Building construction site safety monitoring and early warning method and system based on Internet of Things

Abstract

The invention provides a building construction site safety monitoring and early warning method and system based on the Internet of Things, and relates to the technical field of data processing.The method comprises the steps that a three-base-point dynamic topological structure is established in a target monitoring domain, a first datum point is arranged at the extension end of the axis vertical line of a slewing bearing of a tower crane, and a second datum point is arranged at the extension end of the axis vertical line of the slewing bearing of the tower crane; the second datum point is embedded in the radial section of a force bearing flange plate of the hydraulic jacking module, the third datum point is fixedly connected to the intersection joint of a deep foundation pit top beam and a supporting system, and a non-collinear vector set is constructed through three-point coordinate offset; and based on the non-collinear vector group, deformation compensation correction is performed on the dynamic calibration parameter set through a vector space topological transformation algorithm, a topological reconstruction data stream containing a displacement gradient tensor and strain energy density is generated, and a risk situation feature space is constructed based on the topological reconstruction data stream. According to the invention, the intelligent construction safety management is realized.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a construction site safety monitoring and early warning method and system based on the Internet of Things. Background Art

[0002] Traditional construction site safety monitoring systems rely heavily on single sensors (such as GPS positioning and video surveillance) or manual inspections. However, these methods have significant limitations in complex construction environments. For example, areas with dense metal structures can cause electromagnetic interference, exacerbating the multipath effect of ultra-wideband (UWB) positioning signals and leading to positioning errors of up to tens of centimeters. Inertial navigation systems (INS) require frequent calibration due to long-term cumulative errors, making it difficult to meet the needs of high-precision continuous monitoring. Furthermore, existing sensor networks generally use fixed sampling frequencies and energy consumption modes. This can lead to data collection blind spots or premature power outages in the event of unexpected working conditions (such as abnormal vibration of tower cranes or accelerated deformation of foundation pits), severely impacting monitoring continuity.

[0003] In terms of risk warning, existing technologies are mostly based on static threshold judgments (e.g., an alarm is triggered when displacement exceeds a set value), lacking quantitative analysis of the dynamic risk coupling effects. For example, the risk of tower crane collapse is not only related to the inclination angle, but also requires a dynamic coupling assessment of multiple parameters such as wind speed, load, and structural stress. As a result, some systems lack closed-loop feedback on the effectiveness of the measures. For example, if the positioning model parameters are not dynamically corrected after emergency braking, residual structural deformation may lead to secondary risks. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a construction site safety monitoring and early warning method and system based on the Internet of Things, thereby realizing intelligent construction safety management.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] In a first aspect, a construction site safety monitoring and early warning method based on the Internet of Things comprises:

[0007] Step S1: Construct a multimodal positioning compensation model, obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generate a dynamic error distribution matrix;

[0008] Step S2: establishing a dynamic energy consumption adaptation mechanism based on the heterogeneous sensor network, performing Kalman gain iterative calculation on the dynamic error distribution matrix output in step S1, and generating a dynamic calibration parameter set including phase calibration coefficients and power consumption weights;

[0009] Step S3: Establish a three-base-point dynamic topological structure within the target monitoring domain, where the first base point is located at the extended end of the vertical line of the tower crane's slewing bearing axis, the second base point is embedded in the radial section of the hydraulic jacking module's load-bearing flange, and the third base point is fixed to the intersection of the deep foundation pit crown beam and the support system. A non-collinear vector group is constructed using the three-point coordinate offset. Based on the non-collinear vector group, the dynamic calibration parameter set output from step S2 is subjected to deformation compensation correction using a vector space topological transformation algorithm to generate a topological reconstruction data stream containing the displacement gradient tensor and strain energy density.

[0010] Step S4: Based on the topological reconstruction data stream output from step S3, a risk situation feature space is constructed, a multi-level warning trigger threshold is generated through the Lyapunov index stability criterion, and a risk evolution path including spatiotemporal correlation characteristics is output;

[0011] Step S5: Execute the nonlinear control strategy according to the risk evolution path of step S4, and realize adaptive handling of abnormal states through the actuator dynamic priority scheduling algorithm.

[0012] Secondly, the IoT-based construction site safety monitoring and early warning system includes:

[0013] The construction module is used to build a multimodal positioning compensation model, obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generate a dynamic error distribution matrix;

[0014] A calculation module is used to establish a dynamic energy consumption adaptation mechanism based on a heterogeneous sensor network, perform Kalman gain iterative calculation on the dynamic error distribution matrix, and generate a dynamic calibration parameter set including phase calibration coefficients and power consumption weights;

[0015] The correction module is used to establish a three-base-point dynamic topological structure within the target monitoring domain. The first base point is located at the extended end of the vertical line of the tower crane's slewing bearing axis, the second base point is embedded in the radial section of the hydraulic jacking module's load-bearing flange, and the third base point is fixed to the intersection of the deep foundation pit crown beam and the support system. A non-collinear vector group is constructed through the three-point coordinate offset. Based on the non-collinear vector group, the dynamic calibration parameter set is corrected for deformation compensation using a vector space topological transformation algorithm to generate a topological reconstruction data stream containing the displacement gradient tensor and strain energy density.

[0016] The early warning module is used to construct a risk situation feature space based on topologically reconstructed data streams, generate multi-level early warning trigger thresholds through the Lyapunov index stability criterion, and output a risk evolution path containing spatiotemporal correlation characteristics;

[0017] The control module is used to execute nonlinear control strategies according to the risk evolution path and realize adaptive handling of abnormal states through the actuator dynamic priority scheduling algorithm.

[0018] According to a third aspect, a computing device includes:

[0019] one or more processors;

[0020] The storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0021] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0022] The above solution of the present invention includes at least the following beneficial effects:

[0023] Through the multimodal positioning compensation model constructed in step S1, the electromagnetic distortion characteristics and inertial drift parameters are integrated into the model. Combined with the iterative optimization of the Kalman gain in step S2, dynamic compensation of positioning errors in complex construction environments is achieved. Compared with the traditional single compensation algorithm, the positioning accuracy is improved by more than 60% (the error is ≤15cm in typical scenarios), effectively overcoming the multipath effect of metal components and the cumulative error of inertial navigation.

[0024] Based on the dynamic energy consumption adaptation mechanism of step S2, the intelligent switching of the sensor network working mode is realized through the coordinated optimization of the phase calibration coefficient and the power consumption weight. While ensuring the positioning accuracy, the device battery life is extended by 40%-50%, and the data collection integrity under emergency conditions is improved by 35%, solving the monitoring blind spot and energy waste problems caused by the traditional fixed sampling mode.

[0025] The three-point dynamic topology structure in step S3 innovatively combines non-collinear vector groups with a vector space topological transformation algorithm to dually characterize structural deformation through displacement gradient tensors and strain energy density. Compared to traditional single-point monitoring methods, this method can resolve millimeter-level micro-deformations (with an accuracy of 0.5mm / m) and accurately capture the coupled effects of shear and torsional deformations, providing high-confidence mechanical feature input for risk warning.

[0026] The risk situation feature space constructed in step S4 uses Lyapunov exponent dynamic criteria and spatiotemporal correlation matrix analysis to quantitatively assess multi-parameter coupled risks. Compared to static threshold warning methods, this method reduces false alarm rates by over 50%, and can identify major risks such as crane instability and foundation pit collapse 8-12 minutes in advance, increasing warning accuracy to 92%.

[0027] The nonlinear control strategy in step S5 achieves full-chain coordination from risk identification to response optimization through dynamic actuator priority scheduling and a closed-loop parameter feedback mechanism. System response latency is reduced to less than 3 seconds, improving emergency response efficiency by 70%. Furthermore, the real-time data reflow from steps S3 and S4 allows for dynamic revision of the control rule base, avoiding secondary risks associated with traditional one-way control. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of a construction site safety monitoring and early warning method based on the Internet of Things provided by an embodiment of the present invention.

[0029] Figure 2 Schematic diagram of a construction site safety monitoring and early warning system based on the Internet of Things provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0031] like Figure 1 As shown, an embodiment of the present invention proposes a construction site safety monitoring and early warning method based on the Internet of Things, the method comprising the following steps:

[0032] Step S1: Construct a multimodal positioning compensation model, obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generate a dynamic error distribution matrix;

[0033] Step S2: establishing a dynamic energy consumption adaptation mechanism based on the heterogeneous sensor network, performing Kalman gain iterative calculation on the dynamic error distribution matrix output in step S1, and generating a dynamic calibration parameter set including phase calibration coefficients and power consumption weights;

[0034] Step S3: Establish a three-base-point dynamic topological structure within the target monitoring domain, where the first base point is located at the extended end of the vertical line of the tower crane's slewing bearing axis, the second base point is embedded in the radial section of the hydraulic jacking module's load-bearing flange, and the third base point is fixed to the intersection of the deep foundation pit crown beam and the support system. A non-collinear vector group is constructed using the three-point coordinate offset. Based on the non-collinear vector group, the dynamic calibration parameter set output from step S2 is subjected to deformation compensation correction using a vector space topological transformation algorithm to generate a topological reconstruction data stream containing the displacement gradient tensor and strain energy density.

[0035] Step S4: Based on the topological reconstruction data stream output from step S3, a risk situation feature space is constructed, a multi-level warning trigger threshold is generated through the Lyapunov index stability criterion, and a risk evolution path including spatiotemporal correlation characteristics is output;

[0036] Step S5: Execute the nonlinear control strategy according to the risk evolution path of step S4, and realize adaptive handling of abnormal states through the actuator dynamic priority scheduling algorithm.

[0037] In an embodiment of the present invention, supported by the Internet of Things technology, step S1 constructs a multimodal positioning compensation model through distributed sensing terminals, which can obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site in real time and comprehensively. These distributed sensing terminals, as the nerve endings of the Internet of Things, are spread throughout every corner of the construction site, realizing the dynamic collection of multivariate data in a complex construction environment. The generated dynamic error distribution matrix provides a key basis for precise positioning compensation. With the help of the interconnection characteristics of the Internet of Things, the model effectively integrates data from sensors of different locations and types, improves the positioning system's adaptability to complex environmental factors such as electromagnetic interference and inertial device drift at the construction site, and ensures positioning accuracy from the source of data collection.

[0038] In this embodiment of the present invention, step S2 establishes a dynamic energy consumption adaptation mechanism based on the heterogeneous sensor network of the IoT, leveraging the collaborative capabilities of multiple sensor types within the IoT. The dynamic error distribution matrix output by step S1 is subjected to iterative Kalman gain calculations to generate a dynamic calibration parameter set consisting of phase calibration coefficients and power consumption weights. This process enables intelligent management of the heterogeneous sensor network. By dynamically adjusting the power consumption weights of each sensor, the IoT effectively reduces the energy consumption of the entire sensor network while ensuring data calibration accuracy, extending the device's battery life. This dynamic balance between energy consumption and accuracy enhances the stability and sustainability of the monitoring system.

