Construction site safety risk intelligent early warning system and method based on BIM and big data analysis

Abstract

The invention discloses a construction site safety risk intelligent early warning system and method based on BIM and big data analysis, relates to the technical field of building engineering construction safety, and solves the problem that it is difficult to transmit construction site multi-source data which is collected and preprocessed in real time in real time and carry out space mapping with a BIM model. A rule engine is difficult to carry out initial early warning; a machine learning model is difficult to analyze time series data, predict collapse risks and identify dangerous behaviors; a risk prediction model is difficult to construct and is difficult to integrate into a BIM model; and pushing and closed-loop management are difficult to carry out on the risk early warning information. According to the method, the multi-source data is collected at the construction site, the digital twinborn scene is constructed by mapping the multi-source data to the BIM model by means of space-time alignment, the multi-source data is analyzed and processed by applying technologies such as a rule engine and a machine learning algorithm, and the result is integrated to the BIM model, so that visual risk monitoring and early warning are realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of construction engineering construction safety, and specifically relates to an intelligent early warning system and method for construction site safety risks based on BIM and big data analysis. Background Technique

[0002] Traditional construction site safety management relies on manual inspections and static risk assessments, which have problems such as low efficiency, poor real-time performance, and easy omission of dynamic hazard sources. Although the existing BIM technology can provide a three-dimensional visualization model, it lacks dynamic interaction with real-time data; the application of big data technology in construction safety is mostly limited to historical data analysis and has not formed a closed-loop early warning - investigation mechanism. Therefore, there is an urgent need for an active safety management system that integrates BIM dynamic modeling, real-time data collection, and intelligent analysis.

[0003] The following problems exist in the prior art: It is difficult to transmit the multi-source data of the construction site collected in real time and preprocessed in real time and perform spatial mapping with the BIM model; it is difficult to use a rule engine for initial early warning; it is difficult to use a machine learning model to analyze time-series data, predict collapse risks, and identify dangerous behaviors; it is difficult to construct a risk prediction model and integrate it into the BIM model; it is difficult to push and manage the risk early warning information in a closed loop. Summary of the Invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes an intelligent early warning system and method for construction site safety risks based on BIM and big data analysis;

[0005] For this reason, the present invention provides the following technical solution: An intelligent early warning system for construction site safety risks based on BIM and big data analysis, including a data collection layer, an edge computing layer, a cloud computing layer, and an application layer:

[0006] The data collection layer: By deploying a variety of sensors at the construction site, it collects the Internet of Things data stream in real time, and collects the on-site image data in real time through intelligent cameras; it transmits the collected Internet of Things data stream and on-site image data to the edge computing layer;

[0007] The edge computing layer: Preprocesses the received Internet of Things data stream, including: data screening and cleaning to eliminate interference data and data standardization; preprocesses the received on-site image data, including: filtering and denoising and image enhancement; uses spatio-temporal alignment technology to map the real-time data to the corresponding spatial coordinates of the BIM model to form a digital twin scenario;

[0008] The cloud computing layer: uses a rule engine to preset safety thresholds for primary early warning; uses a machine learning model to analyze time-series data and predict collapse risks; identifies dangerous behaviors based on on-site image data; uses historical accident data to build a risk prediction model, combines adaptive learning and transfer learning technologies, updates the risk prediction model in real time, and integrates the prediction results into the BIM model.

[0009] Furthermore: it also includes: the application layer: pushes early warnings to the manager's APP according to the risk level, and performs visual display in combination with the BIM model and AR glasses; automatically generates inspection work orders, assigns them to responsible persons and tracks the processing progress, and feeds back the processing results to the system to update the risk status of the BIM model, forming a PDCA cycle.

[0010] Furthermore: real-time collects Internet of Things data streams, including: meteorological data, equipment status data, personnel positioning data, and environmental monitoring data. Among them, meteorological data includes: wind speed, temperature, and humidity; equipment status data includes: crane load and voltage; personnel positioning data includes: personnel location information based on UWB ultra-wideband or RFID radio frequency identification; environmental monitoring data includes: dust and noise.

