors from safe operating conditions to various extreme failure conditions.

[0040] The Monte Carlo simulation of disturbances specifically includes: determining the probability distribution of key input parameters based on geological survey reports and indoor test results. Key input parameters include the mechanical parameters and hydraulic parameters of the soil, as well as external boundary conditions. Random sampling combinations are performed within the probability distribution range of each parameter to generate tens of thousands of different parameter inputs. A complete fluid-structure interaction three-dimensional numerical model calculation is performed on each input to generate a potential risk state point cloud database covering multiple working conditions.

[0041] By randomly sampling and combining key uncertainty parameters, tens of thousands of possibilities covering everything from safety to various failure modes are systematically generated, and each state is assigned a clear risk consequence label. This provides a massive amount of high-quality, supervised learning material for subsequent machine learning algorithms, solving the problem that traditional methods rely on limited engineering cases and cannot exhaustively enumerate risks.

[0042] Clustering units include: An improved coarse-grained risk sphere generation algorithm is used to iteratively cluster the potential risk state point cloud database to obtain a risk sphere knowledge base. The risk sphere knowledge base consists of risk spheres with engineering semantic labels, risk consequences, center vectors, and radii. The improved coarse-grained risk particle generation algorithm includes splitting risk particles through a non-random center point selection strategy, iterating until the purity of all risk particles meets a preset purity threshold, and forming a risk particle knowledge base. The improved coarse-grained risk particle generation algorithm includes: Step S1: Take out a risk ball from the head of the risk ball column, count the number of sample points belonging to each risk consequence in the risk ball, and take the risk consequence corresponding to the highest number of sample points as the dominant risk consequence. The initial state of the risk particle array to be processed is the potential risk state point cloud database. Splitting the risk particle array is equivalent to classifying the potential risk state point cloud database, and the sample points are the state vectors of each high-dimensional system. Step S2: Calculate the ratio of the number of sample points of the dominant risk consequence to the total number of sample points in the risk sphere to obtain the purity of the risk sphere. When the purity of the risk sphere is greater than or equal to the preset purity threshold, the risk sphere is stored in the risk sphere knowledge base. When the purity of the risk sphere is less than the preset purity threshold, the risk sphere splitting operation is performed; Step S3: Remove the split risk particles from the list of risk particles to be processed, add all the new sub-risk particles generated after splitting to the end of the list of risk particles to be processed, return and repeat steps S1 and S2 until the list of risk particles to be processed is empty, and output the risk particle knowledge base. This step is equivalent to performing an efficient, single-step k-means split using two carefully selected, most representative seed points. This ensures that the boundaries of the two split sub-risk particles are located precisely in the clash zone where the two risk patterns are most easily confused, resulting in a precise and efficient split.

[0043] The purity threshold is a real number between 0.5 and 1. When the proportion of sample points belonging to the same risk consequence in a risk sphere reaches or exceeds this threshold, the risk sphere is considered pure and does not need to be split. The risk particles to be processed are arranged in a first-in-first-out queue. Initially, the queue contains only one risk particle, which contains all the data in the potential risk state point cloud database.

[0044] The risk particle knowledge base is a collection in which each element is a risk particle object, representing a specific risk pattern of the foundation pit system. The risk particle includes a risk particle ID, risk label, risk consequence, center vector, radius, and sample points. The risk particle ID is a unique identifier, the risk label is an intuitive engineering semantic description (such as initial uniform settlement), the center vector is the most typical state vector of the risk pattern represented by the risk particle, the radius is the fluctuation range of the risk pattern, and the sample points are all high-dimensional system state vectors belonging to the risk particle.

[0045] The core of the improved coarse-grained risk granule generation algorithm lies in selecting split centers through a deterministic strategy based on data distribution characteristics, thereby replacing the random initialization in traditional clustering algorithms and ensuring the stability and high quality of results in each risk granulation process.

[0046] The engineering semantic tags for risk particles include safe operation risk particles, initial uniform settlement risk particles, over-extraction-induced regional settlement risk particles, improper reinjection leading to local heave risk particles, confined water seepage risk particles, support system stress concentration risk particles, and piping burst precursor risk particles.