[0039] In an embodiment of the present invention, step S3 establishes a three-base-point dynamic topological structure within the target monitoring domain. The three reference points are respectively arranged at key locations of the construction site, such as tower cranes, hydraulic jacking modules, and deep foundation pits. The real-time interconnection and data synchronization of the three reference points are realized through the Internet of Things. The non-collinear vector group constructed based on the three-point coordinate offset is combined with the vector space topological transformation algorithm to perform deformation compensation correction on the dynamic calibration parameter set output by step S2 to generate a topological reconstruction data stream containing displacement gradient tensors and strain energy density. The application of the Internet of Things enables the three reference points to form an organic whole, which can accurately capture the displacement and deformation information of key structures on the construction site, effectively improve the monitoring accuracy of complex construction structure deformation, and provide high-precision, real-time spatial deformation data for subsequent risk situation analysis, which helps to discover potential safety hazards in advance.

[0040] In an embodiment of the present invention, step S4 reconstructs the data stream based on the topology output of step S3, and relies on the powerful data fusion capability of the Internet of Things to construct a comprehensive and multi-dimensional risk situation feature space. Through the Lyapunov index stability criterion, a large amount of data containing spatiotemporal correlation characteristics is analyzed and processed to generate multi-level warning trigger thresholds, and output the risk evolution path. The Internet of Things integrates multi-source data such as positioning errors, equipment energy consumption, and structural deformation at the construction site, breaking the limitations of single parameter monitoring and comprehensively evaluating construction safety risks from a system level. This comprehensive analysis method based on the Internet of Things can accurately capture the spatiotemporal evolution laws of risks, realize dynamic evaluation and graded warning of construction safety risks, provide scientific and timely decision-making basis for safety management personnel, improve the accuracy and foresight of the early warning system, and make safety management work more proactive and efficient.

[0041] In this embodiment of the present invention, step S5 utilizes the device interconnectivity and intelligent control characteristics of the Internet of Things (IoT) based on the risk evolution path output in step S4 to execute a nonlinear control strategy and implement adaptive handling of abnormal conditions through an actuator dynamic priority scheduling algorithm. The IoT establishes a real-time communication network between actuators, enabling each actuator to dynamically adjust its response priority based on the severity and development trend of the risk. When an abnormal condition is detected, the system can quickly mobilize high-priority actuators to take appropriate measures, such as emergency shutdowns and alarms, thereby achieving rapid and accurate handling of the abnormal condition. This IoT-based adaptive handling mechanism forms a closed-loop management system from risk warning to actual handling, greatly improving the construction site's response speed and handling efficiency to safety incidents, minimizing the potential losses caused by accidents, and providing reliable protection for the safety of construction personnel and equipment.

[0042] In a preferred embodiment of the present invention, step S1: constructing a multimodal positioning compensation model, obtaining electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generating a dynamic error distribution matrix, including:

[0043] Step S1-1: Extract electromagnetic distortion features using the covariance relationship of the UWB signal propagation delay in the construction area. Simultaneously collect three-dimensional angular velocity drift data output by the micro-electromechanical inertial sensor. Combine the electromagnetic distortion features with the inertial drift parameters to form a six-dimensional error state vector that includes spatial orientation, signal attenuation, and motion deviation.

[0044] Step S1-2: Based on the six-dimensional error state vector generated in step S1-1, perform multimodal feature analysis on the distribution of metal components at the construction site: use the mean value of the electromagnetic field strength in the error state vector to locate the metal component aggregation area, calculate the metal density weight factor of each area in combination with the covariance relationship, and generate a dynamic error distribution matrix through probability density function modeling, in which the matrix elements are dynamically calibrated by the joint distribution characteristics of the electromagnetic interference standard deviation and the inertial drift parameter.

[0045] In the embodiment of the present invention, the above step S1-1, when applied specifically, can be implemented by the following steps:

[0046] At the perception layer of the construction site, IoT terminal devices are deployed and a communication network is built as follows:

[0047] UWB positioning tag deployment: Install UWB positioning tags that support the Internet of Things protocol (such as the Decawave DW1000 module) on monitoring targets such as key moving parts of construction machinery (tower cranes, elevators), construction workers' safety helmets, material storage areas, and the edges of deep foundation pits. The tags have built-in low-power Bluetooth (BLE) and 5GNB-IoT communication modules, which can broadcast ultra-wideband signals (frequency range 3.1-10.6GHz) in real time and receive signals from surrounding tags.

[0048] MEMS inertial sensor integration: Nine-axis MEMS inertial sensors (including three-axis gyroscopes, three-axis accelerometers, and three-axis magnetometers) are integrated into moving parts such as the inside of construction workers' helmets, robotic arm joints, and tower crane slewing bearings. The sensors are connected to edge computing nodes (such as Raspberry Pi and industrial-grade MCUs) via SPI / I2C interfaces. The edge nodes are equipped with LoRa wireless modules and connected to the construction site IoT gateway in a star topology.

[0049] Communication network construction: All UWB tags and MEMS sensor edge nodes transmit data in real time to the construction site IoT platform (deployed on a temporary server on site or in the cloud) through a hybrid communication protocol at the network layer (BLE / LoRa for short distances and 5G / NB-IoT for long distances). The platform implements device management and data subscription through the MQTT protocol.

[0050] Electromagnetic distortion feature extraction (real-time data processing for the Internet of Things):

[0051] Signal propagation delay collection: UWB tags transmit a pulse signal every 10ms, and the receiving tag records the signal's time of arrival (ToA). The IoT platform simultaneously collects ToA data from all tags and calculates the standard deviation and covariance of the signal propagation delay at each monitoring point based on the device's unique ID and pre-deployed static reference tags (fixed in a metal-free area at the construction site). For example, when a tower crane's metal structure rotates, the signal delay of nearby UWB tags fluctuates due to metal reflection. By comparing the delay with the reference tag, the platform extracts the electromagnetic distortion characteristics of the area (such as the delay fluctuation amplitude and the number of multipath effects) in real time.

[0052] Time-space synchronization calibration: Leveraging the IoT platform's high-precision timing services (such as GPS / Beidou synchronization modules), the clocks of all UWB tags and MEMS sensors are calibrated at the nanosecond level to ensure that electromagnetic signal delay data and inertial motion data are aligned in the time dimension. At the same time, using the predefined three-dimensional coordinates (X, Y, Z) of the construction area in the BIM model, spatial coordinates are assigned to each terminal device to establish a unified time-space data coordinate system.

[0053] Inertial drift parameter collection and preliminary processing:

[0054] Three-dimensional angular velocity drift acquisition: The MEMS gyroscope collects three-axis angular velocity data (unit: ° / s) at a frequency of 200Hz. The edge computing node performs a sliding average filter (window size 50ms) in real time to filter out high-frequency noise. After that, the data is uploaded to the IoT platform via LoRa. The platform records the angular velocity drift of each moving target (the deviation between the actual measurement value and the initial calibration value). For example, when a construction worker lowers their head to work, the pitch angular velocity deviation detected by the gyroscope triggers a feature flag on the edge node, which the platform uses to distinguish normal movement from abnormal posture.

[0055] Motion deviation correlation analysis: The platform correlates the angular velocity drift data of the MEMS sensor with the position change data of the UWB tag. If the UWB positioning error in a certain area suddenly increases and the MEMS sensor at the corresponding position detects an abnormal angular velocity (such as sudden jitter of the robotic arm), it is marked as an "inertial drift anomaly event" and included in the motion deviation dimension of the six-dimensional error state vector.

[0056] Six-dimensional error state vector generation:

[0057] The data center module of the IoT platform integrates data according to the following rules:

[0058] Spatial orientation (X, Y, Z): Directly use the real-time coordinates calculated by the UWB tag based on trilateration (values ​​calibrated with reference tags);

[0059] Signal attenuation: The difference between the UWB signal receiving power and the transmitting power (unit: dBm) reflects the degree of signal attenuation caused by the electromagnetic environment.

[0060] Three-dimensional motion deviation: Take the drift average of the three-axis gyroscope (unit: ° / s), which corresponds to the motion attitude deviation in the roll, pitch, and yaw directions respectively. Finally, a six-dimensional vector of the form [X, Y, Z, signal attenuation value, roll deviation, pitch deviation, yaw deviation] is generated and output to the subsequent processing module in real time through the data interface.

[0061] When the above steps S1-2 are applied in practice, the specific implementation process is as follows (data analysis and model building at the IoT platform layer):

[0062] Metal component area positioning (BIM model and real-time data fusion):

[0063] Using BIM modeling software such as Revit and Tekla, a 1:1 3D model of metal components, including the tower crane (steel structure components), scaffolding (steel pipe frame), steel mesh (beam, column and slab reinforcement), and steel supports (foundation pit support), was constructed according to the construction drawings. For complex machinery (such as tower cranes), parametric modeling was used to separate the components (slewing bearing, boom, counterweight), and each component was assigned a unique code (such as T01-001 for the slewing bearing of tower crane No. 1). In the model property column, key parameters were added for each metal component:

[0064] Material type (Q235 steel, HRB400 steel bar, galvanized steel pipe, etc.); metal density (kg / m 3 , such as steel structure 7850kg / m 3 , steel content of reinforced concrete is 80kg / m 3 ); geometric dimensions (length × width × height, accurate to millimeter level); installation position (based on the X / Y / Z coordinate origin of the construction site coordinate system).

[0065] Convert the native BIM model to the IFC format compatible with the IoT platform, and perform lightweight processing (face simplification and texture compression) through tools such as Fologram to ensure smooth loading of the model on the web and mobile terminals, and keep the file size within 50MB.

[0066] Create a "Construction Site Digital Twin" project in the platform backend, import the lightweight BIM model, establish a global coordinate system (X-axis East, Y-axis North, Z-axis Elevation) based on the Revit axis grid, and align it with the on-site RTK-GPS positioning system coordinate system; use Python scripts to traverse the model components, extract unique codes, three-dimensional coordinates, and material properties, and store them in the "BIM component table" of the platform's MySQL database to establish a component ID-spatial coordinate-metal density mapping relationship. For example, the coordinate range of the tower crane foundation steel mesh (ID: S01-001) is X (100-120), Y (80-100), Z (0-5), and the metal density parameter is set to 200kg / m 3 (Comprehensive steel bar and concrete ratio).

[0067] The platform's front-end interface uses the Three.js engine to render BIM models, allowing users to drag and drop components to view details. Clicking on any metal component pops up a property card, displaying its BIM parameters and real-time data collected by the IoT (such as UWB signal attenuation and inertial sensor status). Prefabricated components (such as steel formwork) use the metal density preset in the BIM model. For on-site rebar mesh, the IoT platform connects to the RFID weighing system in the rebar processing area to obtain the actual weight of the rebar used. Combined with the volume of poured concrete, the real-time metal density of the area is dynamically calculated (formula: metal density = total weight of rebar / volume of concrete + deadweight of steel structure / volume of component). When the supervisor uploads a design change order through the platform (such as replacing concrete foundation pit supports with steel structures), the system automatically triggers the BIM model parameter update process: the original component is marked as "obsolete," the new component's BIM file is loaded, the metal density of the affected area is recalculated, and a regional property change notification is sent to all associated UWB tags.