[0011] Furthermore: uses spatio-temporal alignment technology to map real-time data to the corresponding spatial coordinates of the BIM model, forming a digital twin scenario, including the following steps:

[0012] Build a BIM three-dimensional model including construction progress, equipment layout, and personnel movement lines, and associate design changes and construction logs; the BIM three-dimensional model is connected to the Internet of Things data streams in real time, including: meteorological data, equipment status, personnel positioning, and environmental monitoring;

[0013] Establish a unified spatial coordinate system in the BIM three-dimensional model, use the entrance of the construction site or a corner of a building as the origin, and set the directions of the X, Y, and Z axes, which represent the east-west, north-south, and height directions respectively;

[0014] Convert the actual position information of each sensor at the construction site, including: longitude, latitude, and elevation, to the spatial coordinate system of the BIM three-dimensional model using a three-dimensional coordinate transformation formula; by adopting the Network Time Protocol or a precise time synchronization mechanism, keep the time of each sensor and the system synchronized; map the preprocessed Internet of Things data stream to the corresponding spatial coordinates of the BIM three-dimensional model according to the time series and spatial position;

[0015] Based on the BIM model and the initial sensor data, build a digital twin scenario consistent with the initial state of the construction site; in the digital twin scenario, display the building structure, equipment layout, and personnel distribution information of the construction site; according to the real-time data collection of various sensors for the Internet of Things data stream, update the digital twin scenario in real time.

[0016] Further: Using a rules engine, preset safety thresholds for primary warning, including the following steps:

[0017] The preset safety thresholds include: lack of protection for high-altitude operations, intrusion into dangerous areas, equipment overload, exceeding limits, and operating beyond the designated area. The rules engine receives the preprocessed IoT data stream in real time and matches the IoT data stream with predefined rules; when the IoT data stream meets the conditions of the preset safety thresholds, the rules engine triggers the warning mechanism for initial warning.

[0018] Further: Using a machine learning model to analyze time-series data and predict collapse risks, including the following steps:

[0019] Using the LSTM long short-term memory network model in the machine learning model to analyze time-series data includes: the deformation trend of the support structure and predicting collapse risks;

[0020] Collecting historical time-series data on the deformation of the support structure includes: displacement, strain, inclination, cracks, temperature, humidity, and rainfall; cleaning, normalizing, or standardizing the collected time-series data; labeling the time-series data including: labels indicating whether a collapse has occurred;

[0021] Inputting the labeled historical time-series data on the deformation of the support structure into the LSTM model for training, and inputting the time-series data on the deformation of the support structure collected in real time at the construction site into the trained LSTM model to output the predicted deformation trend of the support structure and the probability of collapse risks.

[0022] Further: Identifying dangerous behaviors based on on-site image data, including the following steps:

[0023] Based on the on-site image data collected and preprocessed in real time at the construction site, using a convolutional neural network model to detect and identify dangerous behaviors;

[0024] Collecting a historical on-site image data set and labeling dangerous behavior type labels including: not wearing safety equipment and illegal stacking;

[0025] Inputting the labeled historical on-site image data set into the convolutional neural network model for training; inputting the on-site image data collected and preprocessed in real time into the trained convolutional neural network model, and the convolutional neural network model identifies whether dangerous behaviors occur in the real-time on-site image data and outputs the categories of the identified dangerous behaviors.

[0026] Further: Using historical accident data to build a risk prediction model, combining adaptive learning and transfer learning techniques, updating the risk prediction model in real time, and integrating the prediction results into the BIM model, including the following steps:

[0027] Collect a historical accident dataset, and use an ensemble learning method to fuse the output results of the rule engine, LSTM model, and convolutional neural network model, and integrate them into the historical accident dataset; label the risk level tags for each accident sample in the historical accident dataset, including: red, orange, and yellow;

[0028] Input the labeled historical accident dataset into the risk prediction model for training; input the preprocessed IoT data stream and on-site image data collected in real time into the trained risk prediction model to output the risk event level;

[0029] Utilize an adaptive learning mechanism, according to the newly collected IoT data stream and new on-site image data of the construction site in real time, and use transfer learning technology to transfer the knowledge of the trained risk prediction model to the new construction site scenario;

[0030] Integrate the output result of the risk prediction model into the BIM model to generate a dynamic risk heat map that changes over time; by overlaying real-time risk level data in the BIM model, the risk distribution of the construction site is displayed in real time.

