Deep foundation pit supporting structure dynamic early warning method and system based on deformation monitoring

Abstract

The present application relates to the technical field of foundation pit support structure early warning, in particular to a deep foundation pit support structure dynamic early warning method and system based on deformation monitoring, which deploys a multi-modal sensor array at key support points, activates local intensive monitoring through a geological mutation feature trigger, constructs a space-time feature matrix input into a ConvLSTM model optimized by transfer learning, accurately identifies three types of risks, namely karst cave, soft interlayer and pressure water layer gushing, and locates the instability coordinates; based on a hierarchical response mechanism, matches grouting, dewatering support or structure reinforcement schemes, and dynamically optimizes construction parameters combined with closed-loop feedback. The present application solves the problems of lagging response and rough risk identification of traditional monitoring, realizes early warning, quick disposal and accurate reinforcement of deep foundation pit risks, and significantly improves the construction safety under complex geology.

Description

Technical Field

[0001] This invention relates to the field of early warning technology for foundation pit support structures, specifically a dynamic early warning method and system for deep foundation pit support structures based on deformation monitoring. Background Technology

[0002] Deep foundation pit engineering is a crucial aspect of urban construction, and the stability of its support structure directly affects project safety and the surrounding environment. Under complex geological conditions, such as the presence of abrupt geological changes like karst caves, weak interlayers, or confined aquifers, the mechanical properties and hydrological conditions of the soil and rock mass may change significantly within a short period, posing a threat to the support structure.

[0003] Currently, the deformation monitoring of deep foundation pit support structures mainly adopts the following technical means: deploying sensors such as inclinometers and settlement observation points on the support structure to collect deformation data at regular intervals or low frequencies; in terms of data analysis, the threshold alarm method is mostly used, that is, when the monitored value exceeds the preset static threshold, an alarm is triggered; when identifying geological anomalies, judgments are usually made based on previous geological survey reports and engineering experience; after a dangerous situation occurs, engineers generally hold consultations based on monitoring information, then formulate reinforcement plans and organize construction.

[0004] The aforementioned existing technical solutions have the following shortcomings in practical engineering applications:

[0005] First, in terms of response timeliness, the use of timed low-frequency sampling and static threshold judgment makes it difficult to effectively capture micro-deformation and micro-fracture signals that are precursors to geological abrupt changes, which may lead to alarms lagging behind the development of the danger.

[0006] Second, in terms of risk identification, due to the relatively limited types of available monitoring data and the lack of comprehensive analysis methods for multi-source monitoring data, it is difficult to accurately distinguish different types of geological anomalies (such as karst caves, weak interlayers, and sudden surges of confined aquifers), and it is also difficult to accurately determine the spatial impact range of anomalies.

[0007] Third, in terms of the connection between decision-making and execution, the formulation of reinforcement plans and the determination of parameters after the occurrence of risks mainly rely on human experience. The time from anomaly identification to the implementation of measures is relatively long, and once the reinforcement parameters are determined, it is difficult to dynamically adjust them based on real-time monitoring data. The closed-loop verification of the reinforcement effect is somewhat difficult. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the purpose of this invention is to provide a dynamic early warning method and system for deep foundation pit support structures based on deformation monitoring, which can achieve early identification of geological abrupt changes, real-time risk quantification, graded automatic response, and dynamic closed-loop adjustment.

[0009] The technical solution adopted by this invention to solve its technical problem is:

[0010] A dynamic early warning method for deep foundation pit support structures based on deformation monitoring includes the following steps:

[0011] S1: Deploy a multimodal sensor array at key points of the support structure. The multimodal sensor array includes tilt sensors, strain gauges, microseismic monitors and pore water pressure gauges to simultaneously collect the deformation of the support structure, changes in internal forces, micro-fracture signals of the soil and rock mass and groundwater pressure data.

[0012] S2: Establish a geological abrupt change feature trigger. By comparing the gradient difference of adjacent sensor data in real time, when the difference in deformation between any two sensors in the same horizontal plane exceeds the threshold and the increase in microseismic energy within the preset time window exceeds the preset percentage threshold, the local enhanced monitoring mode is immediately activated, the sampling frequency of the affected area is increased to the preset sampling frequency and three-dimensional laser scanning is started.

[0013] S3: Construct a spatiotemporal feature matrix and input the spatiotemporal feature matrix into a geological mutation prediction model optimized by transfer learning. The geological mutation prediction model is constructed based on a convolutional long short-term memory network and is pre-trained through a training set containing three types of geological mutation samples: karst caves, weak interlayers, and confined aquifers. It outputs the probability of geological anomaly types and the coordinates of potential unstable areas in real time.

[0014] S4: Activate the graded response mechanism based on the geological anomaly type output by the model: When a karst cave is identified, automatically trigger grouting reinforcement decision and generate construction parameters including grouting hole coordinates, grout diffusion radius and pressure threshold; when a confined aquifer surge is identified, initiate dewatering-support joint control and generate dewatering well opening and closing sequence and support axial force compensation value; when a weak interlayer is identified, automatically trigger structural strengthening decision and generate adjusted support structure spacing and axial force distribution parameters.

[0015] As a preferred embodiment, a further technical solution of the present invention is:

[0016] Preferably, the key points of the support structure in step S1 include the supports, joints, connection nodes, main stress-bearing parts, deep foundation pit walls, retaining walls, weak layer areas, the bottom of the foundation pit, and pore water pressure monitoring points around the retaining structure.

[0017] Tilt sensors are deployed at supports, joints, and connection nodes of the support structure to monitor changes in the tilt of the support structure.

[0018] Strain gauges are deployed at the main stress-bearing parts of the supporting structure, including the location of reinforcing bars, beam-column connection nodes, support feet, and bearing surfaces, to monitor changes in internal forces in real time.

[0019] Microseismic monitors are deployed in deep foundation pit walls, retaining walls, and weak layer areas to detect microcracks and rupture signals in the soil and rock mass;

[0020] Pore ​​water pressure gauges are deployed at pore water pressure monitoring points at the bottom of the foundation pit or around the retaining structure to monitor the dynamic changes in groundwater pressure.

[0021] All sensors are connected to the data acquisition system via wireless transmission modules to achieve real-time synchronous data acquisition and transmission.