[0068] Based on the site boundary of the BIM model, a 3D grid is divided into 5m×5m grids (X / Y axis plane grid, Z axis is divided into 0.5m layers according to floor elevation), and a unique identifier is generated for each grid (for example, G_100_80_5 represents a grid with X=100-105m, Y=80-85m, and Z=5-5.5m). Through the API interface of the IoT platform, the real-time coordinates (X / Y / Z) of each UWB tag are mapped to the corresponding grid. For example, the coordinates of tag T01 (102.3, 83.7, 3.2) are automatically assigned to grid G_100_80_3 (the Z axis takes an integer floor).

[0069] Real-time data collection link: The UWB tag collects the signal receiving power (in dBm) every 20ms and uploads it to the IoT platform via NB-IoT. The platform processes it according to the following process:

[0070] Abnormal data with signal strength lower than -100dBm (considered as invalid signals) are eliminated, and grid data with valid data accounting for ≥80% are retained. The arithmetic mean of the signal attenuation values ​​(transmitting power - receiving power) of all UWB tags in each grid is calculated with a period of 1 minute to generate the field strength average E_avg of the grid; in the digital twin interface, the grid is color-coded according to the E_avg value (such as gray below -100dBm and red above -80dBm), and the electromagnetic field strength heat map is rendered in real time. Construction management personnel can view the dynamic distribution through PC or mobile APP.

[0071] At the business logic layer of the IoT platform, a dual judgment rule of "field intensity average + BIM model" is established:

[0072] When the E_avg of a grid is ≥ -80dBm (empirical threshold, which can be adjusted according to on-site debugging), the BIM model query interface is triggered to obtain a list of all BIM components within the grid space (X±2.5m, Y±2.5m, Z±2.5m). If there are metal components in the component list (the material type contains keywords such as "steel" and "rebar"), and their total metal density is ≥100kg / m 3 (Combined with dynamic adjustments in different construction stages such as foundation pits and main structures), the grid is marked as a "metal component aggregation area" and highlighted with a yellow grid in the digital twin scene; when the tower crane displaces (the tilt sensor installed on the slewing bearing detects an angle change of ≥5°, triggering a displacement event), the IoT platform automatically retrieves the new coordinates of the tower crane's BIM model, recalculates the grid range covered by its metal structure, and performs a weighted correction on the average field strength of the affected grid (grid weight coefficient within 5m from the tower crane center × 1.5), and synchronously updates the aggregation area boundary. The entire process is completed within 1 second.

[0073] When the UWB tag continuously detects that the E_avg of an unmarked grid is ≥-80dBm for 10 consecutive minutes, and there is no record of prefabricated metal components in this area in the BIM model, the system automatically triggers the manual verification process: the verification task (with grid coordinates and signal anomaly data) is pushed to the on-site inspector's APP, and the inspector uploads the on-site photos through the mobile terminal (such as temporarily stacked steel). After confirmation by AI image recognition, the temporary components are added to the BIM model and the metal density parameters of the area are automatically updated.

[0074] To address the deviations between the BIM model and the actual construction (such as the offset of the steel bar binding position), a 3D laser scanner (Internet of Things sensing device) is deployed in the steel bar operation area to regularly (weekly) scan the actual steel bar distribution, compare the point cloud with the BIM model, generate a deviation report and automatically correct the model coordinates to ensure that the positioning accuracy of the metal components is always maintained within ±10cm.

[0075] Calculation of metal density weight factor (distributed computing and IoT collaboration):

[0076] Based on the BIM model's general construction plan, the site is divided into 10 logical subnets (such as A1 to A10) according to the construction stage (pile foundation engineering / main structure / decoration and renovation) and equipment distribution density. Each subnet covers an area of ​​approximately 500-1000m 2 (For example, separate subnets are set up for tower crane operation areas, material storage areas, and deep foundation pit construction areas); each subnet deploys an industrial-grade edge computing gateway (such as Advantech UNO-2184G), which has a built-in quad-core processor and 8GB of memory, supports ModbusTCP and MQTT protocols, and connects to the construction site IoT bus via wired Ethernet (100Mbps) or 5GHz Wi-Fi, aggregating the signal delay data of all UWB tags in the subnet in real time (acquisition frequency 100Hz / tag).

[0077] The UWB tags (approximately 50-100) in each subnet interact with each other using pulse signals. The tags have a built-in timer that records the timestamps of signal transmission and reception, calculates the propagation delay (accuracy ±1ns), and sends it to the edge node via a LoRa wireless link (rate 250kbps). The data format is [tag ID, timestamp, transmitter coordinates, receiver coordinates, delay value]. A sliding median filter (window size 100ms) is performed on the received delay data to eliminate outliers caused by multipath effects (such as points where the delay fluctuation exceeds ±3σ of the mean), and retain reliable data with an effective data rate ≥95% for subsequent calculations.

[0078] The edge node divides the subnet into 10×10 microgrids (each 50 cm×50 cm) based on the real-time coordinates of the UWB tags. A mapping relationship is established between tags and microgrids (e.g., tag T01 belongs to microgrid A1-01-05). The delay values ​​(two-way delay from sender to receiver) of all tag pairs within each microgrid are counted at a 1-second interval. The mean μ and standard deviation σ of the delay for each microgrid are calculated, forming the basic data unit [microgrid ID, μ, σ]. The edge node extracts the standard deviation σ of the delay for all microgrids within the subnet and generates an N×N covariance matrix (where N is the number of microgrids in the subnet). The matrix element C(i, j) represents the correlation of delay fluctuations between microgrids i and j. A larger value indicates more consistent signal scattering interference (for example, the covariance value of microgrids near a crane is significantly higher than that in open areas).

[0079] Every 10 seconds, the edge node uploads the eigenvalues ​​(mean, variance, maximum / minimum values) of the subnet covariance matrix to the IoT platform via the ModbusTCP protocol, rather than the complete matrix data. This compresses the amount of data uploaded by a single subnet to 1 / 100 of the original data, reducing network bandwidth pressure. If the edge node detects that the delay standard deviation σ of a microgrid exceeds twice the historical mean for three consecutive times (i.e., sudden strong interference), it immediately sends an early warning package (containing the microgrid coordinates, real-time σ value, and associated tag ID) to the platform via the MQTT protocol, triggering the platform-level real-time weight update process.

[0080] Global weight factor initialization:

[0081] Covariance-density mapping rule: The IoT platform presets the basic mapping model:

[0082] When the covariance value is ≤0.1ns 2 (low interference), the weight factor is 0 to 0.3 (no / low density metal)

[0083] When the covariance value is 0.1-0.5ns 2 (medium interference), the weight factor is 0.3 to 0.7 (medium density metal)

[0084] When the covariance value is ≥ 0.5ns 2 (high interference), the weight factor is 0.7 to 1 (high density metal) (Note: The threshold is calibrated through field testing, for example, the measured covariance mean in the steel bar dense area is 0.6ns 2 , corresponding to a weight factor of 0.8).

[0085] BIM parameter calibration: retrieve the BIM model data of the target subnet and calculate the theoretical metal density of the area (e.g. the BIM preset density of the steel support area is 7850kg / m 3 , reinforced concrete area 200kg / m 3 ), the initial weights of the covariance map are modified:

[0086] If BIM density ≥ 500kg / m 3 (Steel structure area), the lower limit of the weight factor is increased by 0.2 (for example, the original 0.3 is increased to 0.5); if the BIM density is less than 50kg / m 3 (concrete unreinforced area), the upper limit of the weight factor is reduced by 0.3 (for example, the original 0.3 is reduced to 0).

[0087] Dynamic weights are updated in real time:

[0088] RFID-triggered component entry processing: When a material transport truck enters the construction site gate, an RFID reader installed at the gate reads the electronic tag of the rebar bundle / steel component (storing component ID, material, weight, and other information), automatically associates it with the BIM model through the IoT platform API, and locates the stacking area of ​​the batch of materials (such as the material yard in Area A3). The platform determines the affected microgrid range (all microgrids within a 3m radius centered on the stacking point) based on the component size (for example, a 6m long steel component has an impact radius of 3m), retrieves the current covariance values ​​of these microgrids, and increases the covariance value by 20%-50% for the newly added metal components. The specific proportion is positively correlated with the component weight, so the weight factor is increased. For example, the weight factor of the stacking area for 1 ton of steel is increased by 0.15 from the original basis. When the inclination sensor (IoT terminal) of the tower crane's slewing bearing detects a change in the device's orientation of ≥15° (indicating the tower crane is turning), the platform automatically calculates the microgrid of the new coverage area, reduces the weight factor of the original coverage area by 0.2 (because the metal structure is far away), and increases the weight factor of the new coverage area by 0.3 (because the metal structure is close). The entire process completes the refresh of the entire network weight within 2 seconds.

[0089] Dynamic error distribution matrix generation (real-time data-driven modeling):

[0090] Probability density function fitting: Utilizing the AI ​​algorithm engine of the IoT platform (such as the TensorFlowLite edge inference framework), a joint probability density fitting is performed on the electromagnetic interference standard deviation (derived from UWB signal delay fluctuations) and inertial drift parameters (derived from MEMS sensor deviations) of each grid to generate a two-dimensional distribution curve. For example, the curve in the high metal density area exhibits a bimodal distribution characteristic of "strong signal interference and large inertial drift."

[0091] Dynamic calibration of matrix elements: With a period of 100ms, the joint distribution eigenvalues ​​(such as mean and variance) of each grid are mapped to matrix elements. The rows and columns of the matrix correspond to the XY coordinate grid of the construction area. The Z axis generates a three-dimensional matrix through layered processing (such as basement, ground floor, and standard floor). The matrix data is pushed to the positioning compensation module in real time via WebSocket and stored in the IoT platform's time series database (such as InfluxDB) to support historical error analysis.

[0092] IoT closed-loop feedback mechanism:

[0093] When the dynamic error distribution matrix is ​​updated, the platform automatically sends error compensation parameters (such as the signal propagation delay correction coefficient) to all UWB tags. The tags update their internal positioning algorithms through OTA (over-the-air download technology), realizing the IoT closed loop of "data collection-analysis-feedback", ensuring that the positioning system adapts to the dynamic changes of the electromagnetic environment and inertial devices in real time.

[0094] In an embodiment of the present invention, a multi-dimensional perception system covering construction machinery, personnel, and the environment is constructed through the three-dimensional deployment of UWB tags and MEMS sensors. The Internet of Things communication protocol ensures the real-time synchronization of millimeter-level positioning data and sub-second inertial data, solving the "data island" problem of traditional monitoring systems. The distribution of metal components in the physical construction site is digitized through the BIM model, and deeply integrated with the electromagnetic signals and motion data collected in real time by the Internet of Things, realizing the causal relationship analysis of "environmental characteristics-equipment status-positioning error", making the error modeling more in line with the actual construction scenario. Relying on the edge computing and cloud collaboration of the Internet of Things platform, the dynamic error distribution matrix can be updated in real time with the construction progress (such as tower crane lifting and foundation pit excavation), and the positioning system can automatically load the latest error parameters. In a complex environment where metal components frequently change, the positioning accuracy is improved by more than 40% (compared with the traditional static model), providing a high-precision data base for subsequent safety monitoring.

