Intelligent early warning system for safety of buildings around foundation pit construction

Abstract

The invention discloses an intelligent early warning system for safety of buildings around foundation pit construction, and relates to the technical field of construction safety monitoring. The platform adopts a four-layer architecture of a sensing layer, a transmission layer, a platform layer and an application layer. The sensing layer collects data in real time through a multi-class sensor array; the transmission layer wirelessly transmits data by using the Internet of Things technology; the platform layer serves as a core, comprises a data management module, a digital twin modeling module and an intelligent analysis early warning engine, and is responsible for data preprocessing, three-dimensional model driving and intelligent risk study and judgment based on machine learning and a dynamic threshold value; and the application layer provides visual display and early warning push service through a Web terminal and a mobile APP. All-weather automatic monitoring, intelligent advanced early warning and accurate risk prevention and control of the surrounding environment of the foundation pit are achieved, a complete intelligent management closed loop is formed, and the construction safety level is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of civil engineering construction safety monitoring technology, specifically to an intelligent early warning system for the safety of surrounding buildings during foundation pit construction, used for real-time safety monitoring and intelligent early warning of surrounding buildings and structures during foundation pit construction. Background Technology

[0002] With the acceleration of urbanization, the number of deep foundation pit projects is increasing. Foundation pit excavation will cause changes in the stress field and displacement field of the surrounding soil, which will have an adverse effect on nearby buildings and structures (such as houses, bridges, tunnels, underground pipelines, etc.), and may lead to uneven settlement, tilting, cracking or even collapse, seriously threatening people's lives and property safety.

[0003] Currently, safety monitoring of the environment surrounding foundation pits mainly relies on traditional methods, namely, deploying sensors such as settlement points, inclinometers, and joint gauges on buildings and structures, with technicians conducting regular manual inspections and data collection. This method has the following significant drawbacks: 1. Poor timeliness: Data collection is intermittent, making it impossible to achieve continuous monitoring around the clock and difficult to capture rapid deformation induced during critical construction stages.

[0004] 2. Delayed early warning: Data analysis relies on human experience and is highly subjective. Early warnings are often issued only after the deformation has reached a certain level, leaving a very short window for emergency response.

[0005] 3. Information silos: Monitoring data, construction progress data, geological data, etc. are isolated from each other, lacking effective linkage and deep integration, making it difficult to accurately locate risk sources.

[0006] 4. Insufficient global coverage: Point-based monitoring cannot fully reflect the overall deformation state and spatial effects of buildings and structures, resulting in monitoring blind spots.

[0007] In recent years, although some automated monitoring systems have emerged, enabling automatic data collection and transmission, there are still shortcomings in in-depth data mining, intelligent analysis, accurate early warning, and visualization, failing to form a complete intelligent closed loop of "perception-cognition-early warning-decision".

[0008] Therefore, there is an urgent need in this field for a comprehensive platform that can achieve all-weather, fully automated, intelligent analysis and early warning, so as to proactively perceive and accurately control the surrounding environmental risks caused by the entire process of foundation pit construction. Summary of the Invention

[0009] The present invention proposes an intelligent early warning system, equipment and storage medium for the safety of buildings surrounding the foundation pit construction, which can at least solve one of the technical problems in the background art.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: An intelligent early warning system for the safety of buildings surrounding a foundation pit construction site includes a perception layer, a transmission layer, a platform layer, and an application layer connected in sequence. The system architecture is used to execute the following intelligent early warning process: S100. Based on the foundation pit engineering design plan and the surrounding environmental risk assessment report, construct the early warning system perception layer; S200. Construct the transmission layer of the early warning system to transmit the monitoring data collected by the perception layer to the platform layer of the early warning system; S300: Construct an early warning system platform layer to manage monitoring data and drive the digital twin model to update the data; S400 uses an intelligent analysis and early warning engine combined with data updated by a digital twin model to perform trend prediction, coupled analysis, and risk assessment. If the dynamic threshold is exceeded, it will automatically generate early warning information of the corresponding level. S500: Construct an early warning system application layer to guide engineering decisions and emergency responses based on early warning information.

[0011] Furthermore, the sensing layer of the present invention comprises an array of intelligent sensors deployed inside the foundation pit, in the surrounding soil, and on the target building; The sensor array is deployed to cover the foundation pit itself, the surrounding soil, and all target buildings that need to be protected. When the target building involves ancient buildings, a settlement monitoring network is formed using hydrostatic levels with an accuracy of ±0.5mm. Dual-axis tilt sensors are installed at the corners and center of the structure, with a range of not less than ±5° and an accuracy of not less than ±0.001°. Strain sensors are installed on key load-bearing components, with the range selected based on 1.5 times the estimated axial force.