[0022] Preferably, the construction process of the spatiotemporal feature matrix in S3 includes,

[0023] The multimodal data collected by enhanced monitoring are filtered, denoised, and normalized.

[0024] Extract the deformation corresponding to time t and spatial position s from the processed data. Stress changes Microseismic energy changes and pore water pressure changes Integrated into a four-dimensional spatiotemporal feature matrix ;

[0025] in, The row dimension corresponds to the time series, the column dimension corresponds to the spatial point sequence, and the matrix elements correspond to the quantized values ​​of the multimodal monitoring parameters.

[0026] Preferably, the geological mutation prediction model in S3 is as follows:

[0027] A two-branch fusion model is constructed using a Convolutional Long Short-Term Memory (ConvLSTM) network. The first branch is a Convolutional Neural Network (CNN) module, which is used to extract the local spatial correlation features of deformation, stress change, microseismic energy change, and pore water pressure change in the spatiotemporal feature matrix. The second branch is a Long Short-Term Memory (LSTM) network module, which is used to model the time series evolution of the spatiotemporal feature matrix.

[0028] The spatial feature map output by the CNN module and the temporal feature vector output by the LSTM module are concatenated and fused, and then input into the fully connected layer to output the occurrence probability of three types of geological anomalies: karst caves, weak interlayers, and confined aquifers.

[0029] Preferably, the training process for the geological mutation prediction model includes,

[0030] S301: Initialize model parameters, loading the pre-trained model weights trained using historical monitoring data from similar geological regions as the initial convolutional kernel weights. and bias terms ;

[0031] S302: Labels with actual geological anomalies The historical spatiotemporal feature matrix is ​​used as input, and the feature mapping function is applied. Perform forward propagation and output the predicted probability. ;

[0032] S303: Employs the cross-entropy loss function Calculate the error between the predicted value and the actual value, where The total number of training samples;

[0033] S304: Iteratively update the convolutional kernel weights using the backpropagation algorithm. and bias terms until the loss function The model training is complete when the convergence reaches below the preset threshold.

[0034] Preferably, the coordinates of the potential instability area in S3 are determined by the following method, based on the probability of occurrence of three types of geological anomalies output by the trained geological mutation prediction model. Combined with pre-defined spatial influence weighting factors for various anomalies Through formula Calculate the probability of instability at each spatial location. ,Will The spatial coordinates corresponding to the maximum value are marked as the coordinates of the potential instability region.

[0035] Preferably, the specific implementation process of the hierarchical response mechanism in step S4 includes,

[0036] The karst cave response mechanism involves retrieving geological survey data from the potential instability zone coordinates when the model's output probability of karst cave anomalies exceeds a preset probability threshold. Based on the numerical range of the formation permeability coefficient k, the system selects the grouting material. ≥10 -4 cm / s, cement grout should be selected. <10 -4 cm / s, chemical slurry is selected; and the slurry diffusion radius is determined based on real-time monitoring. With the target diffusion radius The difference Dynamically adjust the grouting flow rate until... The absolute value is less than the preset tolerance;

[0037] The confined aquifer inrush response sub-mechanism generates a dewatering well opening / closing sequence and support axial force compensation value when the model outputs an inrush anomaly probability greater than a preset probability threshold; based on the real-time monitored water inrush flow rate... Set precipitation speed Satisfying the relation ,in, A safety factor greater than 1; based on the water level fluctuation range. Set water level monitoring frequency ,when When >0.5m, Set to 1 time / minute, when When ≤0.5m, Set to 1 time / 10 minutes;

[0038] The weak interlayer response sub-mechanism generates an adjusted support structure spacing when the model outputs an anomaly probability of the weak interlayer that exceeds a preset probability threshold. and axial force distribution parameters Set the support spacing adjustment amount With deformation Satisfying the relation ,in, As a positive coefficient, set the axial force compensation value. With change of internal force Satisfying the relation ,in, It is a positive coefficient to counteract the shear slip tendency of the weak interlayer.

[0039] Preferably, it also includes S5: according to the triggering conditions of the graded response mechanism, implement corresponding reinforcement and emergency measures. During the implementation of reinforcement measures, the multimodal sensor array continuously collects microseismic signals, deformation and pore water pressure data in real time and feeds them back to the control system. The control system dynamically verifies the reinforcement effect and iteratively optimizes the construction parameters based on the feedback data until the state of the geologically abnormal area meets the stability judgment conditions.

[0040] Preferably, the specific process of dynamically verifying the reinforcement effect and iteratively optimizing construction parameters based on feedback data includes:

[0041] S501: Real-time acquisition of microseismic energy from feedback data Slurry diffusion radius Precipitation flow Rate of water level change Deformation and changes in internal forces ;

[0042] S502: Match the corresponding dynamic adjustment algorithm based on the currently triggered geological anomaly type.

[0043] If the current scenario is grouting in a karst cave, then follow the formula. Dynamically adjust grouting pressure ,in, These are constants related to the mechanical properties of soil and rock masses; It is the attenuation factor of slurry flow resistance;

[0044] If the current situation is a sudden surge of confined aquifer, then follow the formula. Dynamically adjust the support axial force ,in, For the stiffness-related constants of the support structure, The nonlinear factor affecting hydraulic disturbance;

[0045] If the current scenario is a weak mezzanine, then follow the formula. Dynamically adjust the support spacing and according to the formula Dynamically adjust the support axial force, where, , These are constants related to the structural geometric properties. Sensitivity factor for changes in internal force;

[0046] S503: The adjusted construction parameters are sent to the execution mechanism, and the micro-seismic energy and deformation change trends after parameter adjustment are monitored in real time. When the micro-seismic energy exceeds the preset safety limit or the deformation exceeds the preset warning value, the system automatically suspends the current reinforcement operation and issues an alarm. Construction resumes after the monitoring data returns to the safe range, until all monitoring data are within the stable threshold range for a continuous preset time.