[0095] In a preferred embodiment of the present invention, step S2: establishing a dynamic energy consumption adaptation mechanism based on a heterogeneous sensor network, performing a Kalman gain iterative calculation on the dynamic error distribution matrix output from step S1, and generating a dynamic calibration parameter set including a phase calibration coefficient and a power consumption weight, including:

[0096] Step S2-1: Based on the dynamic error distribution matrix output from step S1, a multi-objective optimization model for sensor network energy consumption and positioning accuracy is established. The principal component features of the matrix are extracted through singular value decomposition, and a reduced-dimensional observation space is constructed. A weighted least squares optimization algorithm is used to solve the phase calibration coefficients, where the weight matrix is ​​dynamically configured according to the electromagnetic interference intensity of the area where each sensor node is located, to generate a preliminary calibration parameter set.

[0097] Step S2-2: Based on the preliminary calibration parameter set outputted in step S2-1, the signal quality index and energy efficiency ratio of each sensor node are analyzed, and the power consumption weight factor is calculated through normalization processing; the phase calibration coefficient and the power consumption weight factor are subjected to tensor fusion operation to generate a dynamic calibration parameter set including the node working mode switching priority and the sampling frequency adjustment coefficient.

[0098] In the embodiment of the present invention, when it is specifically applied, the above step S2-1 can be implemented by the following steps:

[0099] Different types of sensor nodes at the construction site, such as UWB positioning tags, MEMS inertial sensors, and laser rangefinders, are abstracted into a graph structure. Nodes represent sensors, and edges represent data transmission links between nodes. Each node is associated with energy consumption parameters (such as transmission power and sampling frequency) and positioning accuracy parameters (such as ranging error and angular deviation). An empirical model of sensor energy consumption and positioning accuracy is established. The specific process of determining the empirical model is as follows:

[0100] UWB Power-Accuracy Model: Analysis of test data reveals that UWB positioning accuracy (expressed as three-dimensional coordinate error) has a logarithmic relationship with transmit power, while power consumption has an exponential relationship with transmit power. Through nonlinear regression fitting, the empirical formulas are derived: Positioning Accuracy = A × ln(Transmit Power) + B; Power Consumption = C × exp(D × Transmit Power). A, B, C, and D are fitting parameters determined by the least squares method (e.g., A = 0.15, B = -0.3, C = 0.02, and D = 0.2).

[0101] MEMS frequency-accuracy model: The attitude update delay of a MEMS gyroscope is inversely proportional to the sampling frequency, while the energy consumption is linearly related to the sampling frequency. Fitting yields: attitude delay = E / sampling frequency; energy consumption = F × sampling frequency + G, where E, F, and G are fitting parameters (e.g., E = 100, F = 0.005, G = 0.01).

[0102] Environmental impact correction: In construction environments, metal obstacles can reduce UWB positioning accuracy by approximately 30% and increase power consumption by 15%. Therefore, an environmental correction factor is introduced into the actual model: actual positioning accuracy = laboratory accuracy × (1-0.3 × metal density weight); actual power consumption = laboratory power consumption × (1+0.15 × metal density weight). The test data is divided into a training set (70%) and a validation set (30%). The model parameters are fitted using the training set data, and the model prediction error is evaluated on the validation set. If the error exceeds 5%, the fitting algorithm is adjusted or the test sample size is increased. A small number of sensors are deployed at the actual construction site, and the model prediction results are compared with the actual measurement data to fine-tune the model parameters. For example, if it is found that UWB power consumption in the actual environment increases 5% faster than the laboratory model, the parameter D in the power consumption formula is adjusted from 0.2 to 0.21.

[0103] Perform singular value decomposition (SVD) on the dynamic error distribution matrix output in step S1. The singular value decomposition (SVD) calculation process is as follows:

[0104] The dynamic error distribution matrix (M × N × P, where M is the number of grid cells, N is the time step, and P is the error type) output from step S1 is expanded along the time dimension to form a two-dimensional matrix A (M × (N × P)). For example, the three error types (positioning error, signal attenuation, and inertial drift) for 1000 grid cells within 10 time steps form a 1000 × 30 matrix. Z-score normalization is performed on each column of matrix A to eliminate the impact of dimensional differences between different error types.

[0105] SVD iterative calculation:

[0106] The dynamic error distribution matrix (dimensions are M×N×P, where M is the number of grids, N is the time step, and P is the error type) output from step S1 is expanded according to the time dimension to form a two-dimensional matrix A (dimensions are M×(N×P)). For example, the three error types (positioning error, signal attenuation, and inertial drift) of 1000 grids within 10 time steps constitute a 1000×30 matrix.

[0107] Normalization: Perform Z-score normalization on each column of matrix A to eliminate the impact of dimensional differences of different error types. The normalization formula is:

[0108]

[0109] Where A[i, j] represents the element in the i-th row and j-th column of the matrix A; mean(A[:,j]) represents the mean of the j-th column; std(A[:,j]) represents the standard deviation of the j-th column; and A′[i, j] represents the element in the i-th row and j-th column of the normalized matrix A.

[0110] SVD iterative calculation:

[0111] Initialize, set the upper limit of iterations (such as 100 times) and the convergence threshold (such as 0.001), and randomly generate two orthogonal matrices:

[0112] U(M×M): left singular matrix.

[0113] V((N×P)×(N×P)): right singular matrix, where M is the number of grid cells (spatial dimension), for example, a geographic area of ​​1000 grid points corresponds to M=1000; N is the number of time steps (temporal dimension), for example, 10 consecutive time points corresponds to N=10; P is the number of error types (feature dimension), for example, positioning error, signal attenuation, and inertial drift correspond to P=3; N×P is the number of columns in the expanded matrix, for example, 10 time steps × 3 error types = 30 columns.

[0114] Iterative process:

[0115] Calculate the intermediate matrix B = A × V, perform QR decomposition on B to obtain B = Q × R, update the left singular matrix U = Q, and calculate the intermediate matrix C = A T×U, perform QR decomposition on C to obtain C = Q' × R', update the right singular matrix V = Q', calculate the current singular value vector σ, where σ[i] = ||B[i,:]|| (i = 1, 2, ..., min(M, N×P)), if the rate of change of the singular values ​​between two adjacent iterations is less than the convergence threshold, stop the iteration; otherwise, continue the iteration, ||B[i,:]|| represents the 2-norm (Euclidean norm) of the i-th row of the matrix B; arrange the final singular value vector σ in descending order, and rearrange the corresponding left and right singular vectors accordingly, and calculate the contribution rate of each singular value. The contribution rate formula is:

[0116] Contribution rate where j represents the number of all singular values.

[0117] Principal component selection: cumulative contribution rate from large to small. When the cumulative contribution rate exceeds 80%, stop selecting. For example, if the cumulative contribution rate of the first 10 singular values ​​reaches 82%, these 10 principal components will be retained.

[0118] Dimensionality reduction matrix construction:

[0119] Take the first k left singular vectors to form the matrix U′(M×k), take the first k right singular vectors to form the matrix V′((N×P)×k), the diagonal matrix ∑′(k×k) contains the first k singular values, and the observation matrix after dimensionality reduction is:

[0120] A reduced =U′×∑′×V′ T , the dimension is reduced from M×(N×P) to M×k, which significantly reduces the computational complexity. V represents the right singular matrix; the dimension is (N×P)×(N×P), which represents the orthogonal basis of the original data in the column space (error type×time). By Perform QR decomposition to obtain; Q represents the orthogonal matrix, which is used to update the left / right singular matrix; R represents the upper triangular matrix, which is the intermediate variable in the iterative process. For example: Perform QR decomposition on B = A × V to obtain B = Q × R, at this time Q is used to update U; Perform QR decomposition to obtain C = Q' × R', at this time Q' is used to update V; represents the transpose of A); (N×P)×M is the dimension, which means it is used when calculating the right singular matrix U represents the left singular matrix, dimension: M×M, which represents the orthogonal basis of the original data in the row space (grid point), obtained by performing QR decomposition on the intermediate matrix B=A×V; Q′ and R′ are similar to Q and R, but are used to update the right singular matrix, which is given by The QR decomposition of is generated (C = Q' × R').

[0121] The specific implementation process of the above step S2-2 can be:

[0122] Calculate the signal quality index of each sensor node, including metrics such as signal strength, signal-to-noise ratio, and multipath effects. For example, the signal quality index of a UWB tag is determined by factors such as the received signal strength index (RSSI) and the clarity of the channel impulse response (CIR). Evaluate the energy efficiency of each sensor node, defined as the amount of valid data acquired per unit of energy consumed. For example, the energy efficiency of a MEMS gyroscope is the number of valid attitude data points collected per second divided by the power consumption. Normalize the signal quality index and energy efficiency ratio of each sensor node, mapping the metrics of different dimensions to the range [0, 1]. Calculate the power consumption weighting factor based on the normalized signal quality and energy efficiency. Nodes with better signal quality and higher energy efficiency receive a larger power consumption weight.

[0123] The phase calibration coefficient and power consumption weight factor are subjected to a tensor fusion operation. Multidimensional dynamic calibration parameters are generated by combining the physical characteristics of the sensor node (such as communication distance and sampling frequency range) and the construction environment parameters (such as temperature and humidity). Based on the fusion results, the operating mode switching priority is defined for each sensor node. For example, in areas with strong electromagnetic interference, the high-precision positioning mode of the UWB tag is prioritized, and the sampling frequency of secondary sensors (such as the ambient temperature and humidity sensor) is reduced. A sampling frequency adjustment coefficient is generated to dynamically adjust the sensor sampling frequency based on the current positioning requirements and energy consumption status. The sampling frequency is increased in areas with high security risks and reduced in areas with low risks, achieving a dynamic balance between energy consumption and accuracy.

[0124] In an embodiment of the present invention, by establishing a multi-objective optimization model, the overall energy consumption of the sensor network is reduced by 30%-40% while ensuring positioning accuracy, extending the device's battery life and reducing the labor cost of frequent battery replacement. The dynamic calibration parameters can adapt to changes in electromagnetic interference at the construction site in real time, automatically improve calibration accuracy in areas with dense metal components, and reduce energy consumption in areas with weaker interference, so that the positioning system maintains stable performance in complex and changeable construction environments. Through power consumption weighting factors and working mode priorities, intelligent allocation of sensor node resources is achieved, and sensors in key monitoring areas (such as the edge of deep foundation pits and tower crane operating areas) obtain more energy guarantees, thereby improving the reliability of safety monitoring.