[0012] Furthermore, the transmission layer of this invention intelligently selects and combines multiple communication technologies such as 4G / 5G, LoRa, and NB-IoT according to data type, bandwidth requirements, and field environment to form a heterogeneous network; The method of intelligently selecting and combining multiple communication technologies such as 4G / 5G, LoRa, and NB-IoT includes: based on the priority of data packets. Estimated transmission delay and link quality Dynamically select the optimal communication link; select the decision function F selec Represented as:

[0013] in, For link quality indicators of specific technologies, The estimated transmission delay is calculated using α, β, and γ, which are weighting coefficients used to balance link quality, delay, and service priority. Link quality assessment and adaptive coding: The transport layer continuously assesses the quality of the wireless channel, and for narrowband IoT links, its maximum path loss is... Based on the link budget formula, the following is estimated:

[0014] in, For transmission power, and These represent the transmit and receive antenna gains, respectively. For receiver noise figure, To achieve the required signal-to-noise ratio for demodulation This is the fading margin; Based on real-time calculated channel capacity Adaptively adjust the modulation and coding scheme (MCS) to approximate the channel capacity and ensure reliable data transmission.

[0015] in, For channel bandwidth, For receiving signal-to-noise ratio; Probability of successful data packet transmission: For transmissions using forward error correction (FEC) coding, the probability that a data packet will be successfully received. The model is as follows:

[0016] in, The channel bit error rate is closely related to the signal-to-noise ratio (SNR) and the selected modulation scheme (e.g., for BPSK modulation, in an additive white Gaussian noise channel, ...). , The length of the data packet; the transport layer evaluates... To decide whether to perform data fragmentation or retransmission; Energy consumption model: Energy consumed in a single data transmission Simplified to:

[0017] in, For circuit power consumption, The efficiency coefficient of the power amplifier. For transmission distance, This is the path loss index. To optimize transmission time, the transport layer optimizes the transmit power. It employs a low-power sleep-wake mechanism to maximize the overall battery life of sensor nodes.

[0018] Furthermore, the platform layer described in this invention is responsible for in-depth processing, model-driven analysis, and intelligent analysis of the monitoring data; The in-depth processing methods for monitoring data include: State prediction:

[0019]

[0020] Measurement Update:

[0021]

[0022]

[0023] in, This is the state estimate. Let be the error covariance matrix. Here is the state transition matrix. Let u be the control input matrix and u be the control vector. The process noise covariance matrix is... For the observation matrix, To observe the noise covariance matrix, For Kalman gain, These are actual observed values.

[0024] Furthermore, the intelligent analysis method for monitoring data at the platform layer of this invention includes: LSTM neural networks are used to predict the trends of key monitoring indicators for future periods. LSTM utilizes its gating mechanism and cell state... To process time series data, the core formula is as follows: Forgotten Gate:

[0025] Input Gate:

[0026] Candidate cell status:

[0027] Cell status update:

[0028] Output gate:

[0029] Hidden state output:

[0030] in, It is the sigmoid activation function. and For model parameters, for Input at any time, The model ultimately outputs the predicted value through a fully connected layer, which represents the hidden state from the previous time step. ; A risk assessment model is constructed using machine learning ensemble algorithms, which integrates real-time monitoring values ​​X. monitor Predicted rate of change v, construction condition data X construction and environmental data X environment Based on multidimensional features, a comprehensive risk probability score is output. The XGBoost model optimizes the following objective function through additive training:

[0031] in, , The number of leaf nodes. Leaf weights and For regularization parameters; Model output Obtained by conversion using the sigmoid function; Multi-level early warning judgment unit: preset three-level early warning mechanism of yellow, orange and red; Its judgment threshold is dynamically adjusted based on the predicted trend, rate of change, and coupling analysis results; Dynamic threshold Represented as:

[0032] in, This is a static control value. The current rate of change, For the maximum permissible rate of change, The overall risk probability score, and The weighting coefficients are used to comprehensively consider the percentage of monitored values ​​exceeding static thresholds, the duration of predicted value exceeding limits, and other factors when triggering early warnings. The risk zone in which it is located.

[0033] Furthermore, the preset yellow, orange, and red three-level early warning mechanism of the present invention includes: Yellow alert: When the monitored value exceeds the static control value 60% Three consecutive acquisition cycles exceeding the allowable rate, or When the value is >0.3, the system will issue a prompt on the web interface and notify the on-site safety officer to strengthen patrols; Orange alert: When the monitored value exceeds 80% of the values ​​predicted by LSTM will exceed the limit within the next 48 hours, or When the value exceeds 0.6, the system will immediately send an alarm to the project manager and chief engineer via the APP, suggesting that a risk meeting be held and measures such as adjusting the excavation sequence and increasing the monitoring of the project be taken. Red Alert: When the monitored value has exceeded Predictions indicate that the deformation will get out of control or When the threshold is >0.9, the system will forcibly send the highest level alarm to all key responsible persons, require the platform to immediately lock the construction permit process, forcibly stop work in dangerous areas, and activate the emergency plan.