[0047] A dynamic early warning system for deep foundation pit support structures based on deformation monitoring is provided to execute a dynamic early warning method for deep foundation pit support structures based on deformation monitoring, comprising:

[0048] The multimodal sensor array includes tilt sensors, strain gauges, microseismic monitors, and pore water pressure gauges. Tilt sensors are deployed at supports, joints, and connection nodes of the supporting structure to collect tilt deformation data. Strain gauges are deployed at key load-bearing locations of the supporting structure, including reinforcement locations, beam-column connection nodes, support feet, and bearing surfaces, to collect internal force changes. Microseismic monitors are deployed on the deep foundation pit walls, retaining walls, and weak layer areas to collect micro-fracture signals in the soil and rock. Pore water pressure gauges are deployed at pore water pressure monitoring points at the bottom of the foundation pit or around the retaining structure to collect groundwater pressure data.

[0049] The data acquisition and transmission module connects to each sensor in the multimodal sensor array and realizes real-time synchronous acquisition and transmission of multimodal data through wireless transmission.

[0050] The geological mutation feature triggering module is used to compare the gradient difference of adjacent sensor data in real time. When the difference in deformation between any two sensors in the same horizontal plane exceeds the threshold and the increase in microseismic energy within the preset time window exceeds the preset percentage threshold, the local enhanced monitoring mode is activated, the sampling frequency of the affected area is increased to the preset sampling frequency and three-dimensional laser scanning is started.

[0051] The spatiotemporal coupling analysis module is used to fuse multimodal data collected by various sensors into a spatiotemporal feature matrix, which is then input into a geological mutation prediction model optimized by transfer learning. The geological mutation prediction model is constructed based on a convolutional long short-term memory network and is pre-trained using a training set containing three types of geological mutation samples: karst caves, weak interlayers, and confined aquifers. It outputs the probability of geological anomaly types and the coordinates of potential unstable areas in real time.

[0052] The graded response decision module is used to initiate a graded response mechanism based on the geological anomaly type output by the geological mutation prediction model: when a karst cave is identified, grouting reinforcement decision is automatically triggered, generating construction parameters including grouting hole coordinates, grout diffusion radius, and pressure threshold; when a confined aquifer surge is identified, dewatering-support joint control is initiated, generating dewatering well opening and closing sequence and support axial force compensation value; when a weak interlayer is identified, structural strengthening decision is automatically triggered, generating adjusted support structure spacing and axial force distribution parameters.

[0053] The dynamic feedback optimization module is used to dynamically verify the reinforcement effect and iteratively optimize the construction parameters based on the real-time monitoring data fed back by the multimodal sensor array during the implementation of reinforcement measures, until the state of the geologically abnormal area meets the stability judgment conditions.

[0054] The present invention, which adopts the above technical solution, has the following prominent features compared with the prior art:

[0055] This invention effectively improves the timeliness of response in foundation pit engineering under complex geological conditions by synchronously acquiring data from a multimodal sensor array and responding instantly to geological abrupt change feature triggers. When deformation gradient anomalies and microseismic signals surge, the system automatically activates an enhanced monitoring mode and uses three-dimensional laser scanning for spatial reinforcement, significantly improving the timeliness and sensitivity of early identification of geological abrupt changes and providing a more ample time window for subsequent analysis and decision-making. Based on a spatiotemporal coupling analysis model and a geological abrupt change prediction model optimized by transfer learning, this invention enables the identification and location of typical geological anomalies such as karst caves, weak interlayers, and sudden surges in confined aquifers. This model integrates multidimensional spatiotemporal data such as deformation, internal forces, microseismic activity, and water pressure, and outputs the probability of anomaly types and the coordinates of potential instability areas in real time, transforming traditional experience-based judgment into data-driven intelligent prediction, thereby improving the scientific rigor and accuracy of risk assessment.

[0056] Meanwhile, this invention tightly couples the graded response mechanism with the dynamic adjustment of reinforcement measures, achieving a closed-loop linkage from intelligent diagnosis to execution. Based on the geological anomaly type and coordinates output by the spatiotemporal coupling model, the system automatically initiates matching grouting, dewatering, or support structure adjustment plans. During this process, it dynamically calculates and optimizes key construction parameters such as grouting pressure and support axial force based on real-time feedback of multimodal monitoring data, including microseismic energy, water level changes, and structural deformation. This not only achieves targeted response but also ensures that reinforcement measures remain compatible with dynamically changing geological and structural conditions through adaptive parameter adjustment, thereby improving the timeliness, adaptability, and controllability of support system reinforcement under complex geological abrupt changes. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the arrangement of the multimodal sensor array in the deep foundation pit support structure according to an embodiment of the present invention;

[0058] Figure 2 This is a flowchart of the geological catastrophe prediction model based on ConvLSTM and transfer learning in an embodiment of the present invention.

[0059] Figure 3 This is a flowchart of the hierarchical response decision-making process in an embodiment of the present invention;

[0060] Figure 4 This is a flowchart of the closed-loop feedback control in an embodiment of the present invention. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. These embodiments are only used to better understand the technical content of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0062] like Figure 1 The diagram shown illustrates the arrangement of a multimodal sensor array in a deep foundation pit support structure according to an embodiment of the present invention. This embodiment provides a dynamic early warning method for deep foundation pit support structures based on deformation monitoring, comprising the following steps:

[0063] S1: Deployment and Data Acquisition of Multimodal Sensor Arrays; Multimodal sensor arrays are deployed at key points of the support structure. These arrays include tilt sensors, strain gauges, microseismic monitors, and pore water pressure gauges, simultaneously acquiring data on deformation, internal force changes, micro-fracture signals in the soil and rock mass, and groundwater pressure. Key points of the support structure include supports, joints, connection nodes, main stress-bearing areas, deep pit walls, retaining walls, weak layer areas, the bottom of the pit, and pore water pressure monitoring points around the retaining structure.

[0064] The multimodal sensor array is deployed as follows: tilt sensors are deployed at the supports, joints, and connection nodes of the supporting structure to monitor changes in the tilt of the supporting structure. Strain gauges are deployed at key load-bearing locations in the supporting structure, including reinforcement locations, beam-column connection nodes, support feet, and bearing surfaces, to monitor changes in internal forces in real time. Microseismic monitors are deployed in deep foundation pit walls, retaining walls, and weak layers to detect microcracks and rupture signals in the soil and rock mass. Pore ​​water pressure gauges are deployed at pore water pressure monitoring points at the bottom of the foundation pit or around the retaining structure to collect groundwater pressure data. All sensors are connected to the data acquisition system via wireless transmission modules to achieve real-time synchronous data acquisition and transmission.