[0031] Furthermore: It also includes the following steps:

[0032] Divide the risks at the construction site into three levels, including: red, orange, and yellow; according to the predicted results of the risk event level output by the risk prediction model, generate corresponding early warning information, including: risk event, occurrence location, and risk level;

[0033] Push the generated early warning information through the mobile APP of the management personnel. Among them, different levels of early warning information are distinguished by colors and icons, including: red early warning is displayed with a red background and an alarm icon, orange early warning is displayed with an orange background and a warning icon, and yellow early warning is displayed with a yellow background and a prompt icon;

[0034] In the BIM model, automatically mark and highlight the corresponding dangerous areas according to the occurrence location and influence range of the risk event; among them, the highlighting uses colors corresponding to the risk levels, including: red areas represent high-risk areas, orange areas represent medium-risk areas, and yellow areas represent low-risk areas;

[0035] When the patrol personnel wear AR glasses with augmented reality function and conduct patrols at the construction site, fuse the early warning information with the display interface of the AR glasses in real time; when the patrol personnel are in the dangerous area, corresponding risk point markings automatically pop up on the display interface of the AR glasses, and the marking information includes: risk event, risk level, and risk location;

[0036] Automatically generate corresponding risk investigation work orders according to the early warning information, assign them to the responsible persons, and track the progress of handling; the content of the work orders includes: risk events, occurrence locations, risk levels, and discovery times.

[0037] Feed back the processing results to the BIM model, and automatically update the status of the corresponding risk points in the model; through the visual display of the BIM model, the closed-loop risk investigation management process follows the PDCA cycle principle.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] Through dynamic BIM modeling and real-time data access, the present invention realizes real-time visualization and dynamic risk monitoring of the construction site, improving the safety management level; through multi-source data fusion and intelligent analysis, it accurately identifies dangerous events and risk levels, and timely pushes hierarchical early warning information to assist managers in making quick decisions.

[0040] Through the closed-loop risk investigation management and adaptive learning mechanism, the present invention continuously optimizes the risk model, adapts to different construction site scenarios, reduces the false alarm rate, and improves the investigation efficiency. By superimposing real-time risk data on the BIM model, a heat map that changes over time is generated to assist in decision-making; by using edge-cloud collaborative computing technology to process real-time high-frequency data at the edge and perform in-depth analysis in the cloud, the balance between low latency and high precision is guaranteed.

[0041] The present invention can shorten the dangerous identification response time from the hour level to the minute level, reduce the accident rate by more than 40%, and improve the investigation efficiency by 60% through visual closed-loop management. It is applicable to complex engineering scenarios such as large complexes and subway tunnels. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1 It is the system architecture diagram of the present invention;

[0044] Figure 2 It is the method flow diagram of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0045] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] Please refer to Figure 1 , an embodiment of the first aspect of the present invention provides an intelligent early warning system for construction site safety risks based on BIM and big data analysis, including a data acquisition layer, an edge computing layer, a cloud computing layer and an application layer:

[0047] The data acquisition layer: Deploy a variety of sensors at the construction site to collect IoT data streams in real time, and collect on-site image data in real time through intelligent cameras; transmit the collected IoT data streams and on-site image data to the edge computing layer;

[0048] The edge computing layer: Preprocess the received IoT data streams, including: data screening and cleaning to eliminate interference data and data standardization; preprocess the received on-site image data, including: filtering and denoising and image enhancement; use spatio-temporal alignment technology to map real-time data to the corresponding spatial coordinates of the BIM model to form a digital twin scenario;

[0049] The cloud computing layer: Use a rule engine to preset safety thresholds for primary early warning; use a machine learning model to analyze time-series data and predict collapse risks; identify dangerous behaviors based on on-site image data; use historical accident data to build a risk prediction model, combine adaptive learning and transfer learning technologies, update the risk prediction model in real time, and integrate the prediction results into the BIM model.