[0034] In summary, this invention achieves continuous, all-weather, fully automated, and blind-spot-free monitoring of the environment surrounding the foundation pit through intelligent sensors, replacing inefficient manual inspections. Through machine learning and trend prediction, it shifts from "post-event analysis" to "pre-event early warning," identifying potential risks and buying valuable time for intervention. By fusing multi-source information, it deeply integrates monitoring data with construction conditions and geological features, accurately locating risk sources and improving judgment accuracy. Through digital twin technology, it visually visualizes abstract data in a three-dimensional model, making risks "visible, manageable, and controllable," greatly improving management efficiency. Through a multi-level early warning mechanism and mobile push notifications assigning responsibility to individuals, it ensures that early warning information quickly reaches key responsible persons, forming a closed-loop management system of "monitoring-early warning-response-feedback-verification." Attached Figure Description

[0035] Figure 1 This is a system architecture diagram of the intelligent early warning platform of the present invention; Figure 2 This is a flowchart of the intelligent analysis and early warning engine in the platform of this invention; Figure 3 This is a schematic diagram of the digital twin interface of the application layer Web visualization terminal of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0037] like Figure 1 As shown in the figure, the intelligent early warning platform for the safety of buildings surrounding the foundation pit construction described in this embodiment includes a perception layer, a transmission layer, a platform layer and an application layer connected in sequence.

[0038] The sensing layer consists of a variety of intelligent sensor arrays deployed inside the foundation pit, in the surrounding soil, and on the target building (structure), used to collect multi-dimensional monitoring data of the physical world in real time; The sensor array includes, but is not limited to: a hydrostatic level, a micro-differential pressure settlement meter, an inclinometer, an accelerometer, a strain sensor, a crack gauge, a high-definition network camera, and a GNSS receiver.

[0039] The transmission layer employs IoT communication technology to reliably and efficiently transmit data collected by the sensing layer wirelessly to the platform layer. This layer intelligently selects and combines multiple communication technologies, such as 4G / 5G, LoRa, and NB-IoT, based on data type, bandwidth requirements, and the site environment, to form an adaptive heterogeneous network.

[0040] To optimize network performance and reliability, the transport layer employs the following core mechanisms and algorithms: Intelligent selection mechanism for communication technology: based on data packet priority. (e.g., real-time video is of high priority) Periodic settlement data is of ordinary priority. ), estimated transmission delay and link quality (e.g., signal-to-noise ratio SNR), dynamically select the optimal communication link. The selection decision function is F. select It can be represented as:

[0041] in, For link quality indicators of specific technologies, The estimated transmission delay is calculated using α, β, and γ, which are weighting coefficients used to balance link quality, delay, and service priority.

[0042] Link quality assessment and adaptive coding: The transport layer continuously assesses the quality of the wireless channel. For narrowband IoT links (such as NB-IoT, LoRa), the maximum path loss is... (Unit: dB) can be estimated using the link budget formula:

[0043] in, For transmission power, and These represent the transmit and receive antenna gains, respectively. For receiver noise figure, To achieve the required signal-to-noise ratio for demodulation This is the fading margin.

[0044] Based on real-time calculated channel capacity (Given by Shannon's formula), the modulation and coding scheme (MCS) is adaptively adjusted to approximate the channel capacity and ensure the reliability of data transmission.

[0045] in, For channel bandwidth, For received signal-to-noise ratio.

[0046] Data packet successful transmission probability: In complex construction site environments, successful data packet transmission is crucial. For transmissions using forward error correction (FEC) coding, the probability of a data packet being successfully received is... It can be modeled as:

[0047] in, The channel bit error rate is closely related to the signal-to-noise ratio (SNR) and the selected modulation scheme. For BPSK modulation, in an additive white Gaussian noise channel, , This refers to the packet length (in bits). The transport layer evaluates... This determines whether data fragmentation or retransmission is necessary.

[0048] Energy Consumption Model (for Battery-Powered Sensors): For battery-powered sensor nodes deployed in remote locations, energy management at the transport layer is crucial. This includes the energy consumed in a single data transmission. It can be simplified to:

[0049] in, For circuit power consumption, The efficiency coefficient of the power amplifier. For transmission distance, This is the path loss index. For transmission time. The transport layer optimizes the transmit power. (and (Related) and employ a low-power sleep-wake mechanism to maximize the overall battery life of sensor nodes.