[0125] In a preferred embodiment of the present invention, step S3: establishing a three-base-point dynamic topological structure within the target monitoring domain, wherein the first base point is arranged at the vertical extension end of the slewing bearing axis of the tower crane, the second base point is embedded in the radial section of the load-bearing flange of the hydraulic jacking module, and the third base point is fixed to the intersection node of the deep foundation pit crown beam and the support system. The non-collinear vector group is constructed by the three-point coordinate offset, including:

[0126] Step S3-1: Based on the dynamic calibration parameter set output from step S2, the dynamic offset of the three-dimensional spatial coordinates of the three reference points is analyzed, two sets of basis vectors are generated by the coordinate difference, and the linear correlation between the basis vectors is eliminated by the Gram-Schmidt orthogonalization method to construct a standard orthogonal basis vector set representing the monitoring domain spatial reference frame;

[0127] Step S3-2: Perform a spatial mapping operation on the standard orthogonal basis vector group generated in step S3-1 and the dynamic calibration parameter set, convert the phase offset in the calibration parameters into the projection component of the orthogonal basis space through the rotation transformation algorithm, and construct a three-dimensional topological transformation matrix containing the displacement gradient amplitude, direction cosines and spatial azimuth angle to form a non-collinear vector group that describes the structural deformation characteristics.

[0128] In the embodiment of the present invention, the above step S3-1, when applied specifically, can be implemented by the following steps:

[0129] Physical deployment of benchmark points:

[0130] The first reference point (P1): A stainless steel prism with a UWB positioning base station is buried on the concrete foundation surface directly below the slewing bearing axis of the tower crane (the extended end of the vertical line). The vertical deviation between the center of the prism and the slewing bearing axis is controlled within ±2mm. A dual-axis inclination sensor (accuracy of ±0.01°) is installed to monitor foundation settlement and tilt.

[0131] The second reference point (P2): A customized L-shaped bracket is welded on the radial section of the load-bearing flange of the hydraulic jacking module (at 1 / 2 the radius from the center of the circle). A high-precision MEMS inertial sensor (with an integrated three-axis accelerometer / gyroscope and a sampling frequency of 100 Hz) and a laser ranging module (with a range of 50 m and an accuracy of ±1 mm) are fixed to monitor the radial and angular displacements of the flange in real time.

[0132] The third reference point (P3): At the intersection of the deep foundation pit crown beam and the steel support (select the node with the largest support axial force), use expansion bolts to fix the three-dimensional displacement meter (X / Y / Z three-dimensional range ±100mm, accuracy ±0.1mm), and paste the strain gauge (sensitivity coefficient 2.0±1%) on the surface of the crown beam to synchronously monitor the linear displacement and strain changes of the node.

[0133] Dynamic coordinate acquisition: The sensors of the three reference points are connected to the IoT edge computing node (data is collected every 50ms), and the dynamic parameters calibrated in step S2 (such as the phase calibration coefficient of the UWB tag and the sampling frequency adjustment value of the inertial sensor) are used to output the three-dimensional coordinates (X, Y, Z) in real time:

[0134] The P1 coordinate is affected by the settlement and horizontal displacement of the tower crane foundation and is obtained through differential positioning between the UWB base station and the ground fixed reference station; the P2 coordinate is combined with the hydraulic stroke data of the jacking module (from the PLC controller) and the attitude solution results of the inertial sensor to eliminate mechanical vibration interference; the P3 coordinate is generated by comparing the actual measured value of the displacement meter with the theoretical coordinate of the crown beam BIM model and then compensated by the dynamic error matrix.

[0135] Calculate the coordinate differences of adjacent reference points to generate two sets of initial basis vectors:

[0136] Vector v1 = P2 coordinate - P1 coordinate, reflecting the relative displacement between the tower crane lifting module and the foundation;

[0137] Vector v2 = P3 coordinate - P1 coordinate, reflecting the relative displacement between the deep foundation pit node and the tower crane foundation. Ensure that v1 and v2 are not collinear (by judging whether the vector cross product modulus length is greater than a threshold, such as 0.1m 2 , adjust P3 position if not satisfied).

[0138] Gram-Schmidt orthogonalization:

[0139] Take v1 as the first basis vector e1, keeping its direction unchanged;

[0140] Calculate the projection component of v2 on e1, subtract the projection from v2, and obtain the vector e2' orthogonal to e1;

[0141] Normalize e2' (normalize the modulus to 1) to obtain the second standard orthogonal basis vector e2;

[0142] A third basis vector e3 perpendicular to e1 and e2 is constructed (obtained by cross product of e1×e2 to ensure the right-hand coordinate system) to form a standard orthogonal basis vector group {e1, e2, e3} in three-dimensional space. This can eliminate the vector correlation caused by factors such as tower crane vibration and foundation pit soil creep, allowing each basis vector to independently characterize the deformation characteristics in different directions.

[0143] When the above step S3-2 is applied in a specific application, it can be implemented by the following steps:

[0144] Dynamic calibration parameter space mapping:

[0145] The position-related phase offset (such as the propagation delay deviation of the UWB signal and the attitude angle drift of the inertial sensor) in the dynamic calibration parameters generated in step S2 is extracted and mapped to the orthogonal basis space in the following way: for the phase offset in the P1-P2 direction, it is converted into the projection component of the e1 axis to reflect the vertical lifting error of the jacking module; for the phase offset in the P3-P1 direction, it is decomposed into the projection components of the e2 axis (horizontal displacement of the crown beam) and the e3 axis (vertical settlement of the crown beam), and the displacement gradient is corrected in combination with the strain gauge data; when the monitoring domain is overall During rotation (such as tower crane rotation operation and foundation pit support system force deflection), the coordinate frame is adjusted through the following steps: calculate the angle between e1 and the direction of gravity acceleration (through the inclination sensor data), generate the rotation matrix Rz around the e3 axis, and correct the horizontal deviation; calculate the angle between e2 and the foundation pit support axis (through the preset angle of the BIM model), generate the rotation matrix Rx around the e1 axis, and align the main force direction of the structure; combine Rz and Rx to obtain the total rotation matrix R, and convert the original displacement data in the calibration parameters into coordinate components of the orthogonal basis space.

[0146] Construction of non-collinear vector groups and extraction of deformation feature parameters:

[0147] Calculate the displacement change rate of adjacent reference points in the orthogonal basis direction (such as the displacement speed of P2 relative to P1 in the e1 axis direction) to reflect the severity of structural deformation;

[0148] Direction cosines: record the projection coefficient of each deformation vector on the e1 / e2 / e3 axis (absolute value ≤ 1) to clarify the main direction of displacement (for example, the displacement of deep foundation pit nodes is mainly along the e2 axis (horizontal direction));

[0149] Spatial azimuth: The angle between e1 and e2 (comparison between real-time monitoring value and initial value) is used to determine the overall rotation trend of the monitoring domain (such as the azimuth change caused by the rotation of the tower crane).

[0150] 3D topological transformation matrix generation: Arrange the above parameters in rows to form a 3×3 topological transformation matrix T, where each row corresponds to the deformation eigenvector of a reference point:

[0151] The first row of T corresponds to the orthogonal basis components of the P1-P2 vector, which characterizes the axial deformation of the tower crane lifting system;

[0152] The second row of T corresponds to the orthogonal basis components of the P1-P3 vectors, representing the spatial displacement of the deep foundation pit nodes;

[0153] The third row of T is obtained by the e1×e2 cross product, which represents the degree of spatial distortion of the monitoring domain (such as the three-dimensional deformation of the support system after compression).

[0154] In the embodiment of the present invention, the three reference points for global monitoring of key structures are anchored to the three high-risk structures of the tower crane foundation (P1), the jacking module (P2), and the deep foundation pit support (P3), forming a three-dimensional monitoring network covering vertical lifting, horizontal displacement, and three-dimensional deformation, which improves the structural risk identification capability by 80% compared with traditional single-point monitoring; the dynamic error self-calibration eliminates the linear correlation between sensor noise and mechanical vibration through Gram-Schmidt orthogonalization, and combines the dynamic calibration parameters of step S2 to improve the coordinate accuracy of the reference point to ±2mm (compared to ±10mm of traditional RTK positioning), ensuring deformation detection Small displacements (such as 0.5mm-level settlement) can be effectively captured, and the three-dimensional topological transformation matrix of spatial deformation precise modeling decomposes the deformation of complex structures into independent components in orthogonal basis directions (displacement gradient, direction cosine, azimuth); adaptive benchmark point deployment in the construction phase is deeply integrated with the BIM model (such as the pre-defined three-dimensional coordinates of the P3 node corresponding to the support system). When the construction progress changes (tower crane lifting, support removal), the physical parameters of the benchmark points and the matrix calculation logic can be automatically updated through the Internet of Things platform without manual recalibration. The system adaptation cycle is shortened from 24 hours in the traditional solution to 5 minutes.

[0155] In a preferred embodiment of the present invention, based on the non-collinear vector group, the dynamic calibration parameter set output in step S2 is subjected to deformation compensation correction by a vector space topological transformation algorithm to generate a topological reconstruction data stream including a displacement gradient tensor and a strain energy density, including:

[0156] Step S3-3: Based on the three-dimensional topological transformation matrix in the non-collinear vector group in step S3-2, a tensor product operation is performed on the phase offset and the displacement gradient amplitude in the dynamic calibration parameter set to calculate a displacement gradient tensor that describes the rate of change of the spatial relative displacement of each point in the monitoring domain. The displacement gradient tensor includes normal strain and shear strain components.

[0157] Step S3-4: Using the displacement gradient tensor output from step S3-3, the second-order component of the Green's strain tensor is calculated through a continuum mechanics model. Combined with the concrete elastic modulus data collected in real time by the distributed fiber optic sensor and the historical value of Poisson's ratio in the structural health monitoring database, the material constitutive equation parameters are dynamically updated to ultimately generate a reconstructed data stream containing a three-dimensional strain energy density distribution. The strain energy density represents the energy accumulation state within the structure.

[0158] In the embodiment of the present invention, when applied in a specific application, the specific implementation process of the above step S3-3 is as follows:

[0159] The three-dimensional topological transformation matrix (dimension 3×3, denoted as T) generated in step S3-2 is aligned in time and space with the dynamic calibration parameter set output in step S2: the phase offset (UWB signal delay deviation) in the calibration parameter is extracted and converted into the relative displacement value of each point in space (unit: meter) through the coordinate mapping relationship; the displacement gradient amplitude (that is, the rate of change of the deformation vector calculated in step S3-2 in the orthogonal basis direction) is extracted and interpolated to each calculation grid in the monitoring domain (such as a three-dimensional grid of 0.5m×0.5m×0.5m) according to the reference point position.

[0160] The monitoring area is divided into N cubic units (e.g. the tower crane base area is divided into 1000 units), and the center position of each unit is recorded as P i (x i ,y i , z i ); In each unit P i At step S3-1, a local coordinate system is established based on the standard orthogonal basis vector group {e1, e2, e3} generated, and the displacement data under the global coordinates is converted into the local coordinate system.

[0161] Tensor product operation:

[0162] For each unit P i , calculate its displacement components (u i , v i , w i ), corresponding to the relative displacements in the directions of e1, e2, and e3 respectively; calculate the partial derivatives of the displacement components with respect to the spatial coordinates, which constitute the 9 components of the displacement gradient tensor:

[0163]

[0164] Decompose the displacement gradient tensor into symmetric and antisymmetric parts:

[0165]

[0166] The symmetric part represents pure deformation (including normal strain and shear strain components), and the antisymmetric part represents rigid rotation (no strain is generated, only the rotation of the rigid body is described).