[0065] S2: Geological abrupt change feature triggering and enhanced monitoring; establish a geological abrupt change feature trigger, and the system compares the data gradient difference of adjacent sensors in the same horizontal plane in real time. When the difference in deformation between any two sensors in the same horizontal plane exceeds the preset threshold L1 (e.g., L1 is set to 2mm), and the microseismic energy increases by more than 50% within a 10-second time window, the system immediately determines it as a precursor to a geological abrupt change and activates the local enhanced monitoring mode.

[0066] At this point, the system increases the sampling frequency of the affected area from the usual 1 time / minute to 10Hz (10 times per second) and simultaneously starts the 3D laser scanning equipment. Figure 4 As shown, 3D laser scanning is used to perform high-precision spatial deformation modeling of the affected area, obtain full-field displacement cloud maps of the support structure and surrounding rock surface, and supplement the spatial continuous deformation information of sensor points. The point cloud data obtained by scanning is registered, filtered, and calculated for deformation, and then used to verify and correct the spatial deformation distribution between adjacent sensors, and serves as a supplementary basis for spatial correlation analysis in the spatiotemporal feature matrix. However, it is not directly used as the input feature of the geological mutation prediction model to avoid computational redundancy caused by heterogeneous fusion of point cloud data and sensor data. In this embodiment, the conventional sampling frequency is set to 1 time / minute, and the sampling frequency is increased to 10Hz under the local enhanced monitoring mode. The ratio of the conventional frequency to the high-frequency frequency is not less than 50 times to ensure that transient signal changes during the geological mutation process can be captured with a sufficiently high temporal resolution after the enhanced monitoring is triggered. In actual engineering, those skilled in the art can adjust the specific values ​​of the conventional frequency and the high-frequency frequency according to factors such as the size of the foundation pit, sensor type, data storage and transmission capabilities. The principle is that the high-frequency sampling frequency should be at least one order of magnitude higher than the conventional frequency.

[0067] S3: Spatiotemporal Coupling Analysis and Geological Aberration Prediction; A spatiotemporal coupling analysis model is used to process enhanced monitoring data. First, the collected multimodal data is filtered, denoised, and normalized, and then a spatiotemporal feature matrix is ​​constructed.

[0068] like Figure 2 The diagram shown illustrates the processing flow of the geological catastrophe prediction model based on ConvLSTM and transfer learning in this embodiment of the invention. The system integrates the processed data into a four-dimensional spatiotemporal feature matrix. ;in, The row dimension corresponds to the time series, and the column dimension corresponds to the spatial point sequence; the matrix elements correspond to the quantized values ​​of the multimodal monitoring parameters.

[0069] The matrix is ​​input into a geological catastrophe prediction model optimized through transfer learning. This model employs a two-branch fusion structure built with ConvLSTM: the first branch is a CNN module that extracts local spatial correlation features from the data; the second branch is an LSTM module that models the time-series evolution patterns. The two feature sets are concatenated and then input into a fully connected layer.

[0070] The training sets for the three types of geological anomalous sample sets were derived from historical monitoring data and geological exploration verification data from multiple deep foundation pit projects. Specifically, karst cave samples were collected from cave exposure data verified by drilling or geophysical exploration in foundation pit projects in karst-developed areas; weak interlayer samples were collected from excavation verification data of foundation pits containing clay interlayers or fractured zones; and confined aquifer surge samples were collected from surge precursor data verified by on-site emergency response records in foundation pit projects experiencing surge hazards. Geological anomaly labels in each sample were marked by geological engineers based on verification results from core drilling, borehole television, or pressure water tests, with no fewer than 200 sets of samples for each type of anomaly. The training process for the geological anomalous prediction model includes...

[0071] S301: Initialize model parameters, loading the pre-trained model weights trained using historical monitoring data from similar geological regions as the initial convolutional kernel weights. and bias terms .

[0072] S302: Labels with actual geological anomalies The historical spatiotemporal feature matrix is ​​used as input, and the feature mapping function is applied. Perform forward propagation and output the predicted probability. .

[0073] S303: Employs the cross-entropy loss function Calculate the error between the predicted value and the actual value, where This represents the total number of training samples.

[0074] S304: Iteratively update the convolutional kernel weights using the backpropagation algorithm. and bias terms until the loss function The model training is complete when the convergence reaches below the preset threshold.

[0075] In real-time monitoring applications, the trained geological mutation prediction model is deployed on a GPU computing card (such as an NVIDIA T4 or equivalent) on a backend server, with a single inference calculation taking less than 0.5 seconds. Combined with data transmission and preprocessing time, the total latency from sensor data acquisition to model output prediction results does not exceed 2 seconds, meeting the time requirements for real-time early warning of geological mutations in deep foundation pit engineering. Those skilled in the art can select computing devices with different performance levels and adjust the sampling frequency and inference batches accordingly based on actual computing power configuration and engineering needs to achieve response latency requirements matching specific engineering projects.

[0076] During real-time monitoring, the model outputs the probability of occurrence of three types of geological anomalies. Through formula Calculate the probability of instability at each spatial location. ,Will The spatial coordinates corresponding to the maximum value are marked as the coordinates of the potential instability region, where, This serves as a weighting factor for various spatial anomalies. For example, in a certain monitoring session, the system calculated that the probability of a karst cave occurring at coordinates (12.5, 34.8, -20.0) was 0.85, and subsequently designated this coordinate as a key prevention and control area.

[0077] S4: Hierarchical response decision-making and parameter generation; such as Figure 3 As shown, this is a flowchart of the hierarchical response decision-making process in an embodiment of the present invention. The system initiates the hierarchical response mechanism based on the geological anomaly type output by S3:

[0078] (1) Karst cave response sub-mechanism: When the karst cave anomaly probability output by the model is greater than the preset probability threshold, the system retrieves geological survey data at the coordinates of the potential instability area and selects the grouting material according to the numerical range of the formation permeability coefficient k. ≥10 -4 cm / s, cement grout should be selected. <10 -4 cm / s, chemical pulp was selected; of which, 10 -4 cm / s is the minimum permeability coefficient threshold that cement grout can effectively penetrate into the formation under conventional grouting pressure. This value is determined based on conventional experience in the selection of grouting materials in the "Technical Specification for Building Foundation Treatment," and can be adjusted accordingly based on engineering geological conditions during actual construction. It is also based on the real-time monitored grout diffusion radius. With the target diffusion radius The difference Dynamically adjust the grouting flow rate until... The absolute value is less than the preset tolerance.