[0050] The platform layer, serving as the core processing hub, is deployed on a cloud server or local server and is responsible for in-depth data processing, model-driven operations, and intelligent analysis. The platform layer includes: Data governance module: Used to perform preprocessing such as cleaning, filtering, normalization and time alignment on the received multi-source and heterogeneous real-time data, so as to provide high-quality and standardized datasets for subsequent analysis.

[0051] Data cleaning: A Z-score-based outlier detection method is used to automatically identify and remove abnormal data points caused by momentary sensor malfunctions or signal interference. For data points... The formula for calculating its Z-score is:

[0052] in, μ The mean of the data sequence. σ Let | be the standard deviation. When |>3, it is judged as an outlier and removed.

[0053] Data filtering: A Kalman filter is used to smooth the data, effectively eliminating random noise and preserving the true trend of data changes. Its core state-space model and iterative process are as follows: State prediction:

[0054]

[0055] Measurement Update:

[0056]

[0057]

[0058] in, This is the state estimate. Let be the error covariance matrix. Here is the state transition matrix. Let u be the control input matrix and u be the control vector. The process noise covariance matrix is... For the observation matrix, To observe the noise covariance matrix, For Kalman gain, These are actual observed values.

[0059] Data normalization: The min-max normalization method is used to uniformly transform monitoring data of different dimensions and orders of magnitude to the [0,1] interval, as shown in the following formula:

[0060] in, and These are the minimum and maximum values ​​of the monitoring indicator in the historical dataset, respectively.

[0061] Data alignment: Based on a unified timestamp, time series alignment is performed on multi-source heterogeneous data from different sensors. Interpolation methods (such as linear interpolation) are used to ensure data consistency along the time axis for two adjacent time points. and Data and At the target time point interpolation for:

[0062] Digital Twin Modeling Module: Based on BIM and GIS technologies, this module constructs a high-precision 3D visualization model that includes the foundation pit, surrounding soil, and buildings / structures. Through spatial mapping functions, it dynamically associates and drives real-time monitoring data with corresponding components in the model.

[0063] For settlement data, the model's elevation changes are driven. Let the original elevation of a point be... Real-time settlement value Then its real-time elevation in the model for:

[0064] For tilt data, the model's pose is rotated. The tilt angles θ and φ, measured by tilt sensors, are converted into a rotation matrix of the model in three-dimensional space. (Represented in Euler angles or quaternions) to visualize the small tilt of the model.

[0065] Spatial interpolation algorithms (such as Inverse Distance Weighting (IDW) or Kriging) are used to transform discrete point monitoring data into a continuous spatial distribution field, which is then rendered as a contour map on the model surface. In the IDW method, unsampled points... value The calculation formula is:

[0066] in, Known monitoring points The value, It is a point arrive distance, It is an exponential parameter (usually 2).

[0067] Intelligent Analysis and Early Warning Engine: This engine is the intelligent core of the platform layer, integrating trend prediction, coupled analysis, and multi-level early warning judgment functions.

[0068] Trend Prediction Unit: Employs an LSTM (Long Short-Term Memory) neural network to predict the trends of key monitoring indicators (such as subsidence and tilt) over future time periods (e.g., 24-72 hours). LSTM utilizes its gating mechanism (forget gate)... Input gate Output gate ) and cell state To process time series data, the core formula is as follows: Forgotten Gate:

[0069] Input Gate:

[0070] Candidate cell status:

[0071] Cell status update:

[0072] Output gate:

[0073] Hidden state output:

[0074] in, It is the sigmoid activation function. and For model parameters, for Input at any time, This represents the hidden state from the previous time step. The model ultimately outputs the predicted value through a fully connected layer. .

[0075] Coupling Analysis Unit: A risk assessment model is constructed using machine learning ensemble algorithms (such as XGBoost). This model integrates real-time monitoring values ​​X... monitor Predicted rate of change v, construction condition data X construction and environmental data X environment Based on multidimensional features, a comprehensive risk probability score is output. The XGBoost model optimizes the following objective function through additive training:

[0076] in, , The number of leaf nodes. Leaf weights and These are the regularization parameters. The model's output. It is obtained by conversion using the sigmoid function.

[0077] Multi-level early warning judgment unit: A preset three-level early warning mechanism of yellow, orange, and red. Its judgment thresholds are not fixed values, but rather intelligent thresholds that are dynamically adjusted based on predicted trends, rates of change, and coupling analysis results. Dynamic thresholds. It can be represented as:

[0078] in, This is a static control value. The current rate of change, For the maximum permissible rate of change, The overall risk probability score, and The weighting coefficients are used to determine the early warning trigger conditions, which comprehensively consider the percentage of monitored values ​​exceeding static thresholds, the duration of predicted value exceeding limits, and other factors. The risk zone in which it is located.