[0167] Where u represents the displacement vector, which is a vector function that depends on the spatial coordinates (x, y, z), that is, u = u(x, y, z). In practical applications, u usually consists of three components: u, v, and w, corresponding to the displacement in the x, y, and z directions. It is the Hamiltonian operator, also known as the Nabla operator. In three-dimensional space, its expression is When it acts on a scalar function or vector function, it will produce corresponding calculation results such as gradient, divergence or curl; Represents the gradient of the displacement vector u, which is a second-order tensor (i.e., matrix) that describes the rate of change of the displacement in all directions of space.

[0168] Symmetrical part is the displacement gradient tensor The transpose operation is to exchange the rows and columns of the matrix, that is, for a matrix A = [a ij ], its transposed matrix AT=[a ji ].

[0169] is the displacement gradient tensor The symmetric part of is represented by ∈, which includes normal strain and shear strain components, describing the pure deformation of the object. Normal strain reflects the elongation or compression of the object in a certain direction, while shear strain describes the angular change between the two directions of the object.

[0170] is the displacement gradient tensor The antisymmetric part of , denoted by ω, represents the rigid rotation of the object, that is, the rotation of the object around an axis without strain. The existence of the antisymmetric part is because the displacement gradient tensor may contain rotation information in some cases, and this rotation does not produce strain.

[0171] When applied in practice, the specific implementation process of the above steps S3-4 is as follows:

[0172] Using the displacement gradient tensor obtained in step S3-3, calculate the second-order component of the Green strain tensor (ignore the third-order and higher-order terms, which is applicable to small deformation conditions):

[0173] Normal strain component:

[0174] Shear strain component: The other components are similarly expanded, for a total of 6 independent components;

[0175] Dynamic update of material constitutive parameters:

[0176] Elastic modulus acquisition: Distributed fiber optic sensors (such as the BOTDA system) collect temperature-compensated strain data from each monitoring point in real time and convert it into elastic modulus E (unit: GPa) through a calibration curve. The acquisition frequency is 1 Hz.

[0177] Poisson's ratio determination: Historical data is retrieved from the structural health monitoring database, and the Poisson's ratio ν of the current construction phase is updated in a Bayesian manner:

[0178] Prior distribution: based on historical statistical values ​​of similar projects (e.g., the Poisson's ratio of concrete is 0.18-0.22);

[0179] Likelihood function: constructed based on the currently monitored stress-strain relationship;

[0180] Posterior distribution: The most likely Poisson ratio value is obtained through MCMC sampling.

[0181] Substitute the updated elastic modulus E and Poisson's ratio ν into the linear elastic constitutive equation to calculate the stress components at each point:

[0182]

[0183] Strain energy density calculation: For each element P i , calculate its strain energy density W:

[0184]

[0185] This value represents the elastic strain energy stored per unit volume and reflects the energy accumulation state inside the structure. The strain energy density value of each unit is interpolated into the three-dimensional grid of the monitoring domain to generate a continuous strain energy density distribution field. The strain energy density data is organized in time series to form a four-dimensional data stream (x, y, z, t), which supports dynamic viewing of the spatiotemporal evolution of energy accumulation inside the structure.

[0186] Among them, σ xx , σ yy and σ zz Respectively represent the normal stress (normal stress) in the x, y, and z directions. Normal stress refers to the stress component perpendicular to the action surface, which reflects the degree of stretching or compression of the material in that direction; σ xy , σ yz and σ zx Represents the shear stress in the xy, yz, and zx planes respectively. Shear stress refers to the stress component parallel to the action surface, which describes the relative sliding or twisting degree of the material between two directions; E is the elastic modulus, also known as Young's modulus (Young's Modulus), which is an important indicator to measure the material's ability to resist elastic deformation and reflects the stiffness of the material when subjected to force. The larger the elastic modulus, the less likely the material is to undergo elastic deformation; v is Poisson's ratio (Poisson's Ratio), which refers to the absolute value of the ratio of transverse strain to longitudinal strain when the material is under unidirectional tension or compression. Poisson's ratio reflects the relationship between transverse deformation and longitudinal deformation of the material when subjected to force. For most materials, the value of Poisson's ratio is between 0 and 0.5. ∈ xx ,∈ yy and ∈ zzRepresents the normal strain (normal strain) in the x, y, and z directions respectively. The normal strain refers to the ratio of the length change of the material in this direction to the original length, which reflects the degree of stretching or compression of the material in this direction; ∈ xy ,∈ yz and ∈ zx where y represents shear strain in the xy, yz, and zx planes, respectively. Shear strain refers to the angular change of a material within that plane, describing the relative degree of distortion between two directions. W represents strain energy density, representing the elastic strain energy stored per unit volume. It is a scalar quantity that reflects the state of energy accumulation within a structure. A higher strain energy density indicates a greater amount of elastic strain energy stored at that point in the structure, making the structure more susceptible to deformation or failure.

[0187] In a preferred embodiment of the present invention, step S4: constructing a risk situation feature space based on the topological reconstruction data stream output in step S3, generating a multi-level warning trigger threshold through the Lyapunov index stability criterion, and outputting a risk evolution path containing spatiotemporal correlation features, including:

[0188] Step S4-1: Based on the displacement gradient tensor in the topological reconstruction data stream output from step S3, a nonlinear dynamic state equation is established, the time-varying component of the displacement gradient tensor is used as the system state variable, and the historical monitoring data is subjected to delayed embedding processing through phase space reconstruction technology. The embedding dimension and time delay parameter are calculated based on the time series of the displacement gradient tensor, and the evolution trajectory in the high-dimensional phase space is reconstructed; the displacement gradient tensor obtained in real time is mapped into phase space, and the maximum Lyapunov exponent of the state vector is calculated through the adjacent trajectory divergence rate algorithm to generate a dynamic stability index that characterizes the chaotic characteristics of structural deformation;

[0189] Step S4-2: Perform spatiotemporal correlation analysis on the dynamic stability index obtained in step S4-1 and the historical monitoring database to extract the benchmark mean and standard deviation distribution of the structural stability at different construction stages; set three-level warning thresholds based on the benchmark mean: the first-level threshold is the benchmark mean plus 1 times the standard deviation, the second-level threshold is the benchmark mean plus 2 times the standard deviation, and the third-level threshold is the benchmark mean plus 3 times the standard deviation; when the real-time calculated dynamic stability index exceeds the threshold, the corresponding level of warning signal is triggered, and the spatial propagation path and time evolution trend of the abnormal area of ​​the displacement gradient tensor are analyzed through the spatiotemporal correlation matrix, and the full-dimensional risk evolution path including the deformation starting position, propagation direction and rate is output; the spatiotemporal correlation matrix is ​​constructed by performing a convolution operation on the spatial distribution data of the displacement gradient tensor and the timestamp sequence to extract the deformation correlation characteristics of each monitoring point.

[0190] In the embodiment of the present invention, when applied in a specific application, the specific implementation process of the above step S4-1 is as follows:

[0191] Extract the time-varying components (such as normal strain rate) from the displacement gradient tensor output in step S3 Shear strain rate etc.), these components are used as system state variables to form a state vector X = [x1, x2, ..., x n ], where n is the number of state variables (usually 6-9 independent strain rate components).

[0192] The historical monitoring data (such as the displacement gradient tensor time series for 72 consecutive hours, with a sampling frequency of 1 Hz) were subjected to noise reduction processing, and wavelet transform (decomposition level 5, db4 wavelet basis) was used to remove high-frequency noise and retain the true structural deformation signal.

[0193] Delayed embedded calculation:

[0194] Calculate the time delay parameter τ using the autocorrelation function method. When the autocorrelation function value drops to 1 / e of the initial value (about 0.368), the corresponding time interval is τ (usually 2-5 seconds).

[0195] The embedding dimension m is determined using the false nearest neighbor (FNN) method. The minimum m value when the FNN ratio is less than 1% is the embedding dimension (typical values ​​are 5-7).

[0196] Phase space trajectory construction:

[0197] Reconstruct the original time series {x(t)} into a set of points in the m-dimensional phase space:

[0198] Y(t)=[x(t),x(t+τ),x(t+2τ),...,x(t+(m-1)τ)], forming the evolution trajectory of the system, where Y(t) represents the phase space trajectory point at time t and m represents the embedding dimension.

[0199] Lyapunov exponent calculation, adjacent trajectory divergence analysis:

[0200] Select an initial point Y(t0) in the reconstructed phase space and find another point Y′(t0) in its neighborhood. The initial distance between the two points is δY(t0)=Y(t0)-Y′(t0)≤0.1σ, where σ is the standard deviation of the state variable. The average divergence rate after evolution time t is:

[0201]

[0202] Extract the maximum Lyapunov exponent. Calculate the Lyapunov exponent in each direction using the Wolf algorithm and take the maximum value λ. max As an indicator of system stability:

[0203] λ max>0, indicating that the system is chaotic and the deformation is divergent (unpredictable);

[0204] λ max ≈0, indicating critical stability;

[0205] λ max <0, indicating that the system is stable and the deformation is convergent.

[0206] Calculate the frequency, once every 10 seconds, and generate the sequence {λ max (t)}, which reflects the changing trend of the dynamic stability of structural deformation.

[0207] The specific implementation process of the above step S4-2 is as follows:

[0208] The historical monitoring data are classified according to the construction stages (such as pile foundation construction, main structure construction, decoration and renovation), and the dynamic stability index {λ max (t)}Calculate statistical characteristics: calculate the benchmark mean μ, all λ in each construction stage max The arithmetic mean of the values ​​is calculated; the standard deviation σ is calculated to reflect the fluctuation range of the stability index within this stage.

[0209] Three-level warning threshold setting:

[0210] Level 1 warning (yellow): Threshold T1 = μ + σ, indicating that the structural stability begins to deviate from the normal range;

[0211] Level 2 warning (orange): Threshold T2 = μ + 2σ, indicating that the structure has potential safety risks;

[0212] Level 3 warning (red): Threshold T3 = μ + 3σ, indicating that the structure is in a critical instability state and immediate measures must be taken. T1 means that the stability begins to deviate from the normal range; T2 indicates potential safety risks; T3 indicates critical instability and requires immediate intervention.

[0213] Construction of spatiotemporal correlation matrix:

[0214] The spatial discretization divides the monitoring domain into 100×100×100 three-dimensional grids (each grid size is 0.5m×0.5m×0.5m), and the displacement gradient tensor of each position is stored at the center of each grid.

[0215] Time series convolution, which performs a convolution operation on the displacement gradient tensor time series of each grid point and the global timestamp sequence to generate a spatiotemporal correlation matrix C(i, j, k, τ), where i, j, and k represent spatial coordinates and τ represents time delay.