[0079] (2) Confined aquifer surge response sub-mechanism: When the probability of confined aquifer surge anomaly output by the model is greater than the preset probability threshold, the system generates a dewatering well opening and closing sequence and a support axial force compensation value; based on the real-time monitored water flow rate... Set precipitation speed Satisfying the relation ,in, A safety factor greater than 1; based on the water level fluctuation range. Set water level monitoring frequency ,when When >0.5m, Set to 1 time / minute, when When ≤0.5m, The setting is 1 time / 10 minutes. The 0.5m water level fluctuation threshold and the corresponding monitoring frequency are empirical values ​​determined based on conventional practices of deep foundation pit engineering dewatering monitoring. The purpose is to increase monitoring frequency when the water level changes rapidly in order to capture sudden changes in water pressure. In actual projects, these values ​​can be adjusted according to the depth of the foundation pit, the height of the confined water head, and the requirements of the surrounding environment.

[0080] (3) Weak interlayer response sub-mechanism: When the probability of weak interlayer anomaly output by the model is greater than the preset probability threshold, the system generates an adjusted support structure spacing. and axial force distribution parameters Set the support spacing adjustment amount With deformation Satisfying the relation ,in, As a positive coefficient, set the axial force compensation value. With change of internal force Satisfying the relation ,in, It is a positive coefficient to counteract the shear slip tendency of the weak interlayer.

[0081] S5: Dynamic closed-loop adjustment and feedback control; during the implementation of reinforcement measures, the system enters a closed-loop control process (see...). Figure 4 According to the triggering conditions of the graded response mechanism, corresponding reinforcement and emergency measures are implemented. During the implementation of reinforcement measures, the multimodal sensor array continuously collects microseismic signals, deformation and pore water pressure data in real time and feeds them back to the control system. The control system dynamically verifies the reinforcement effect and iteratively optimizes the construction parameters based on the feedback data until the state of the geologically abnormal area meets the stability judgment conditions.

[0082] The stability criteria are as follows: the microseismic energy in the affected area is below a preset safety threshold for 24 consecutive hours, the rate of change of deformation is below a preset rate limit for 24 consecutive hours, and the amplitude of pore water pressure fluctuation is below a preset amplitude limit for 24 consecutive hours. A stable state is determined when all three criteria are met simultaneously. The preset safety threshold, preset rate limit, and preset amplitude limit are pre-set by the site engineer based on the project level, foundation pit design requirements, and relevant specifications.

[0083] The specific process of dynamically verifying the reinforcement effect and iteratively optimizing construction parameters based on feedback data includes:

[0084] S501: Real-time acquisition of microseismic energy from feedback data Slurry diffusion radius Precipitation flow Rate of water level change Deformation and changes in internal forces .

[0085] S502: Match the corresponding dynamic adjustment algorithm based on the currently triggered geological anomaly type.

[0086] If the current scenario is grouting in a karst cave, then follow the formula. Dynamically adjust grouting pressure ,in, These are constants related to the mechanical properties of soil and rock masses; This is the slurry flow resistance attenuation factor.

[0087] For grouting scenarios in karst caves, the relevant constants of the mechanical properties of the rock and soil are as follows: and slurry flow resistance attenuation factor Calibration was achieved through on-site grouting tests; specifically, before formal grouting construction, at least three sets of test grouting under different grouting pressures were conducted in non-critical areas, and the microseismic energy of the grouting area was collected in real time. and the actual diffusion radius of the slurry Substituting the experimental data into the formula and fitting it using the least squares method, we obtain... and The value of ; in this embodiment The value is 0.85. The value is 0.6.

[0088] If the current situation is a sudden surge of confined aquifer, then follow the formula. Dynamically adjust the support axial force ,in, For the stiffness-related constants of the support structure, This refers to the nonlinear factor affecting hydraulic disturbance.

[0089] For scenarios involving sudden inrush of confined aquifers, the stiffness-related constants of the support structure are as follows: The influence of hydraulic disturbance on nonlinear factors The calibration was achieved through a combination of on-site pumping tests and load monitoring of the support structure. Specifically, before formal construction, pumping tests of no less than three drawdown levels were conducted around the foundation pit, with the drawdown flow rate collected simultaneously. Rate of water level change and measured values ​​of support axial force Substituting the experimental data into the formula and fitting it using the least squares method, we obtain... and The value of ; in this embodiment The value is 1.2 × 10 3 , The value is 0.75.

[0090] If the current scenario is a weak mezzanine, then follow the formula. Dynamically adjust the support spacing and according to the formula Dynamically adjust the support axial force, where, , These are constants related to the structural geometric properties. This is the sensitivity factor to changes in internal forces. This represents the deformation of the support structure as monitored in real time in the weak interlayer region, expressed in millimeters (mm). It is acquired by tilt sensors and / or 3D laser scanners and reflects the displacement change of the support structure relative to its initial stable state at the current moment.

[0091] For weak interlayer scenarios, structural geometric property-related constants , and internal force change sensitivity factor The calibration was performed using a combination of numerical simulation and on-site monitoring data. Specifically, a finite element numerical model was established based on the foundation pit design data to simulate the support structure response of the weak interlayer under different deformation and internal force variations, and multiple sets of data were extracted. , , The corresponding data were fitted using the least squares method. , and The value of ; in this embodiment The value is 0.5. The value is 2.0 × 10 3 , The value is 1.2.

[0092] The system monitors data trends in real time after parameter adjustments. A safety limit is set: if the microseismic energy exceeds 10... 3If the deformation exceeds 50mm, the system will automatically suspend operations and issue an alarm. Construction will resume after the data returns to a safe range, continuing until all monitoring data remain within the stable threshold range for 24 consecutive hours, confirming that the geological anomaly area has stabilized.