[0050] Specifically, the hardware deployment includes: installing inclination sensors and load sensors at key positions such as tower cranes and scaffolding, integrating positioning tags into workers' safety helmets, and deploying intelligent cameras at entrances and exits. By deploying a variety of sensors at the construction site, including temperature sensors, pressure sensors, displacement sensors, etc., the Internet of Things data stream is collected in real time. Deploying various sensors at key positions on the construction site ensures that various data during the construction process, such as temperature, humidity, pressure, displacement, etc., can be comprehensively and accurately collected. The collected Internet of Things data stream is preliminarily processed at the edge computing layer to remove interference information such as outliers and error data, ensuring the accuracy and reliability of the data. At the same time, the data is standardized to meet the requirements of subsequent analysis and processing. The on-site image data collected by the intelligent camera is subjected to filtering, denoising, and image enhancement processing to improve the clarity and quality of the image, so as to more accurately identify and analyze the on-site situation. The preprocessed real-time data is spatially and temporally aligned with the BIM model, that is, the data is accurately mapped to the corresponding spatial coordinate positions in the BIM model to form a digital twin scene consistent with the actual construction site. This helps to more intuitively display the real-time state of the construction site and provides a basis for subsequent risk analysis and early warning. At the cloud computing layer, a rule engine is used to perform primary early warning judgments on the real-time data according to preset safety thresholds. When the data exceeds the set safety range, the corresponding early warning mechanism is triggered to promptly remind the management personnel to pay attention to potential safety risks. A machine learning model is used to analyze the time-series data to predict the occurrence probability of major safety risks such as collapses. Through learning historical data and pattern recognition, the model can detect potential risk trends in advance and provide a more scientific decision-making basis for construction safety management. Based on the on-site image data, image recognition technology is used to identify dangerous behaviors of construction workers, such as not wearing safety helmets and operating violations. Historical accident data, including information such as accident types, occurrence times, locations, and causes, is collected and sorted to construct a training data set for the risk prediction model. Machine learning algorithms are used to train the historical accident data to construct an initial risk prediction model. During the actual application process, combined with adaptive learning and transfer learning technologies, the model is updated in real time according to new data and situations, enabling it to continuously adapt to changes in the construction site and improving the accuracy and reliability of the prediction.

[0051] In this embodiment, it further includes: the application layer: pushing early warnings to the manager APP according to the risk level, and performing visual display in combination with the BIM model and AR glasses; automatically generating inspection work orders, assigning them to responsible persons and tracking the processing progress, and after the processing results are fed back to the system, updating the risk status of the BIM model to form a PDCA cycle.

[0052] Specifically, according to the risk level, corresponding early warning information is pushed through the managers' APP to ensure that managers can receive the early warning notifications in a timely manner and take corresponding measures. The early warning information is integrated into the BIM model and visually displayed through AR glasses. Managers can intuitively see the real-time risk status of the construction site, as well as the specific early warning locations and relevant information through the AR glasses, improving the monitoring and management efficiency of the construction site. The system automatically generates inspection work orders and assigns them to the corresponding responsible persons according to the division of responsibilities. The responsible persons need to conduct on-site inspections and handling according to the requirements of the work orders and promptly feedback the progress and results of the handling. After the handling results are fed back to the system, the system automatically updates the risk status in the BIM model, forming a PDCA cycle. Through continuous cyclic improvement, the construction safety management process is continuously optimized, and safety risks are reduced.

[0053] In this embodiment, the Internet of Things data stream is collected in real time, including: meteorological data, equipment status data, personnel positioning data, and environmental monitoring data. Among them, the meteorological data includes: wind speed, temperature, and humidity; the equipment status data includes: crane load and voltage; the personnel positioning data includes: personnel location information based on UWB ultra-wideband or RFID radio frequency identification; the environmental monitoring data includes: dust and noise.