[0047] The core innovation of this invention lies in its application of an improved coarse-grained risk sphere generation algorithm to extract knowledge from this massive point cloud database. Traditional clustering algorithms (such as k-means) rely on random initial points, leading to unstable results. The improved coarse-grained risk sphere generation algorithm of this invention employs a deterministic center point selection strategy. In a risk sphere to be split, containing samples with multiple risk consequences, the algorithm calculates the geometric center of the outlier samples with the highest risk consequences, i.e., the outlier mean point, which represents the most dangerous trend within that risk sphere. Then, the algorithm calculates the distance from all sample points with lower risk consequences to this outlier mean point and finds the sample point at the median of the distance ranking. This median point, due to its robustness in data distribution, is selected as a new, more representative cluster center. Finally, based on this stable cluster center, fine-tuning is performed to complete the split.

[0048] Through repeated iterations, the clustering process continues until all samples within the generated risk spheres belong to the same risk consequence. These final stable, coarse-grained risk spheres are then assigned clear engineering semantic labels and risk consequences, collectively forming the system's initial risk sphere knowledge base.

[0049] The engineering semantic tags for risk particles are named according to typical risk patterns in foundation pit engineering, and have intuitive engineering significance. For example, the initial uniform settlement risk particle represents the normal state in the early stage of foundation pit excavation, where the settlement rate is similar and within a safe range. The over-drainage-induced regional settlement risk particle represents a risk pattern where excessive local dewatering leads to an abnormally accelerated settlement rate in a specific area. The confined water surge precursor risk particle is a high-risk pattern, whose feature vector may show a sharp drop in pore water pressure at a certain point at the bottom of the pit, accompanied by abnormal fluctuations in the water level of nearby observation wells. These semantic tags enable managers to instantly understand the complex state of the system.

[0050] The risk granulocyte splitting operation includes the following sub-steps: Step S21: Inside the risk sphere to be split, form a heterogeneous sample set by combining all sample points that do not belong to the dominant risk consequences, calculate the arithmetic mean of all sample points in the heterogeneous sample set, and obtain the virtual heterogeneous mean point. Step S22: Form a sample set of all sample points belonging to the dominant risk consequence into a sample set of the same kind, calculate the high-dimensional Euclidean distance from each sample point in the sample set to the virtual outlier mean point, obtain a list of high-dimensional Euclidean distances, sort the list of high-dimensional Euclidean distances in ascending order, obtain the median distance of the list of high-dimensional Euclidean distances, obtain the sample point in the sample set of the same kind that is closest to the median distance based on the list of high-dimensional Euclidean distances, and use the sample point that is closest to the median distance as the initial stable cluster center of the new risk sphere representing the dominant risk consequence that will be split off. Step S23: Based on the high-dimensional Euclidean distance list, obtain the sample points in the heterogeneous sample set that are closest to the initial stable cluster center, and obtain the initial auxiliary cluster center. Use the initial stable cluster center and the initial auxiliary cluster center as the split center. Create an empty list of sub-risk particles, which includes the list of first cluster core sub-risk particles and the list of first cluster core sub-risk particles; Traverse each sample point in the original risk sphere to be split, and calculate the high-dimensional Euclidean distance between each sample point and the initial stable cluster center and the initial auxiliary cluster center to obtain the first cluster center distance and the second cluster center distance. When the distance between the first cluster centers is less than the distance between the second cluster centers, the corresponding sample point is assigned to the list of risk particles of the first cluster center. When the distance between the first cluster centers is greater than or equal to the distance between the second cluster centers, the corresponding sample point is assigned to the list of risk particles for the second cluster centers.

[0051] The virtual outlier mean point represents the center of gravity or the most typical manifestation of all more dangerous states within the current mixed-state risk sphere. The virtual outlier mean point is a data-driven and stable reference point that provides a reliable bullseye for subsequent targeted splitting, completely avoiding the uncertainty brought about by random selection.

[0052] Choosing the median sample point as the cluster center provides strong robustness. It effectively resists the interference of a few extreme outliers in the same sample set, which are safe points that are either particularly far from or particularly close to the center of gravity of danger. This ensures that the selected new center best represents the general state of the risk consequences, rather than a special state.

[0053] By employing a deterministic, non-random centroid selection strategy (locating outlier means and selecting median representative points), the stability of the clustering process and the high quality of the results are ensured. It can efficiently and robustly translate massive, unordered state point clouds into a series of risk particles with clear engineering semantics and well-defined boundaries, realizing the extraction of structured knowledge from raw data and laying the foundation for subsequent pattern recognition and decision-making.