[0093] S503: The adjusted construction parameters are sent to the execution mechanism, and the micro-seismic energy and deformation change trends after parameter adjustment are monitored in real time. When the micro-seismic energy exceeds the preset safety limit or the deformation exceeds the preset warning value, the system automatically suspends the current reinforcement operation and issues an alarm. Construction resumes after the monitoring data returns to the safe range, until all monitoring data are within the stable threshold range for a continuous preset time.

[0094] When the model simultaneously identifies two or more types of geological anomalies (e.g., concurrent karst cavern and confined aquifer inrush), the system automatically activates a collaborative response mechanism. Based on the superimposed effects of various anomalies, the system uses a weighted fusion algorithm to comprehensively calculate the coupling parameters of grouting pressure, precipitation rate, and support axial force. For example, while initiating precipitation to reduce the confined aquifer head, the grouting pressure is appropriately increased to fill the voids in the cavern, preventing secondary collapse caused by stress redistribution due to precipitation. The system dynamically optimizes the parameter ratios of the combined scheme based on real-time monitoring data to ensure the comprehensive response effect under complex geological conditions, until all monitoring data continuously return to a stable threshold range within a preset time period, confirming the restoration of stability in the geological anomaly area. This completes a full closed-loop control cycle of "monitoring-early warning-decision-reinforcement-feedback."

[0095] A dynamic early warning system for deep foundation pit support structures based on deformation monitoring is provided to execute a dynamic early warning method for deep foundation pit support structures based on deformation monitoring, comprising:

[0096] Multimodal sensor arrays, such as Figure 1 As shown, the system is responsible for acquiring physical signals; specifically, it includes tilt sensors, strain gauges, microseismic monitors, and pore water pressure gauges. The outputs of each sensor are connected to the field edge gateway via an RS485 bus or LoRa / 5G wireless sensor network. Tilt sensors are deployed at the supports, joints, and connection nodes of the supporting structure to collect the tilt deformation of the supporting structure. Strain gauges are deployed at the main stress-bearing parts of the supporting structure, including the location of reinforcing bars, beam-column connection nodes, support feet, and bearing surfaces, to collect changes in the internal forces of the supporting structure. Microseismic monitors are deployed on the walls of deep foundation pits, retaining walls, and weak layers to collect micro-fracture signals in the soil and rock. Pore water pressure gauges are deployed at pore water pressure monitoring points at the bottom of the foundation pit or around the retaining structure to collect groundwater pressure data.

[0097] The data acquisition and transmission module is implemented by the on-site edge computing gateway. The gateway has a built-in ARM architecture processor, analog-to-digital conversion circuit and cache unit. It connects to each sensor in the multimodal sensor array through wireless transmission. It performs analog-to-digital conversion, filtering, noise reduction and normalization on the analog signals output by the sensors to realize real-time synchronous acquisition and edge caching of multimodal data. After receiving the enhanced monitoring command, it switches the sampling frequency of the corresponding area sensor from the normal mode to the 10Hz high-frequency sampling mode. At the same time, it sends a start scanning command to the 3D laser scanner. The acquired data is encrypted and uploaded to the back-end server through optical fiber or 5G private network.

[0098] The geological mutation feature triggering module, implemented by the embedded control logic in the edge computing gateway, is used to compare the gradient difference of deformation data collected by adjacent sensors in the same horizontal plane in real time. When the difference in deformation between any two sensors exceeds the preset threshold L1, and the microseismic energy increases by more than 50% within a 10-second time window, a geological mutation triggering signal is immediately generated, activating the local enhanced monitoring mode. The multimodal monitoring data and three-dimensional laser scanning data before and after the triggering time are packaged and uploaded to the spatiotemporal coupling analysis module.

[0099] Spatiotemporal coupling analysis module, such as Figure 2 As shown, the system is implemented by a background geological mutation early warning server. The server is equipped with a GPU computing card and a large-capacity storage array. The storage array contains a geological mutation prediction model optimized through transfer learning. This module is used to further fuse the monitoring data collected by various sensors and preprocessed by the edge gateway into a spatiotemporal feature matrix, which is then input into the geological mutation prediction model built based on a convolutional long short-term memory network. The model uses the GPU computing card to perform inference calculations and makes predictions using a training set containing three types of geological mutation samples: karst caves, weak interlayers, and confined aquifer inrush. It outputs the occurrence probability and potential instability area coordinates of the three types of geological anomalies in real time and pushes the prediction results to the hierarchical response decision module.

[0100] Hierarchical response decision module, such as Figure 3 As shown, the system is implemented by a field-programmable logic controller (PLC) or an industrial microcontroller. The controller has three pre-stored contingency plan libraries: grouting reinforcement, dewatering-support joint control, and structural reinforcement. This module is used to automatically match and call the corresponding contingency plan based on the geological anomaly type and instability coordinates output by the geological mutation early warning server, and generate executable construction parameters: when a karst cave is identified, a grouting construction instruction containing the grouting hole coordinates, grout diffusion radius, and pressure threshold is generated; when a confined aquifer surge is identified, a dewatering support instruction with a dewatering well opening and closing sequence and support axial force compensation value is generated; when a weak interlayer is identified, a structural reinforcement instruction with adjusted support structure spacing and axial force distribution parameters is generated, and the instructions are sent to the corresponding grouting pump, dewatering equipment, and hydraulic support actuator via fieldbus.

[0101] Dynamic feedback optimization module, such as Figure 4 As shown, the system is integrated into the graded response decision module and the on-site execution equipment, and is implemented by a closed-loop control algorithm and a real-time monitoring data channel. During the implementation of reinforcement measures, this module continuously collects feedback data on microseismic energy, grout diffusion radius, precipitation flow, water level change rate, deformation, and internal force changes through a multi-modal sensor array. Based on different geological anomaly types, it calls the corresponding dynamic adjustment algorithm, calculates and updates construction parameters in real time, and then sends them to the execution mechanism for adaptive adjustment. When the microseismic energy exceeds the preset safety limit or the deformation exceeds the preset warning value, an emergency stop control signal is immediately generated and sent to the execution equipment, and an audible and visual alarm is issued. The system continues until all monitoring data are within the stable threshold range for a preset period of time, at which point it is determined that the geological anomaly area has returned to stability, and the system automatically exits the enhanced monitoring mode and returns to the normal monitoring state.