[0079] The application layer provides users with an interactive interface and early warning services, including: Web visualization terminal: Displays monitoring data, risk levels and spatial distribution in a three-dimensional manner in a digital twin model, and highlights early warning information in the form of sound, light and color change; Mobile App: Pushes early warning information to designated responsible persons, and supports remote viewing of monitoring status and receiving alarm notifications; Automatic report generation module: Generates monitoring and analysis reports on demand.

[0080] like Figure 2 As shown, based on the aforementioned intelligent early warning platform, this invention provides a method for operating an intelligent early warning platform, which includes the following steps: S100. Based on the foundation pit engineering design plan and the surrounding environmental risk assessment report, construct the early warning system perception layer; S200. Construct the transmission layer of the early warning system to transmit the monitoring data collected by the perception layer to the platform layer of the early warning system; S300: Construct an early warning system platform layer to manage monitoring data and drive the digital twin model to update the data; S400 uses an intelligent analysis and early warning engine combined with data updated by a digital twin model to perform trend prediction, coupled analysis, and risk assessment. If the dynamic threshold is exceeded, it will automatically generate early warning information of the corresponding level. S500: Construct an early warning system application layer to guide engineering decisions and emergency responses based on early warning information.

[0081] The following provides a detailed explanation of each step: S100. Based on the foundation pit engineering design plan and the surrounding environmental risk assessment report, construct the early warning system perception layer; Before the foundation pit project started, the monitoring points were optimized and laid out according to the design plan and the surrounding environmental risk assessment report.

[0082] 1. Deployment location: The sensor array needs to cover the foundation pit body (such as the top of the support piles, the cap beam, and the support), the surrounding soil (such as the soil inclinometer tube and the pore water pressure gauge), and all target buildings (structures) that need to be protected (such as foundations, load-bearing columns, exterior walls, and existing cracks).

[0083] 2. Sensor selection and configuration: For structures sensitive to settlement, such as historical buildings, a settlement monitoring network should be formed using hydrostatic levels with an accuracy higher than ±0.5 mm.

[0084] Dual-axis tilt sensors are installed at the corners and center of the structure, with a range of not less than ±5° and an accuracy of not less than ±0.001°.

[0085] Strain sensors should be installed on critical load-bearing components (such as foundation pit supports), and the range should be selected based on 1.5 times the estimated axial force of the design.

[0086] After deployment, all sensors need to undergo initial value acquisition and calibration, and their unique ID, physical location, range, alarm initial threshold, and other information should be entered into the platform.

[0087] 3. Data Acquisition Strategy: The data acquisition frequency is typically set to once per minute to once per hour. During critical phases such as foundation pit excavation, dewatering, and foundation slab pouring, the system automatically increases the acquisition frequency to once per minute to achieve high-frequency monitoring. Work data is automatically obtained from the project management system via a standard API interface.

[0088] S200. Construct the transmission layer of the early warning system to transmit the monitoring data collected by the perception layer to the platform layer of the early warning system; To build a highly reliable and adaptive field IoT network, specific networking solutions include: L1. Select widely distributed sensors with small data volume and low power consumption requirements (such as sedimentation meters and inclinometers) and build a low-power wide area network using LoRa or NB-IoT technology. A single gateway can cover an area with a radius of 2-3 kilometers.

[0089] L2. For high-bandwidth data streams such as high-definition video surveillance, use 5GCPE or wired networks for transmission.

[0090] L3. Deploy an industrial-grade gateway on-site to handle protocol conversion, initial data encapsulation, and local caching.

[0091] Transmission Guarantee: The gateway's built-in communication strategy engine dynamically selects the optimal link based on a preset decision function, as follows:

[0092] Meanwhile, forward error correction and retransmission mechanisms are employed to ensure a high probability of successful data packet transmission in complex construction site environments. Maintained at over 99.9%. All transmitted data is encrypted to ensure information security.

[0093] S300: Construct an early warning system platform layer to manage monitoring data and drive the digital twin model to update the data; The platform-level data governance module standardizes the massive amounts of multi-source data that are imported, providing a high-quality data foundation for subsequent analysis.

[0094]

[0095] 1. Data Cleaning: An automated outlier detection algorithm based on Z-score is used. For the time series data of any monitoring point { , ,…, }, calculate its mean and standard deviation For data points If its Z-score value| |=|( one ) / If the value is greater than 3, it is considered a gross error, automatically removed, and logged. The system supports manual review and recovery.