[0054] The decision-making module includes a risk pattern matching unit and a state trajectory prediction unit: The risk pattern matching unit is used to calculate the Euclidean distance between the current high-dimensional system state vector and the center vector of each risk particle in the risk particle knowledge base in real time in the high-dimensional feature space, and to determine the risk particle to which the current high-dimensional system state vector belongs and the corresponding risk consequences based on the nearest neighbor principle. The state trajectory prediction unit is used to continuously track the movement trajectory of a high-dimensional system state vector in the risk particle space based on a recurrent neural network algorithm. The movement trajectory includes the movement speed and direction, and predicts the probability that the high-dimensional system state vector will enter a higher-level risk consequence risk particle within a preset time period in the future.

[0055] The preset time period is set to 6 hours in the future; The core function of the state trajectory prediction unit is to elevate the evolution of the entire foundation pit system's state from static diagnosis to dynamic prediction based on a recurrent neural network algorithm capable of processing time-series data. This unit continuously captures and tracks the continuous movement trajectory of the high-dimensional system state vector, which describes the overall system's operational state, within an abstract risk sphere space. It extracts key dynamic features from this trajectory, including movement speed representing the severity of state changes and movement direction revealing the trend of state evolution. These trajectory features, rich in dynamic information, are input into a deep learning model pre-trained on massive amounts of simulated data. The deep learning model has learned and mastered the complex nonlinear mapping relationship between various historical trajectory features and future risk consequences. It calculates and outputs a probabilistic prediction result in real time, clearly indicating the probability that the current system state will enter various risk spheres representing higher-level dangerous consequences within a preset time period. This provides crucial decision-making basis for proactive, preventative interventions.

[0056] Trajectory features are extracted from the continuous high-dimensional system state vector time series. The trajectory features include the moving speed of the state vector and the moving direction pointing towards the center of the high-risk particle. The time series containing trajectory features is input into a pre-trained recurrent neural network model, and the model outputs the probability of the high-dimensional system state vector entering a higher-level risk consequence risk particle within a preset time period in the future.

[0057] In this embodiment, the risk pattern matching unit is the core of diagnosis. It determines the category of the condition by calculating the distance between the current state vector and the center points of all risk particles in the knowledge base. The state trajectory prediction unit is the core of prognosis. It not only looks at the current state but also focuses on the trend of state changes. By analyzing the path and speed of the state vector moving between different risk particles, it can predict the direction of risk evolution. For example, even if a state point is still within the warning risk particle, but its trajectory moves at high speed towards the dangerous risk particle, a higher level of warning can be issued in advance.

[0058] The risk pattern matching unit achieves a cognitive upgrade from single-point threshold judgment to multi-dimensional system pattern recognition through high-dimensional spatial distance calculation, resulting in more accurate decision-making and stronger anti-interference capabilities. The state trajectory prediction unit, by introducing a recurrent neural network, achieves a leap from static diagnosis to dynamic prediction, enabling the prediction of risk evolution trends and probabilities, thus providing a possibility for proactive immunization.

[0059] The control module includes a status trigger mode and a trajectory trigger mode; The state triggering mode includes: when the risk pattern matching unit determines that the current high-dimensional system state vector has entered the interior of a certain risk particle, the response strategy bound to the risk particle is immediately and automatically triggered and executed based on the risk particle-policy mapping library; The trajectory triggering mode includes: when the state trajectory prediction unit predicts that the current high-dimensional system state vector will enter the interior of a certain risk particle within a preset time, the response strategy bound to the risk particle is executed in advance based on the risk particle-strategy mapping library to achieve proactive immune intervention; The risk particle-policy mapping library includes: When the state vector of a high-dimensional system is determined to enter the subsidence risk sphere caused by over-pumping, a recharge compensation strategy is triggered. The recharge compensation strategy includes: automatically increasing the recharge flow rate of recharge wells near the subsidence over-limit area, and simultaneously reducing the pumping power of dewatering wells within a specified range. When the state vector of a high-dimensional system is determined to enter the stress concentration risk particle of the support system, the stress release and support strengthening strategy is triggered. The stress release and support strengthening strategy includes: sending a level 2 alarm to the project management personnel, suggesting local earthwork backfilling and adding temporary supports in the corresponding area of ​​the stress concentration risk particle of the support system, and increasing the collection frequency of relevant axial force monitoring points. When the state vector of the high-dimensional system is determined to enter the risk sphere of the confined water surge precursor, an emergency locking and alarm strategy is triggered. The alarm strategy includes: immediately and automatically shutting down all dewatering wells in the risk area, starting all backup reinjection wells at full power, sending the highest level alarm to all senior management personnel of the project through sound and light, SMS and application software, and automatically locking the access control at the pit entrance.