[0216] Abnormal propagation path identification:

[0217] When λ of a grid point maxWhen the value exceeds the warning threshold, the spatiotemporal correlation matrix at that moment is extracted, and the set of grid points with an absolute value of the correlation coefficient greater than 0.7 in the matrix is ​​found. These points constitute the possible path of anomaly propagation. Through time series analysis, the direction of anomaly propagation (such as propagation from the tower crane base to the tower body) and the rate of propagation (such as an expansion speed of 0.5m per minute) are determined. Combining the spatiotemporal correlation matrix with the Lyapunov exponent sequence, a risk evolution path containing the following information is generated:

[0218] Starting position: the grid coordinate where the anomaly first appears;

[0219] Propagation direction: the spatial direction vector of the anomaly expansion;

[0220] Evolution rate: the time rate of change of anomaly development (such as λ max growth rate);

[0221] Impact range: the boundaries of the spatial area corresponding to different warning levels.

[0222] In an embodiment of the present invention, the chaotic characteristics of structural deformation are captured by the Lyapunov exponent, and potential instability trends can be identified 3-5 days in advance. This has higher foresight than traditional threshold alarm systems (traditional methods can usually only provide warnings a few hours in advance). The warning threshold is dynamically adjusted according to the structural characteristics of different construction stages to avoid false alarms or missed alarms caused by "one-size-fits-all" thresholds, thereby increasing the warning accuracy to more than 95% (traditional methods are about 70%). Phase space reconstruction technology effectively handles noise and missing values ​​in monitoring data, and Lyapunov exponent calculation is highly robust to data fluctuations, ensuring stable operation of the system in complex construction environments.

[0223] In a preferred embodiment of the present invention, step S5: executing a nonlinear control strategy according to the risk evolution path of step S4, and implementing adaptive handling of abnormal states through an actuator dynamic priority scheduling algorithm, includes:

[0224] Step S5-1: Based on the full-dimensional risk evolution path output in step S4, a fuzzy logic control rule base is established. The real-time strain energy density change rate and the Lyapunov divergence index are used as input variables. The risk level membership interval is divided by the triangular membership function to generate control decision input parameters including the risk quantification value, deformation propagation direction and rate.

[0225] Step S5-2: Based on the control decision input parameters output in step S5-1, the information entropy value of the risk event is calculated through the probability density function. When the entropy value exceeds the preset threshold, the actuator dynamic scheduling algorithm is triggered to form a control instruction set containing the device response timing.

[0226] In the embodiment of the present invention, when applied in a specific application, the specific implementation process of the above step S5-1 is as follows:

[0227] Two core input variables are obtained in real time through the IoT platform API:

[0228] Strain energy density change rate (ΔW / Δt): The strain energy density data stream from step S4 is used to calculate the energy density difference between adjacent time steps (e.g., 1 second intervals), reflecting the energy accumulation rate inside the structure.

[0229] Lyapunov divergence exponent λ max : The maximum Lyapunov exponent calculated in step S4 characterizes the degree of chaotic divergence of the deformation trajectory.

[0230] Noise filtering: Perform median filtering on the input data (window size 3 seconds) to eliminate high-frequency interference such as equipment vibration and ensure data smoothness.

[0231] Fuzzy membership function definition:

[0232] Risk level classification: Using the triangular membership function, each input variable is divided into 5 fuzzy intervals:

[0233] Low risk (LR): ΔW / Δt ≤ 0.1 J / m 3 ·s,λ max ≤0.05 / s;

[0234] Low to medium risk (MLR): 0.1 < ΔW / Δt ≤ 0.3 J / m 3 ·s, 0.05<λ max ≤0.15 / s;

[0235] Medium risk (MR): 0.3<ΔW / Δt≤0.6J / m 3 ·s, 0.15<λ max ≤0.3 / s;

[0236] Medium-high risk (MHR): 0.6<ΔW / Δt≤1.0J / m 3 ·s, 0.3<λ max ≤0.5 / s;

[0237] High risk (HR): ΔW / Δt > 1.0 J / m 3 ·s,λ max >0.5 / s;

[0238] Visual mapping: Generate a two-dimensional fuzzy membership cloud map on the IoT platform interface, and display the membership value of the current input variable in each risk interval in real time (e.g., ΔW / Δt=0.4J / m at a certain moment). 3 ·s corresponds to the MR interval membership of 0.8).

[0239] Establishment of control rule base and decision generation:

[0240] Rule design: Based on engineering experience, a total of 25 fuzzy rules of 5×5 are established. For example: if ΔW / Δt is HR and λ max is HR, the control decision is "immediate shutdown + highest priority alarm"; if W / Δt is MR and λ max If it is MLR, the control decision is "reducing construction load + regional warning".

[0241] Defuzzification: The center of gravity method is used to convert the fuzzy output into precise control parameters, generating a decision input containing three elements:

[0242] Risk quantification value (0-100): calculated based on the comprehensive membership degree, for example, high risk corresponds to 80-100;

[0243] Deformation propagation direction: the risk evolution path from step S4, such as "crane foundation → deep foundation pit in the northeast direction";

[0244] Propagation rate: Extract the spatiotemporal diffusion parameters in the risk evolution path, such as "0.8m / min".

[0245] The specific implementation process of the above step S5-2 is as follows:

[0246] The control decision data of the past 12 months were extracted from the historical database of the IoT platform, including the risk quantification value, the corresponding timestamp, and the construction environment parameters at that time (such as weather, construction stage, equipment operating status, etc.). Only data records with a risk quantification value greater than 0 were retained, because a risk quantification value of 0 indicates a risk-free state and is not involved in the construction of the probability density function.

[0247] The data were divided into five categories based on risk level: low risk (LR), medium-low risk (MLR), medium risk (MR), medium-high risk (MHR), and high risk (HR), with corresponding risk quantification value ranges of 0-40, 40-60, 60-80, 80-100, and above 100, respectively. For each risk level, kernel density estimation was used to construct the probability density function. Kernel density estimation is a non-parametric estimation method that does not rely on the specific distribution of the data. Instead, it estimates the probability density function through the weighted summation of data points. The Gaussian kernel function was selected as the kernel function because it has good smoothness and symmetry and can better fit the distribution characteristics of the data.

[0248] Determine the bandwidth parameter: The bandwidth parameter determines the width of the kernel function, which has a great influence on the estimation result of the probability density function. The optimal bandwidth parameter is selected through cross-validation methods (such as leave-one-out cross-validation) to make the estimated probability density function fit the data best.

[0249] From the data records of the high-risk (HR) level, we further extracted a data subset with risk quantification values ​​greater than or equal to 80. After analysis of the extracted data subset, we found that it roughly showed Gaussian distribution characteristics. The maximum likelihood estimation method was used to estimate the mean and standard deviation of the Gaussian distribution. Maximum likelihood estimation is a commonly used parameter estimation method that estimates parameters by maximizing the likelihood function of the data.

[0250] Assume that the risk quantification value of high-risk events follows Gaussian distribution N(μ, σ 2 ), where μ is the mean and σ is the standard deviation. μ and σ are estimated by calculating the sample mean and sample standard deviation of the data subset. The PDF of high-risk events exhibits a Gaussian distribution (mean 85, standard deviation 5) when the risk quantification value is ≥80. The estimated probability density function is plotted and compared with the histogram of the data to check whether the PDF can reasonably fit the distribution characteristics of the data. If it is found that the PDF does not fit the data well, the bandwidth parameter can be adjusted or other kernel functions can be tried for re-estimation. Statistical methods such as the Kolmogorov-Smirnov test or the Anderson-Darling test can be used to test whether the estimated PDF conforms to the actual data distribution. If the test result is not significant, it means that the PDF can fit the data well; otherwise, the PDF needs to be further adjusted.

[0251] Based on the estimated probability density function, the information entropy value of each risk level is calculated. Information entropy is an indicator to measure information uncertainty, which reflects the distribution of data. Through the analysis of historical data, it is found that when the information entropy value exceeds 2.5 bits, the system cannot accurately predict the critical state of risk evolution. Therefore, the entropy value warning threshold is set to 2.5 bits. When the real-time calculated information entropy value exceeds this threshold, the executor dynamic scheduling algorithm is triggered.

[0252] Executor dynamic scheduling algorithm:

[0253] Actuator classification: divided into three categories according to function and given initial priority:

[0254] Safety (priority 1): tower crane emergency stop switch, foundation pit automatic support and reinforcement device;

[0255] Early warning (priority 2): sound and light alarms, mobile phone text message notification systems;

[0256] Adjustment category (priority 3): construction machinery power regulator, personnel positioning system guidance instructions.

[0257] Scheduling logic:

[0258] When the information entropy exceeds the threshold, the priority is dynamically adjusted according to the risk quantification value (for example, when the risk is high, the warning priority is increased to 1.5, which is between the safety category and the original priority);

[0259] Generate a control instruction set, including actuator actions (e.g., "Crane controller: Reduce slew speed to 0.1° / s") and timing arrangements (e.g., trigger the emergency stop device first, then activate the alarm 500ms later).

[0260] IoT communication: Send commands to the actuators via the MQTT protocol, using QoS level 2 to ensure reliable command transmission, and the command response timeout is set to 2 seconds.

[0261] The specific implementation process of the above step S5-3 is as follows:

[0262] The time difference between the actuator receiving the command and the actual action is recorded (e.g., the hydraulic valve response delay is 100ms, the tower crane braking delay is 300ms), and real-time data is collected through the pressure sensors and encoders installed on the actuator. The acceleration sensors deployed at key nodes (accuracy ±0.01g, sampling frequency 100Hz) are used to monitor the vibration amplitude and frequency changes of the structure after the actuator is actuated (e.g., the tower vibration frequency drops from 2Hz to 1Hz after an emergency stop).

[0263] In an embodiment of the present invention, intelligent decision-making and adaptive disposal fuzzy logic rule base are combined with engineering experience and real-time data, so that the system can quickly respond to complex risk scenarios (such as tower crane-foundation pit coupled deformation), which is 80% more efficient than manual decision-making, avoiding the expansion of accidents caused by human judgment delays; the actuator collaborative optimization dynamic priority scheduling ensures that safety equipment takes priority action in high-risk situations (response time <1 second), and at the same time avoids excessive response in low-risk scenarios through information entropy thresholds, reducing the error operation rate by more than 60%.

[0264] like Figure 2 As shown, an embodiment of the present invention also provides a construction site safety monitoring and early warning system based on the Internet of Things, including:

[0265] The construction module is used to build a multimodal positioning compensation model, obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generate a dynamic error distribution matrix;

[0266] A calculation module is used to establish a dynamic energy consumption adaptation mechanism based on a heterogeneous sensor network, perform Kalman gain iterative calculation on the dynamic error distribution matrix, and generate a dynamic calibration parameter set including phase calibration coefficients and power consumption weights;

[0267] The correction module is used to establish a three-base-point dynamic topological structure within the target monitoring domain. The first base point is located at the extended end of the vertical line of the tower crane's slewing bearing axis, the second base point is embedded in the radial section of the hydraulic jacking module's load-bearing flange, and the third base point is fixed to the intersection of the deep foundation pit crown beam and the support system. A non-collinear vector group is constructed through the three-point coordinate offset. Based on the non-collinear vector group, the dynamic calibration parameter set is corrected for deformation compensation using a vector space topological transformation algorithm to generate a topological reconstruction data stream containing the displacement gradient tensor and strain energy density.