[0096] 2. Data Filtering: Kalman filtering is used to smooth the data in real time. Taking settlement data as an example, the state vector is defined as settlement value and settlement rate. The state prediction equation is then used... and measurement update equation The iterative process effectively filters out random noise in the signal and outputs the optimal estimate. (Process noise covariance) and observation noise covariance The sensor needs to be calibrated during the on-site commissioning phase, depending on its accuracy.

[0097] 3. Data Normalization and Alignment: The min-max normalization method is used to unify data of different dimensions into the [0,1] interval. The formula is as follows: Meanwhile, using the BeiDou satellite timekeeping as a benchmark, all data are stamped with a unified timestamp, and linear interpolation is used to solve the problem of slight asynchrony in the acquisition time of different sensors, forming a well-organized dataset that can be used for fusion analysis.

[0098] The digital twin modeling module serves as a bridge connecting the physical world and the virtual space.

[0099] 1. Model Construction: Based on design drawings and on-site survey data, a detailed 3D model of the foundation pit support structure and surrounding buildings is created using BIM software (such as Revit), down to the main beams, columns, and slabs, with a model refinement level of LOD350 or higher. Simultaneously, topographic, geomorphological, and underground pipeline data of the project site and its surrounding area are integrated using a GIS platform. Finally, the lightweight BIM model is imported into the GIS scene, forming a complete BIM-GIS integrated digital twin foundation.

[0100] 2. Data Mapping and Driving: A. Geometric-driven: Preprocessed monitoring data is injected into the model in real time via API. For example, settlement values. Directly drive the elevation of the corresponding points in the model, making it change from the original elevation. Become The tilt angle data, through coordinate transformation, drives a slight three-dimensional rotation of the building model.

[0101] B. Visualization Rendering: Using spatial interpolation algorithms such as the inverse distance weighting method, discrete monitoring point data is interpolated into continuous field data, which is then rendered on the model surface as a color cloud map, intuitively displaying the spatial distribution patterns of settlement and displacement. Early warning status is highlighted through color changes (green-yellow-orange-red) and flashing effects on model components.

[0102] S400 uses an intelligent analysis and early warning engine combined with data updated by a digital twin model to perform trend prediction, coupled analysis, and risk assessment. If the dynamic threshold is exceeded, it will automatically generate early warning information of the corresponding level. This step is the "brain" of the platform, and its internal logic flow is as follows: 1. Single-indicator trend prediction (LSTM unit): Model Training: For each key monitoring point, at least 30 days of normal operating condition data are collected in advance as a training set, and supervised training is performed using an LSTM neural network. The network structure typically consists of two LSTM layers (64 / 128 neurons) and one fully connected output layer, with ReLU and Tanh activation functions used.

[0103] Online prediction: The engine loads a pre-trained model, takes in the data sequence of the most recent 72 hours, and makes rolling predictions of the changing trends over the next 24-72 hours. And simultaneously calculate the rate of change. .

[0104] 2. Multi-indicator coupling analysis and comprehensive risk assessment (XGBoost unit): Feature engineering: The model's input feature vector includes: a) real-time values ​​and rates of change of various monitoring indicators; b) future values ​​and prediction rates predicted by LSTM; c) current construction conditions (excavation depth, pit location, support removal status, etc.); d) environmental data (groundwater level, temperature); e) spatial correlation features of adjacent measuring point data.

[0105] Model Inference: Load the pre-trained XGBoost classification model, input the above feature vector, and the model outputs a comprehensive risk probability score between 0 and 1. This score comprehensively reflects the likelihood of harmful deformation of the structure under the current operating conditions.

[0106] 3. Dynamic threshold judgment and multi-level early warning decision-making: Threshold dynamic adjustment: The early warning judgment unit does not use a fixed threshold, but rather a dynamic threshold. The judgment is made, where α and β are empirical coefficients (usually set between 0.1 and 0.3), which allows the threshold to be automatically tightened or loosened according to the risk development trend.

[0107] Warning triggering logic: Yellow alert: When the monitored value exceeds the static control value 60%, or Three consecutive acquisition cycles exceeding the allowable rate, or The system triggers when the value exceeds 0.3. A notification is displayed on the web interface, and the on-site safety officer is notified to increase patrols.

[0108] Orange alert: When the monitored value exceeds 80% of the values ​​predicted by LSTM will exceed the limit within the next 48 hours, or The alert is triggered when the value exceeds 0.6. The system immediately sends an alert to the project manager and chief engineer via the app, suggesting a risk meeting and taking engineering measures such as adjusting the excavation sequence and increasing monitoring frequency.

[0109] Red Alert: When the monitored value has exceeded Or the prediction shows that the deformation will get out of control, or Triggered at >0.9. The system will forcibly push the highest level alert (including APP push, SMS and voice call) to all key responsible persons, and require the platform to immediately lock the construction permit process, forcibly stop work in dangerous areas, and activate the emergency plan.