[0060] In this embodiment, a pre-established risk particle-strategy mapping library is used to predefine and bind one or more standardized, automatically executable response strategies and emergency plans for each non-safety level risk particle. The preset time is set between 4 and 8 hours; State-triggered mode is a type of responsive control. When the GPS positioning of the high-dimensional system's state vector indicates that it has entered a dangerous area (risk particle), the system immediately activates the standard emergency plan corresponding to that area.

[0061] The trajectory-triggered mode is a proactive and preventative control measure. When the state trajectory prediction unit predicts that the navigation route of the state vector will enter a dangerous area in the future (e.g., within 6 hours), the system will not wait until it enters the dangerous area to take action. Instead, it will proactively intervene, for example, by slightly increasing the backflow rate or slightly reducing the precipitation rate, attempting to change the trajectory of the state vector and deviate it from the dangerous direction. This preventative mechanism is a key guarantee for the safety of this system.

[0062] In this embodiment, when the state vector of the high-dimensional system is determined to have entered the subsidence risk sphere caused by over-pumping, it indicates that the precipitation in a certain area is too intense. A reinjection compensation strategy is executed, specifically including: increasing the flow rate of reinjection well No. 3 in area A by 20%, and simultaneously reducing the frequency of the inverters in precipitation wells No. 5 and No. 6 within a 50-meter radius by 15%. For stress concentration risk spheres in the support system, this may mean uneven stress on the support or impending overload. A level-two alarm is pushed to the APP of the project chief engineer and structural engineer, with clear suggestions. Simultaneously, the collection frequency of relevant monitoring points is automatically adjusted from once per hour to once every 5 minutes to closely monitor its development.

[0063] When the state vector of a high-dimensional system is determined to enter the pre-emergence risk sphere of confined water surge, this is the highest level of danger signal. An emergency lockdown and alarm strategy is immediately executed, requiring no manual confirmation. Power to all dewatering pumps in the risk area is automatically cut off, and all backup reinjection wells are activated at full power for backpressure sealing. Simultaneously, the highest-level alarm is issued to all pre-set senior management personnel through all available channels (audio-visual, SMS, APP, and telephone voice), and can even be linked to the on-site access control system to automatically lock the pit entrance and prevent personnel from entering. These are all pre-coded enforcement logics in a smart contract.

[0064] The state-triggered mode ensures a rapid and standardized response to risks that have occurred. More innovative is the trajectory-triggered mode, which, based on the prediction of movement trajectories, can intervene in advance with small doses, eliminating a large number of risks in their infancy, achieving proactive immunity, and fundamentally improving the safety of the system.

[0065] The real high-dimensional system state vector data generated throughout the construction process will be continuously added to the potential risk state point cloud database. The improved coarse-grained risk particle generation algorithm was re-run on the supplemented database to obtain the supplemented and corrected risk particle knowledge base.

[0066] The supplementary correction includes fine-tuning the center and radius of existing risk spheres, generating new risk spheres, and merging or deleting risk spheres that have been verified as invalid due to sample sparsity.

[0067] In this embodiment, all real monitoring data during construction is continuously fed into the system's online learning unit, labeled with risk outcomes. The model reconstruction unit periodically (e.g., weekly) reruns the coarse-grained risk particle generation algorithm on a larger dataset containing these new cases. This process is akin to an experienced doctor reviewing cases, resulting in the self-evolution of the knowledge base: existing disease definitions (risk particles) may be revised to be more accurate; new complications (new risk particles) that never appeared in the simulation may be discovered; and pseudo-diseases simulated but never occurred in practice will be eliminated.