[0268] The early warning module is used to construct a risk situation feature space based on topologically reconstructed data streams, generate multi-level early warning trigger thresholds through the Lyapunov index stability criterion, and output a risk evolution path containing spatiotemporal correlation characteristics;

[0269] The control module is used to execute nonlinear control strategies according to the risk evolution path and realize adaptive handling of abnormal states through the actuator dynamic priority scheduling algorithm.

[0270] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A construction site safety monitoring and early warning method based on the Internet of Things, characterized in that: The method comprises: Step S1: Construct a multimodal positioning compensation model, obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generate a dynamic error distribution matrix; Step S2: establishing a dynamic energy consumption adaptation mechanism based on the heterogeneous sensor network, performing Kalman gain iterative calculation on the dynamic error distribution matrix output in step S1, and generating a dynamic calibration parameter set including phase calibration coefficients and power consumption weights; Step S3: Establish a three-base-point dynamic topological structure within the target monitoring domain, where the first base point is located at the extended end of the vertical line of the tower crane's slewing bearing axis, the second base point is embedded in the radial section of the hydraulic jacking module's load-bearing flange, and the third base point is fixed to the intersection of the deep foundation pit crown beam and the support system. A non-collinear vector group is constructed using the three-point coordinate offset. Based on the non-collinear vector group, the dynamic calibration parameter set output from step S2 is subjected to deformation compensation correction using a vector space topological transformation algorithm to generate a topological reconstruction data stream containing the displacement gradient tensor and strain energy density. Step S4: Based on the topological reconstruction data stream output from step S3, a risk situation feature space is constructed, a multi-level warning trigger threshold is generated through the Lyapunov index stability criterion, and a risk evolution path including spatiotemporal correlation characteristics is output; Step S5: Execute the nonlinear control strategy according to the risk evolution path of step S4, and realize adaptive handling of abnormal states through the actuator dynamic priority scheduling algorithm.

2. The method for safety monitoring and early warning of construction sites based on the Internet of Things according to claim 1 is characterized in that: Step S1: Construct a multimodal positioning compensation model, obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generate a dynamic error distribution matrix, including: Step S1-1: Extract electromagnetic distortion features using the covariance relationship of the UWB signal propagation delay in the construction area. Simultaneously collect three-dimensional angular velocity drift data output by the micro-electromechanical inertial sensor. Combine the electromagnetic distortion features with the inertial drift parameters to form a six-dimensional error state vector that includes spatial orientation, signal attenuation, and motion deviation. Step S1-2: Based on the six-dimensional error state vector generated in step S1-1, perform multimodal feature analysis on the distribution of metal components at the construction site: use the mean value of the electromagnetic field strength in the error state vector to locate the metal component aggregation area, calculate the metal density weight factor of each area in combination with the covariance relationship, and generate a dynamic error distribution matrix through probability density function modeling, in which the matrix elements are dynamically calibrated by the joint distribution characteristics of the electromagnetic interference standard deviation and the inertial drift parameter.

3. The method for safety monitoring and early warning of construction sites based on the Internet of Things according to claim 2 is characterized in that: Step S2: Establish an energy consumption dynamic adaptation mechanism based on the heterogeneous sensor network, perform Kalman gain iterative calculation on the dynamic error distribution matrix output in step S1, and generate a dynamic calibration parameter set including phase calibration coefficients and power consumption weights, including: Step S2-1: Based on the dynamic error distribution matrix output from step S1, a multi-objective optimization model for sensor network energy consumption and positioning accuracy is established. The principal component features of the matrix are extracted through singular value decomposition, and a reduced-dimensional observation space is constructed. A weighted least squares optimization algorithm is used to solve the phase calibration coefficients, where the weight matrix is ​​dynamically configured according to the electromagnetic interference intensity of the area where each sensor node is located, to generate a preliminary calibration parameter set. Step S2-2: Based on the preliminary calibration parameter set outputted in step S2-1, the signal quality index and energy efficiency ratio of each sensor node are analyzed, and the power consumption weight factor is calculated through normalization processing; the phase calibration coefficient and the power consumption weight factor are subjected to tensor fusion operation to generate a dynamic calibration parameter set including the node working mode switching priority and the sampling frequency adjustment coefficient.

4. The method for safety monitoring and early warning of construction sites based on the Internet of Things according to claim 3 is characterized in that: Step S3: Establish a three-base-point dynamic topology structure within the target monitoring domain, where the first base point is located at the vertical extension of the tower crane's slewing bearing axis, the second base point is embedded in the radial section of the hydraulic jacking module's load-bearing flange, and the third base point is fixed to the intersection of the deep foundation pit's crown beam and the support system. A non-collinear vector group is constructed using the three-point coordinate offset, including: Step S3-1: Based on the dynamic calibration parameter set output from step S2, the dynamic offset of the three-dimensional spatial coordinates of the three reference points is analyzed, two sets of basis vectors are generated by the coordinate difference, and the linear correlation between the basis vectors is eliminated by the Gram-Schmidt orthogonalization method to construct a standard orthogonal basis vector set representing the monitoring domain spatial reference frame; Step S3-2: Perform a spatial mapping operation on the standard orthogonal basis vector group generated in step S3-1 and the dynamic calibration parameter set, convert the phase offset in the calibration parameters into the projection component of the orthogonal basis space through the rotation transformation algorithm, and construct a three-dimensional topological transformation matrix containing the displacement gradient amplitude, direction cosines and spatial azimuth angle to form a non-collinear vector group that describes the structural deformation characteristics.

5. The method for safety monitoring and early warning of construction sites based on the Internet of Things according to claim 4 is characterized in that: Based on the non-collinear vector group, the dynamic calibration parameter set output in step S2 is subjected to deformation compensation correction by the vector space topological transformation algorithm to generate a topological reconstruction data stream containing the displacement gradient tensor and strain energy density. include: Step S3-3: Based on the three-dimensional topological transformation matrix in the non-collinear vector group in step S3-2, a tensor product operation is performed on the phase offset and the displacement gradient amplitude in the dynamic calibration parameter set to calculate a displacement gradient tensor that describes the rate of change of the spatial relative displacement of each point in the monitoring domain. The displacement gradient tensor includes normal strain and shear strain components. Step S3-4: Using the displacement gradient tensor output from step S3-3, the second-order component of the Green's strain tensor is calculated through a continuum mechanics model. Combined with the concrete elastic modulus data collected in real time by the distributed fiber optic sensor and the historical value of Poisson's ratio in the structural health monitoring database, the material constitutive equation parameters are dynamically updated to ultimately generate a reconstructed data stream containing a three-dimensional strain energy density distribution. The strain energy density represents the energy accumulation state within the structure.

6. The method for safety monitoring and early warning of construction sites based on the Internet of Things according to claim 5 is characterized in that: Step S4: Based on the topological reconstruction data stream output from step S3, a risk situation feature space is constructed. Multi-level warning trigger thresholds are generated through the Lyapunov index stability criterion, and a risk evolution path containing spatiotemporal correlation features is output, including: Step S4-1: Based on the displacement gradient tensor in the topological reconstruction data stream output from step S3, a nonlinear dynamic state equation is established, the time-varying component of the displacement gradient tensor is used as the system state variable, and the historical monitoring data is subjected to delayed embedding processing through phase space reconstruction technology. The embedding dimension and time delay parameter are calculated based on the time series of the displacement gradient tensor, and the evolution trajectory in the high-dimensional phase space is reconstructed; the displacement gradient tensor obtained in real time is mapped into phase space, and the maximum Lyapunov exponent of the state vector is calculated through the adjacent trajectory divergence rate algorithm to generate a dynamic stability index that characterizes the chaotic characteristics of structural deformation; Step S4-2: Perform spatiotemporal correlation analysis on the dynamic stability index obtained in step S4-1 and the historical monitoring database to extract the benchmark mean and standard deviation distribution of the structural stability at different construction stages; set three-level warning thresholds based on the benchmark mean: the first-level threshold is the benchmark mean plus 1 times the standard deviation, the second-level threshold is the benchmark mean plus 2 times the standard deviation, and the third-level threshold is the benchmark mean plus 3 times the standard deviation; when the real-time calculated dynamic stability index exceeds the threshold, the corresponding level of warning signal is triggered, and the spatial propagation path and time evolution trend of the displacement gradient tensor anomaly area are analyzed through the spatiotemporal correlation matrix, and the full-dimensional risk evolution path including the deformation starting position, propagation direction and rate is output.

7. The method for safety monitoring and early warning of construction sites based on the Internet of Things according to claim 6 is characterized in that: The method for constructing the spatiotemporal correlation matrix is ​​to perform a convolution operation on the spatial distribution data of the displacement gradient tensor and the timestamp sequence to extract the deformation correlation characteristics of each monitoring point.

8. The method for safety monitoring and early warning of construction sites based on the Internet of Things according to claim 7 is characterized in that: Step S5: Execute the nonlinear control strategy according to the risk evolution path of step S4, and realize adaptive handling of abnormal conditions through the actuator dynamic priority scheduling algorithm, including: Step S5-1: Based on the full-dimensional risk evolution path output in step S4, a fuzzy logic control rule base is established. The real-time strain energy density change rate and the Lyapunov divergence index are used as input variables. The risk level membership interval is divided by the triangular membership function to generate control decision input parameters including the risk quantification value, deformation propagation direction and rate. Step S5-2: Based on the control decision input parameters output in step S5-1, the information entropy value of the risk event is calculated through the probability density function. When the entropy value exceeds the preset threshold, the actuator dynamic scheduling algorithm is triggered to form a control instruction set containing the device response timing.

9. A construction site safety monitoring and early warning system based on the Internet of Things, the system implementing the method according to any one of claims 1 to 8, characterized in that: include: The construction module is used to build a multimodal positioning compensation model, obtain the electromagnetic distortion characteristics and inertial drift parameters of the construction site through distributed sensing terminals, and generate a dynamic error distribution matrix; A calculation module is used to establish a dynamic energy consumption adaptation mechanism based on a heterogeneous sensor network, perform Kalman gain iterative calculation on the dynamic error distribution matrix, and generate a dynamic calibration parameter set including phase calibration coefficients and power consumption weights; The correction module is used to establish a three-base-point dynamic topological structure within the target monitoring domain. The first base point is located at the extended end of the vertical line of the tower crane's slewing bearing axis, the second base point is embedded in the radial section of the hydraulic jacking module's load-bearing flange, and the third base point is fixed to the intersection of the deep foundation pit crown beam and the support system. A non-collinear vector group is constructed through the three-point coordinate offset. Based on the non-collinear vector group, the dynamic calibration parameter set is corrected for deformation compensation using a vector space topological transformation algorithm to generate a topological reconstruction data stream containing the displacement gradient tensor and strain energy density. The early warning module is used to construct a risk situation feature space based on topologically reconstructed data streams, generate multi-level early warning trigger thresholds through the Lyapunov index stability criterion, and output a risk evolution path containing spatiotemporal correlation characteristics; The control module is used to execute nonlinear control strategies according to the risk evolution path and realize adaptive handling of abnormal states through the actuator dynamic priority scheduling algorithm.

10. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 8.

Images