[0110] S500: Construct an application layer for the early warning system, and combine early warning information to guide engineering decisions and emergency responses; Based on the early warning information, early warnings are issued through the web terminal and mobile APP of the intelligent early warning platform application layer to guide engineering decisions and emergency response; Once an alert is generated, the application layer immediately initiates a multi-channel release mechanism.

[0111] 1. Web-based visual terminal: such as Figure 3 As shown, in the platform's 3D interface, warning points and related components immediately change color and flash according to their severity level. A warning dialog box pops up in the center of the interface, listing the warning location, current value, predicted value, rate of change, risk probability, and handling suggestions. Simultaneously, the associated real-time video stream automatically pops up in the sidebar.

[0112] 2. Mobile App: Warning information will be sent to the responsible person's mobile phone via push notification. The notification must contain key information and require the recipient to click "Confirm Receipt". For orange and above warnings, the system will also send SMS messages; for red warnings, an automated voice call will be added to ensure that the information is delivered.

[0113] 3. Report generation: The platform automatically packages all relevant data (raw data, analysis process, model output, on-site video snapshots) of the early warning event into a structured "Early Warning Event Report" (PDF format), which is archived and available for download for post-event analysis and accountability.

[0114] After on-site personnel take appropriate measures based on the early warning information, they must use the "Responsibility Feedback" module of the mobile app to send the response status (e.g., "Excavation has been suspended," "XX cubic meters of grouting reinforcement has been completed") and on-site photos back to the platform. The platform will mark the response measures as a new "operating event" on the timeline and continuously track subsequent monitoring data changes to quantitatively assess the effectiveness of the measures. All of this data will be fed back to the intelligent analysis and early warning engine for incremental learning and adaptive optimization of the model, thus forming a continuously evolving intelligent management closed loop.

[0115] In summary, this invention achieves continuous, all-weather, fully automated, and blind-spot-free monitoring of the environment surrounding the foundation pit through intelligent sensors, replacing inefficient manual inspections. Through machine learning and trend prediction, it shifts from "post-event analysis" to "pre-event early warning," identifying potential risks and buying valuable time for intervention. By fusing multi-source information, it deeply integrates monitoring data with construction conditions and geological features, accurately locating risk sources and improving judgment accuracy. Through digital twin technology, it visually visualizes abstract data in a three-dimensional model, making risks "visible, manageable, and controllable," greatly improving management efficiency. Through a multi-level early warning mechanism and mobile push notifications assigning responsibility to individuals, it ensures that early warning information quickly reaches key responsible persons, forming a closed-loop management system of "monitoring-early warning-response-feedback-verification."

[0116] It is understood that the systems, devices, and storage media provided in the embodiments of the present invention correspond to the methods provided in the embodiments of the present invention, and the explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0117] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0118] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0119] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0120] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent early warning system for the safety of surrounding buildings during foundation pit construction, characterized in that, The system architecture comprises a perception layer, a transmission layer, a platform layer and an application layer connected in sequence, and is used for executing the following intelligent early warning process: S100, constructing a perception layer of the early warning system according to a foundation pit engineering design scheme and a surrounding environment risk assessment report; S200, constructing a transmission layer of the early warning system, and transmitting monitoring data collected by the perception layer to a platform layer of the early warning system; S300, constructing the platform layer of the early warning system to govern the monitoring data, and driving a digital twin model to update the data; S400, performing trend prediction, coupling analysis and risk research and judgment by an intelligent analysis and early warning engine in combination with the data updated by the digital twin model, and if the dynamic threshold is exceeded, automatically generating early warning information of a corresponding level; S500, constructing an application layer of the early warning system, and combining the early warning information to make engineering decisions and emergency responses.