[0068] The implementation steps of the integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata include: Step S1: Before construction, a potential risk state point cloud is generated through large-scale fluid-structure interaction numerical simulation, and an improved coarse-grained risk particle generation algorithm is applied to granulate the risk, construct an initial risk particle knowledge base containing multiple risk modes, and obtain the risk particle-policy mapping. Step S2: During construction, the sensing module collects monitoring data in real time and dynamically constructs a high-dimensional system state vector at the current moment; Step S3: Through the decision module, the current high-dimensional system state vector is pattern matched with the risk particle knowledge base to determine the risk particle and risk consequences to which it belongs, and to predict its evolution trajectory. Step S4: When it is determined or predicted that the system state enters a non-safety risk particle, the control module automatically triggers and executes the bound response strategy or active immune intervention. Step S5: Throughout the entire construction lifecycle, continuously collect real data and periodically iterate and optimize the risk particle knowledge base through model reconstruction units to achieve the self-evolution of the system.

[0069] This invention, in terms of precipitation management, moves beyond simply pumping water blindly to reach target levels. It identifies specific risk spheres caused by over- or under-pumping and automatically triggers the optimal pumping-irrigation combination strategy associated with that sphere. This achieves precise, dynamic, and environmentally friendly precipitation that meets the requirements of pit operations while minimizing environmental disturbance. In monitoring and early warning, it completely changes the traditional methods' problems of false alarms due to data disturbances and missed alarms due to slow perception of systemic risks. Through pattern matching in high-dimensional space, it can accurately identify complex risk precursors composed of multiple factors and data from multiple measurement points. More importantly, through state trajectory prediction, the system can not only react when a risk occurs (state triggering) but also predict the evolution trend of the risk and intervene in advance (trajectory triggering). This proactive immunity capability eliminates a large number of potential risks in their infancy, achieving a fundamental shift in safety management from post-event remediation to pre-event prevention. In terms of environmental control, the system uses the deformation of the surrounding environment (such as ground subsidence and building tilt) as a key component of the high-dimensional system state vector, and generates risk spheres directly related to environmental protection, such as local uplift caused by improper recharge. Once the system state shows a tendency to enter such spheres, the control module will proactively adjust the recharge strategy to ensure the integrated achievement of environmental protection and engineering safety goals. This greatly enhances the initiative and safety of the integrated management of dewatering, monitoring, and environmental protection in water-rich strata subway foundation pits.

[0070] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A three-in-one intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata, characterized in that: It includes a perception module, a risk particle knowledge base module, a decision-making module, and a control module: The sensing module is used to collect and fuse multi-source heterogeneous monitoring data of the foundation pit project in real time, and dynamically form a high-dimensional system state vector describing the overall operating status. The risk sphere knowledge base module is used to iteratively cluster the potential risk state point cloud database through an improved coarse-grained risk sphere generation algorithm to obtain the risk sphere knowledge base. The risk sphere knowledge base consists of risk spheres with engineering semantic labels, risk consequences, center vectors, and radii. The decision module is used to calculate in real time the distance between the high-dimensional system state vector and the center vector of each risk particle in the risk particle knowledge base, determine the risk particle to which the current high-dimensional system state vector belongs and the corresponding risk consequences, and predict the evolution trajectory of the high-dimensional system state vector in the risk particle space. The control module is used to automatically execute the control strategy bound to the risk pellet based on the decision result of the decision module, and to remotely regulate the precipitation and recharge equipment.

2. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 1, characterized in that, The sensing module includes an underground monitoring unit, a surface monitoring unit, a macroscopic monitoring unit, and a dynamic inspection unit: The underground monitoring unit is used to acquire soil layer data, geological parameters, retaining structure displacement, structural stress, pore water pressure and groundwater level through an automated hydrostatic level, pore water pressure gauge, inclinometer, support axial force gauge, intelligent water level gauge and flow meter. The soil layer data includes soil layer classification, soil layer position, soil layer thickness, soil layer distribution and interrelationship between soil layers. The soil layer classification includes miscellaneous fill, silty clay, sand and gravel layer and bedrock. The geological parameters include permeability coefficient and elastic modulus. The ground monitoring unit is used to acquire surface subsidence, building tilt angle, pipeline strain, and hydrological conditions through a high-precision GPS receiver, building tilt sensor, crack gauge, and pipeline strain sensor. The hydrological conditions include rainfall intensity and external water supply.

3. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 2, characterized in that, The risk particle knowledge base module includes a database unit and a clustering unit: The database unit includes: Based on the data collected by the sensing module, a three-dimensional numerical model of fluid-structure interaction is constructed using geotechnical engineering software. A high-dimensional system state vector output by the three-dimensional numerical model of fluid-structure interaction with Monte Carlo simulation of disturbance is used to generate a potential risk state point cloud database. The fluid-structure interaction three-dimensional numerical model includes a geometric model, multiphysics coupling control equations, a material constitutive model and parameters, and boundary and initial conditions: The three-dimensional spatial geometric model integrates the geological body and the engineering entity in a unified coordinate system by performing point-to-point coordinate correspondence. The three-dimensional spatial geometric model provides an accurate solution domain for physical field calculations. The geological body includes stratigraphic structure, hydrological boundaries, and topography; The engineering entity includes the foundation pit and support structure, the three-dimensional model structure of dewatering and recharge, and the BIM model of the surrounding environment; The multiphysics field coupling control equations include the solid mechanical field and the porous medium seepage field. The control equation of the solid mechanical field is the static equilibrium equation of the soil and structure solved based on the elastoplastic mechanics theory. The control equation of the porous medium seepage field is the motion equation of groundwater in the porous medium solved based on the non-Darcy seepage model for high-speed flow. The constitutive model and parameters of the materials include soil materials, structural materials and interface elements. The constitutive model of the soil materials is a hardened soil model. The parameters of the soil materials include mechanical parameters and hydraulic parameters. The mechanical parameters include effective cohesion, effective internal friction angle, elastic modulus and Poisson's ratio. The hydraulic parameters include saturated permeability coefficient, porosity and specific yield. The constitutive model of structural materials is a linear elastic model, and the parameters of structural materials include elastic modulus, Poisson's ratio, and density. The constitutive model of the interface element is used to simulate the relative slip and friction effects at the interface between the soil material and the structural material. The parameters of the interface element include the interface friction angle, cohesion, normal and tangential stiffness. Boundary conditions and initial conditions include: Boundary conditions include mechanical boundaries and hydraulic boundaries. Mechanical boundaries include: Fixed constraints are zero displacement in all directions. The vertical boundaries around the fluid-structure interaction 3D numerical model are set as normal constraints. Normal constraints allow vertical displacement and restrict horizontal displacement. The top surface of the fluid-structure interaction 3D numerical model is a free boundary. The hydraulic boundary includes: The bottom and far-field vertical boundaries of the model are set as impermeable boundaries, the top of the aquifer at the far-field boundary is set as a constant head boundary, and the positions of the dewatering wells and recharge wells are defined as flow boundaries and time-varying head boundaries. Initial conditions include geostress equilibrium conditions and initial water field conditions. Geostress equilibrium conditions include geostress equilibrium calculations performed before any excavation and dewatering. Initial water field conditions include the groundwater level and pore water pressure distribution under natural conditions.

4. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 3, characterized in that, The fluid-structure interaction three-dimensional numerical model closely links the solid mechanical field and the seepage field based on the effective stress principle, forming a two-way influence: The two-way influence includes the effect of seepage on deformation and the effect of deformation on seepage. The effect of seepage on deformation includes: Rainfall causes a decrease in pore water pressure. According to Terzaghi's effective stress principle, the effective stress of the soil skeleton increases accordingly. The increase in effective stress leads to soil compression, resulting in consolidation settlement and deformation of the retaining structure. The effects of deformation on seepage include: Deformation caused by excavation unloading and soil consolidation alters the void ratio of the soil, which in turn affects the permeability coefficient. The change in the permeability coefficient, in turn, changes the distribution of the seepage field. The fluid-structure interaction three-dimensional numerical model transforms the static geometric model into a dynamic construction process simulation through step-by-step construction. The transformation process includes: Calculate the initial ground stress and water pressure field, identify the retaining structure in the fluid-structure interaction three-dimensional numerical model, simulate the dewatering process by manually setting the flow rate parameters of the dewatering wells, simulate excavation after the hydraulic field reaches a new equilibrium or reaches a specified time, activate the support of the soil layer corresponding to the specified depth after excavation to a specified depth, and apply prestress. Repeat the above steps until the foundation pit is excavated to the bottom of the pit. Output a spatiotemporal four-dimensional data field set, which includes stress cloud map, displacement vector or settlement cloud map, pore water pressure isosurface vector map, seepage velocity vector map and structural internal force distribution map. The structural internal forces include bending moment, axial force and shear force. The spatiotemporal four-dimensional data field set is subjected to Monte Carlo simulation perturbation to obtain the high-dimensional system state vector.

5. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 4, characterized in that, The point cloud database of potential risk states generated through Monte Carlo simulation includes: The spatiotemporal four-dimensional data field set is sliced ​​in one dimension according to the time axis to obtain the high-dimensional system state vector corresponding to each time slice. Each high-dimensional system state vector represents the state characteristics of the foundation pit in a certain scenario and at a certain moment. At the same time, the state characteristics are marked with risk consequences, including safety, danger and warning. Obtain the probability distribution of the high-dimensional system state vector corresponding to each time slice, and perform cyclic random simulation of the state based on the probability distribution to generate a time series set; The risk consequences are calculated for each time series set, and each time series is assigned a risk consequence label. The risk consequence calculation logic includes: When the maximum surface settlement is greater than 60 mm, or the axial force of the support is greater than 1.2 times the conventional design value of the axial force of the support, or the maximum displacement of the retaining structure is greater than 100 mm, the risk consequence of the current time series is labeled as dangerous. When the maximum surface settlement is greater than 40 mm and less than 60 mm, or the axial force of the support is greater than 1.05 times the conventional design value of the axial force of the support, or the maximum displacement of the retaining structure is greater than 70 mm and less than 100 mm, the risk consequence label assigned to the current time series is warning. When the maximum surface settlement is less than 40 mm, the axial force of the support is less than 1.05 times the conventional design value of the axial force of the support, and the maximum displacement of the retaining structure is less than 70 mm, the risk consequence of the current time series is labeled as safe. All high-dimensional system state vectors with risk consequences labels generated in the simulation are stored in the potential risk state point cloud database. The risk consequences are categorized into three levels: Danger is the highest level, Alert is the next highest level, and Safety is the lowest level.

6. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 5, characterized in that, The clustering unit includes: An improved coarse-grained risk sphere generation algorithm is used to iteratively cluster the potential risk state point cloud database to obtain a risk sphere knowledge base. The risk sphere knowledge base consists of risk spheres with engineering semantic labels, risk consequences, center vectors, and radii. The improved coarse-grained risk particle generation algorithm includes: Step S1: Take out a risk particle from the head of the risk particle column to be processed, count the number of sample points belonging to each risk consequence in the risk particle, and obtain the risk consequence corresponding to the highest number of sample points as the dominant risk consequence. The initial state of the risk particle array to be processed is the potential risk state point cloud database. Splitting the risk particle array is equivalent to classifying the potential risk state point cloud database, and the sample points are the state vectors of each high-dimensional system. Step S2: Calculate the ratio of the number of sample points of the dominant risk consequence to the total number of sample points in the risk sphere to obtain the purity of the risk sphere. When the purity of the risk sphere is greater than or equal to a preset purity threshold, the risk sphere is stored in the risk sphere knowledge base. When the purity of the risk sphere is less than a preset purity threshold, a risk sphere splitting operation is performed. Step S3: Remove the split risk particles from the list of risk particles to be processed, add all the new sub-risk particles generated after splitting to the end of the list of risk particles to be processed, return and repeat steps S1 and S2 until the list of risk particles to be processed is empty, and output the risk particle knowledge base. The engineering semantic tags for risk particles include safe operation risk particles, initial uniform settlement risk particles, over-extraction-induced regional settlement risk particles, improper reinjection leading to local heave risk particles, confined water seepage risk particles, support system stress concentration risk particles, and piping burst precursor risk particles.

7. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 6, characterized in that, The risk granulocyte splitting operation includes the following sub-steps: Step S21: Inside the risk sphere to be split, form a heterogeneous sample set by combining all sample points that do not belong to the dominant risk consequences, calculate the arithmetic mean of all sample points in the heterogeneous sample set, and obtain the virtual heterogeneous mean point. Step S22: Form a sample set of all sample points belonging to the dominant risk consequences, calculate the high-dimensional Euclidean distance from each sample point in the sample set to the virtual outlier mean point, obtain a list of high-dimensional Euclidean distances, sort the list of high-dimensional Euclidean distances in ascending order, obtain the median distance of the list of high-dimensional Euclidean distances, obtain the sample point in the sample set that is closest to the median distance based on the list of high-dimensional Euclidean distances, and use the sample point that is closest to the median distance as the initial stable cluster center of the new risk sphere representing the dominant risk consequences that will be split off. Step S23: Based on the high-dimensional Euclidean distance list, obtain the sample points in the heterogeneous sample set that are closest to the initial stable cluster center, and obtain the initial auxiliary cluster center. Use the initial stable cluster center and the initial auxiliary cluster center as the split center. Create an empty list of sub-risk particles, the list of sub-risk particles including a list of first cluster core sub-risk particles and a list of first cluster core sub-risk particles; Traverse each sample point in the original risk sphere to be split, and calculate the high-dimensional Euclidean distance between each sample point and the initial stable cluster center and the initial auxiliary cluster center to obtain the first cluster center distance and the second cluster center distance. When the distance between the first cluster centers is less than the distance between the second cluster centers, the corresponding sample point is assigned to the list of risk particles of the first cluster center. When the distance between the first cluster centers is greater than or equal to the distance between the second cluster centers, the corresponding sample point is assigned to the list of risk particles for the second cluster centers.

8. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 7, characterized in that, The decision-making module includes a risk pattern matching unit and a state trajectory prediction unit: The risk pattern matching unit is used to calculate the Euclidean distance between the current high-dimensional system state vector and the center vector of each risk particle in the risk particle knowledge base in real time in the high-dimensional feature space, and to determine the risk particle to which the current high-dimensional system state vector belongs and the corresponding risk consequences based on the nearest neighbor principle. The state trajectory prediction unit is used to continuously track the movement trajectory of a high-dimensional system state vector in the risk particle space based on a recurrent neural network algorithm. The movement trajectory includes the movement speed and direction, and predicts the probability that the high-dimensional system state vector will enter a higher-level risk consequence risk particle within a preset time period in the future.

9. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 8, characterized in that, The control module includes a status trigger mode and a trajectory trigger mode: The state triggering mode includes: when the risk pattern matching unit determines that the current high-dimensional system state vector has entered the interior of a certain risk particle, the response strategy bound to the risk particle is immediately and automatically triggered and executed based on the risk particle-strategy mapping library; The trajectory triggering mode includes: when the state trajectory prediction unit predicts that the current high-dimensional system state vector will enter the interior of a certain risk particle within a preset time, the response strategy bound to the risk particle will be executed in advance based on the risk particle-strategy mapping library; The risk particle-policy mapping library includes: When the state vector of the high-dimensional system is determined to enter the subsidence risk particle of the area caused by over-pumping, the recharge compensation strategy is triggered. The recharge compensation strategy includes: automatically increasing the recharge flow of the recharge wells near the subsidence over-limit area, and at the same time reducing the pumping power of the dewatering wells within the specified range. When the state vector of the high-dimensional system is determined to enter the stress concentration risk particle of the support system, the stress release and support strengthening strategy is triggered. The stress release and support strengthening strategy includes: sending a level 2 alarm to the project management personnel, suggesting local earthwork backfilling and adding temporary supports in the corresponding area of ​​the stress concentration risk particle of the support system, and increasing the collection frequency of relevant axial force monitoring points. When the state vector of the high-dimensional system is determined to enter the risk sphere of the confined water surge precursor, an emergency locking and alarm strategy is triggered. The alarm strategy includes: immediately and automatically shutting down all dewatering wells in the risk area, starting all backup reinjection wells at full power, sending the highest level alarm to all senior management personnel of the project through sound and light, SMS and application software, and automatically locking the access control at the pit entrance.

10. The integrated intelligent management system for monitoring and protecting the environment of subway foundation pits in water-rich strata as described in claim 9, characterized in that: The real high-dimensional system state vector data generated throughout the construction process will be continuously added to the potential risk state point cloud database. The improved coarse-grained risk particle generation algorithm was re-run on the supplemented database to obtain the supplemented and corrected risk particle knowledge base.

Images