[0102] In summary, this invention constructs a dynamic closed-loop control system encompassing the entire process of "perception-analysis-decision-execution-feedback." By leveraging the three-dimensional perception of a multimodal sensor array and the millisecond-level response of geological abrupt change triggers, it solves the problem of delayed identification of micro-deformation and micro-fracture precursors using traditional monitoring methods. Based on a spatiotemporal coupling analysis model and a ConvLSTM algorithm optimized through transfer learning, it achieves accurate classification and spatial localization of three typical geological anomalies: karst caves, weak interlayers, and sudden inrushes of confined aquifers, replacing the fuzzy judgments of traditional empirical methods. More importantly, through the deep coupling of a hierarchical response mechanism and a dynamic closed-loop optimization algorithm, reinforcement measures are transformed from static presets into an adaptive process that evolves in real time with the geological and structural conditions. This invention not only significantly shortens the time window from hazard identification to emergency response but also greatly enhances the proactive defense capabilities and inherent safety level of deep foundation pits under complex geological abrupt changes through quantitative parameter control, demonstrating extremely high engineering application value and promising prospects for widespread adoption.

[0103] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall fall within the scope of the present invention.

Claims

1. A dynamic early warning method for deep foundation pit support structures based on deformation monitoring, characterized in that, Includes the following steps: S1: Deploy a multimodal sensor array at key points of the support structure. The multimodal sensor array includes tilt sensors, strain gauges, microseismic monitors and pore water pressure gauges to simultaneously collect the deformation of the support structure, changes in internal forces, micro-fracture signals of the soil and rock mass and groundwater pressure data. S2: Establish a geological abrupt change feature trigger. By comparing the gradient difference of adjacent sensor data in real time, when the difference in deformation between any two sensors in the same horizontal plane exceeds the threshold and the increase in microseismic energy within the preset time window exceeds the preset percentage threshold, the local enhanced monitoring mode is immediately activated, the sampling frequency of the affected area is increased to the preset sampling frequency and three-dimensional laser scanning is started. S3: Construct a spatiotemporal feature matrix and input the spatiotemporal feature matrix into a geological mutation prediction model optimized by transfer learning. The geological mutation prediction model is constructed based on a convolutional long short-term memory network and is pre-trained through a training set containing three types of geological mutation samples: karst caves, weak interlayers, and confined aquifers. It outputs the probability of geological anomaly types and the coordinates of potential unstable areas in real time. S4: Activate the graded response mechanism based on the geological anomaly type output by the model: When a karst cave is identified, automatically trigger the grouting reinforcement decision and generate construction parameters including grouting hole coordinates, grout diffusion radius and pressure threshold; when a confined aquifer surge is identified, initiate the dewatering-support joint control and generate the dewatering well opening and closing sequence and support axial force compensation value; when a weak interlayer is identified, automatically trigger the structural strengthening decision and generate the adjusted support structure spacing and axial force distribution parameters.

2. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 1, characterized in that: The key points of the support structure in step S1 include the supports, joints, connection nodes, main stress-bearing parts, deep foundation pit walls, retaining walls, weak layer areas, the bottom of the foundation pit, and pore water pressure monitoring points around the retaining structure. Tilt sensors are deployed at supports, joints, and connection nodes of the support structure to monitor changes in the tilt of the support structure. Strain gauges are deployed at the main stress-bearing parts of the supporting structure, including the location of reinforcing bars, beam-column connection nodes, support feet, and bearing surfaces, to monitor changes in internal forces in real time. Microseismic monitors are deployed in deep foundation pit walls, retaining walls, and weak layer areas to detect microcracks and rupture signals in the soil and rock mass; Pore ​​water pressure gauges are deployed at pore water pressure monitoring points at the bottom of the foundation pit or around the retaining structure to monitor the dynamic changes in groundwater pressure. All sensors are connected to the data acquisition system via wireless transmission modules to achieve real-time synchronous data acquisition and transmission.

3. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 1, characterized in that: The construction process of the spatiotemporal feature matrix in S3 includes, The multimodal data collected by enhanced monitoring are filtered, denoised, and normalized. Extract the deformation corresponding to time t and spatial position s from the processed data. Stress changes Microseismic energy changes and pore water pressure changes Integrated into a four-dimensional spatiotemporal feature matrix ; in, The row dimension corresponds to the time series, the column dimension corresponds to the spatial point sequence, and the matrix elements correspond to the quantized values ​​of the multimodal monitoring parameters.

4. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 1, characterized in that: The geological mutation prediction model in S3 is as follows: A two-branch fusion model is constructed using a Convolutional Long Short-Term Memory (ConvLSTM) network. The first branch is a Convolutional Neural Network (CNN) module, which is used to extract the local spatial correlation features of deformation, stress change, microseismic energy change and pore water pressure change in the spatiotemporal feature matrix. The second branch is the Long Short-Term Memory (LSTM) module, which is used to model the time series evolution of the spatiotemporal feature matrix. The spatial feature map output by the CNN module and the temporal feature vector output by the LSTM module are concatenated and fused, and then input into the fully connected layer to output the occurrence probability of three types of geological anomalies: karst caves, weak interlayers, and confined aquifers.

5. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 4, characterized in that: The training process for geological catastrophe prediction models includes, S301: Initialize model parameters, loading the pre-trained model weights trained using historical monitoring data from similar geological regions as the initial convolutional kernel weights. and bias terms ; S302: Labels with actual geological anomalies The historical spatiotemporal feature matrix is ​​used as input, and the feature mapping function is applied. Perform forward propagation and output the predicted probability. ; S303: Employs the cross-entropy loss function Calculate the error between the predicted value and the actual value, where The total number of training samples; S304: Iteratively update the convolutional kernel weights using the backpropagation algorithm. and bias terms until the loss function The model training is complete when the convergence reaches below the preset threshold.

6. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 5, characterized in that: The coordinates of the potential instability region in S3 are determined using the following method, based on the probability of occurrence of three types of geological anomalies output by the trained geological mutation prediction model. Combined with pre-defined spatial influence weighting factors for various anomalies Through formula Calculate the probability of instability at each spatial location. ,Will The spatial coordinates corresponding to the maximum value are marked as the coordinates of the potential instability region.

7. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 1, characterized in that: The specific implementation process of the hierarchical response mechanism described in step S4 includes, The karst cave response mechanism involves retrieving geological survey data from the potential instability zone coordinates when the model's output probability of karst cave anomalies exceeds a preset probability threshold. Based on the numerical range of the formation permeability coefficient k, the system selects the grouting material. ≥10 -4 cm / s, cement grout should be selected. <10 -4 cm / s, chemical slurry is selected; and the slurry diffusion radius is determined based on real-time monitoring. With the target diffusion radius The difference Dynamically adjust the grouting flow rate until... The absolute value is less than the preset tolerance; The confined aquifer inrush response sub-mechanism generates a dewatering well opening / closing sequence and support axial force compensation value when the model outputs an inrush anomaly probability greater than a preset probability threshold; based on the real-time monitored water inrush flow rate... Set precipitation speed Satisfying the relation ,in, A safety factor greater than 1; based on the amplitude of water level fluctuations. Set water level monitoring frequency ,when When >0.5m, Set to 1 time / minute, when When ≤0.5m, Set to 1 time / 10 minutes; The weak interlayer response sub-mechanism generates an adjusted support structure spacing when the model outputs an anomaly probability of the weak interlayer that exceeds a preset probability threshold. and axial force distribution parameters Set the support spacing adjustment amount With deformation Satisfying the relation ,in, As a positive coefficient, set the axial force compensation value. With change of internal force Satisfying the relation ,in, It is a positive coefficient to counteract the shear slip tendency of the weak interlayer.

8. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 1, characterized in that: It also includes S5: Based on the triggering conditions of the graded response mechanism, corresponding reinforcement and emergency measures are implemented. During the implementation of reinforcement measures, the multimodal sensor array continuously collects microseismic signals, deformation and pore water pressure data in real time and feeds them back to the control system. The control system dynamically verifies the reinforcement effect and iteratively optimizes the construction parameters based on the feedback data until the state of the geologically abnormal area meets the stability judgment conditions.

9. The dynamic early warning method for deep foundation pit support structures based on deformation monitoring according to claim 8, characterized in that: The specific process of dynamically verifying the reinforcement effect and iteratively optimizing construction parameters based on feedback data includes: S501: Real-time acquisition of microseismic energy from feedback data slurry diffusion radius Precipitation flow Rate of water level change Deformation and changes in internal forces ; S502: Match the corresponding dynamic adjustment algorithm based on the currently triggered geological anomaly type. If the current scenario is grouting in a karst cave, then follow the formula. Dynamically adjust grouting pressure ,in, These are constants related to the mechanical properties of soil and rock masses; It is the attenuation factor of slurry flow resistance; If the current situation is a sudden surge of confined aquifer, then follow the formula. Dynamically adjust the support axial force ,in, For the stiffness-related constants of the support structure, The nonlinear factor affecting hydraulic disturbance; If the current scenario is a weak mezzanine, then follow the formula. Dynamically adjust the support spacing and according to the formula Dynamically adjust the support axial force, where, , These are constants related to the structural geometric properties. Sensitivity factor for changes in internal force; S503: The adjusted construction parameters are sent to the execution mechanism, and the micro-seismic energy and deformation change trends after parameter adjustment are monitored in real time. When the micro-seismic energy exceeds the preset safety limit or the deformation exceeds the preset warning value, the system automatically suspends the current reinforcement operation and issues an alarm. Construction resumes after the monitoring data returns to the safe range, until all monitoring data are within the stable threshold range for a continuous preset time.

10. A dynamic early warning system for deep foundation pit support structures based on deformation monitoring, used to execute the dynamic early warning method for deep foundation pit support structures based on deformation monitoring as described in any one of claims 1-9, characterized in that: include, The multimodal sensor array includes tilt sensors, strain gauges, microseismic monitors, and pore water pressure gauges. The tilt sensors are deployed at the supports, joints, and connection nodes of the supporting structure to collect the tilt deformation of the supporting structure. The strain gauges are deployed at the main stress-bearing parts of the supporting structure, including the location of reinforcing bars, beam-column connection nodes, support feet, and bearing surfaces, to collect the internal force changes of the supporting structure. The microseismic monitors are deployed in the deep foundation pit walls, retaining walls, and weak layer areas to collect micro-fracture signals in the soil and rock mass. Pore ​​water pressure gauges are deployed at pore water pressure monitoring points at the bottom of the foundation pit or around the retaining structure to collect groundwater pressure data; The data acquisition and transmission module connects to each sensor in the multimodal sensor array and realizes real-time synchronous acquisition and transmission of multimodal data through wireless transmission. The geological mutation feature triggering module is used to compare the gradient difference of adjacent sensor data in real time. When the difference in deformation between any two sensors in the same horizontal plane exceeds the threshold and the increase in microseismic energy within the preset time window exceeds the preset percentage threshold, the local enhanced monitoring mode is activated, the sampling frequency of the affected area is increased to the preset sampling frequency and three-dimensional laser scanning is started. The spatiotemporal coupling analysis module is used to fuse multimodal data collected by various sensors into a spatiotemporal feature matrix, which is then input into a geological mutation prediction model optimized by transfer learning. The geological mutation prediction model is constructed based on a convolutional long short-term memory network and is pre-trained using a training set containing three types of geological mutation samples: karst caves, weak interlayers, and confined aquifers. It outputs the probability of geological anomaly types and the coordinates of potential unstable areas in real time. The graded response decision module is used to initiate a graded response mechanism based on the geological anomaly type output by the geological mutation prediction model: when a karst cave is identified, grouting reinforcement decision is automatically triggered, generating construction parameters including grouting hole coordinates, grout diffusion radius, and pressure threshold; when a confined aquifer surge is identified, dewatering-support joint control is initiated, generating dewatering well opening and closing sequence and support axial force compensation value; when a weak interlayer is identified, structural strengthening decision is automatically triggered, generating adjusted support structure spacing and axial force distribution parameters. The dynamic feedback optimization module is used to dynamically verify the reinforcement effect and iteratively optimize the construction parameters based on the real-time monitoring data fed back by the multimodal sensor array during the implementation of reinforcement measures, until the state of the geologically abnormal area meets the stability judgment conditions.

Images