2. The intelligent pre-warning system for safety of surrounding buildings in foundation pit construction according to claim 1, characterized in that, The perception layer comprises an intelligent sensor array arranged in a foundation pit, surrounding soil and target buildings; The sensor array arrangement covers the foundation pit body, surrounding soil and all target buildings to be protected; When the target buildings involve ancient buildings, a static level gauge with a higher accuracy than ±0.5mm is used to form a settlement monitoring network; A dual-axis tilt sensor is arranged at a structure corner and a middle part, with a range greater than ±5° and an accuracy higher than ±0.001°; A strain sensor is arranged on a key stress member, and a range is selected according to 1.5 times of a design axial force estimated value. 3.The intelligent pre-warning system for the safety of the surrounding buildings in the foundation pit construction according to claim 1, characterized in that, The transmission layer intelligently selects and combines 4G / 5G, LoRa and NB-IoT communication technologies according to data types, bandwidth requirements and field environments to form a heterogeneous network; Among them, the intelligent selection and combination of 4G / 5G, LoRa, NB-IoT multiple communication technology methods include: according to the priority of data packet , estimated transmission delay And link quality , dynamically select the best communication link; the selection decision function F select Is expressed as: wherein, is a link quality indicator of a specific technology, is its estimated transmission delay, and a, b, g are weight coefficients for balancing link quality, delay and traffic priority; Link quality assessment and adaptive coding: The transport layer continuously assesses the wireless channel quality, for narrowband IoT links, its maximum path loss , estimated according to the link budget formula: wherein is the transmit power, and are the transmit and receive antenna gains, respectively, is the receiver noise figure, is the required signal-to-noise ratio for demodulation, is the fading margin; According to the real-time calculated channel capacity , the modulation and coding scheme (MCS) is adaptively adjusted to approach the channel capacity, ensuring the reliability of data transmission: wherein is the channel bandwidth, is the received signal-to-noise ratio; Data packet success transmission probability: for a transmission using forward error correction (FEC) coding, the probability that a data packet is successfully received Modelled as: wherein, is the channel bit error rate, which is closely related to the signal-to-noise ratio SNR and the selected modulation mode. For BPSK modulation, in an additive white Gaussian noise channel, , is the data packet length; the transport layer decides whether to perform data fragmentation or retransmission by evaluating . Energy consumption model: energy consumed for one data transmission Simplify to: wherein, is the circuit power consumption, is the power amplifier efficiency coefficient, is the transmission distance, is the path loss exponent, is the transmission time, the transmission layer maximizes the overall endurance time of the sensor node by optimizing the transmission power and adopting a low-power sleep-wake mechanism.

4. The intelligent pre-warning system for safety of surrounding buildings in foundation pit construction according to claim 1, characterized in that, The platform layer is responsible for deep processing, model driving and intelligent analysis of monitoring data; The deep processing method of the monitoring data comprises: state prediction: measurement update: wherein, is a state estimate, is an error covariance matrix, is a state transition matrix, is a control input matrix, u is a control vector, is a process noise covariance matrix, is an observation matrix, is an observation noise covariance matrix, is a Kalman gain, is an actual observation. 5.The intelligent pre-warning system for safety of surrounding buildings in foundation pit construction according to claim 4, characterized in that, The intelligent analysis method of the platform layer of the monitoring data comprises: The LSTM neural network is used for trend prediction of key monitoring indexes in future periods. The LSTM processes time series data through its gating mechanism and cell state , and the core formula is as follows: forget gate: input gate: Candidate cell states: Cell status update: Output gate: Hidden state output: wherein, sigmoid is a sigmoid activation function, and are model parameters, is is an input at time t, is a hidden state at the previous time t-1, and the model outputs a prediction value through a fully connected layer ; The risk evaluation model is constructed by a machine learning ensemble algorithm, the risk evaluation model fuses a plurality of dimensional features such as real-time monitoring values X monitor , a predicted change rate v, construction working condition data X construction , and environmental data X environment , and outputs a comprehensive risk probability score , and the XGBoost model optimizes the following objective function through an additive training method: wherein, , is the number of leaf nodes, is the leaf weight, and is a regularization parameter; Output of the model Obtained by conversion through a sigmoid function; a multi-level early warning judgment unit: a preset yellow, orange and red three-level early warning mechanism; The judgment threshold is dynamically adjusted according to the prediction trend, change rate and coupling analysis result; Dynamic threshold is represented as: wherein, is a static control value, is a current rate of change, is a maximum allowed rate of change, is a combined risk probability score, and is a weighting factor, the warning trigger condition takes into account the percentage of the monitoring value that exceeds the static threshold, the time that the predicted value is out of range, and the risk interval in which it is located. 6.The intelligent pre-warning system for safety of surrounding buildings in foundation pit construction according to claim 5, characterized in that, The preset yellow, orange and red three-level early warning mechanism comprises: Yellow warning: when the monitoring value exceeds the static control value 60% of the static control value, The continuous 3 collection periods exceed the allowed rate, or > 0.3, the system gives a prompt on the Web interface and notifies the on-site safety officer to strengthen patrol; Orange alert: When the monitored value exceeds 80% of the values ​​predicted by LSTM will exceed the limit within the next 48 hours, or When the value exceeds 0.6, the system will immediately send an alarm to the project manager and chief engineer via the APP, suggesting that a risk meeting be held and measures such as adjusting the excavation sequence and increasing the monitoring of the project be taken. Red Alert: When the monitoring value has exceeded , the prediction shows that the deformation will be out of control or > 0.9, the system forces the highest level of alarm to all core persons in charge, and requires the platform to immediately lock the construction license process, forcibly stop the work in dangerous areas, and start the emergency plan.

Images