[0054] Specifically, meteorological sensors such as anemometers, thermometers, and hygrometers are deployed at key positions such as tower cranes and scaffolding in the construction site to collect meteorological data in real time and transmit the data to the data acquisition layer; load sensors and voltage sensors are installed at key parts of the crane to collect equipment status data and transmit it to the control unit for preliminary processing; construction workers wear UWB tags or RFID tags, and the personnel positioning data is collected through UWB base stations or RFID readers; dust sensors and noise sensors are deployed in each area of the construction site to collect environmental monitoring data. The data acquisition layer transmits various types of data to the edge computing layer through wired or wireless communication methods for preprocessing operations such as data screening, cleaning, and standardization to eliminate interfering data and improve data quality, providing accurate and reliable data support for subsequent intelligent early warning of safety risks.

[0055] In this embodiment, the real-time data is mapped to the corresponding spatial coordinates of the BIM model using the spatio-temporal alignment technology to form a digital twin scenario, including the following steps:

[0056] Construct a BIM three-dimensional model including construction progress, equipment layout, and personnel movement lines, and associate design changes and construction logs; the BIM three-dimensional model is connected to the Internet of Things data stream in real time, including: meteorological data, equipment status, personnel positioning, and environmental monitoring;

[0057] Establish a unified spatial coordinate system in the BIM 3D model, with the entrance of the construction site or a corner of a building as the origin, and set the directions of the three axes X, Y, and Z, which represent the east-west, north-south, and height directions respectively;

[0058] Convert the actual position information of each sensor at the construction site, including longitude, latitude, and elevation, to the spatial coordinate system of the BIM 3D model using the 3D coordinate transformation formula; keep the time of each sensor and the system synchronized by adopting the Network Time Protocol or the Precision Time Synchronization mechanism; map the preprocessed Internet of Things data stream to the corresponding spatial coordinates of the BIM 3D model according to the time series and spatial position;

[0059] Build a digital twin scenario consistent with the initial state of the construction site based on the BIM model and the initial sensor data; in the digital twin scenario, display the building structure, equipment layout, and personnel distribution information of the construction site; update the digital twin scenario in real time according to the real-time data collection of the Internet of Things data stream by various sensors.

[0060] Specifically, develop a BIM plug-in to interface with mainstream project management software such as Revit and Navisworks to achieve seamless data interaction. According to the construction design drawings and the actual on-site situation, use BIM modeling software to build a BIM three-dimensional model containing information such as construction progress, equipment layout, and personnel movement lines. During the modeling process, integrate detailed information such as the construction progress plan, the specifications and locations of equipment, and the movement trajectories of personnel into the model, and associate design changes and construction logs to ensure the accuracy and timeliness of the model. Through various sensors deployed at the construction site, such as meteorological sensors, equipment status sensors, personnel positioning tags, and environmental monitoring sensors, real-time collect meteorological data, equipment status data, personnel positioning data, and environmental monitoring data; transmit the data collected by these sensors to the data acquisition layer through a wired or wireless communication network, and then the data acquisition layer transmits the data to the edge computing layer and the cloud computing layer for processing. In the BIM three-dimensional model, take the entrance of the construction site or a corner of a building as the origin to establish a unified spatial coordinate system. Set the X-axis direction as the east-west direction, the Y-axis direction as the north-south direction, and the Z-axis direction as the height direction to ensure that the coordinate system is consistent with the actual geographical orientation of the construction site, facilitating subsequent data mapping and scene construction. Obtain the actual position information of each sensor at the construction site, including longitude, latitude, and elevation. Use the three-dimensional coordinate transformation formula to transform this actual position information into the spatial coordinate system of the BIM three-dimensional model. For example, a transformation matrix from the geographic coordinate system to the BIM coordinate system can be adopted, and the geographic coordinates of the sensor can be converted into three-dimensional coordinates in the BIM model through matrix operations to ensure that the data collected by the sensor can be associated with the BIM model at the correct spatial position. To ensure the consistency of the data collected by each sensor with the system time, use the Network Time Protocol (NTP) or a precise time synchronization mechanism to synchronize the time of each sensor and the system. By setting up an NTP server or using a precise time synchronization device, keep the clocks of each sensor highly consistent with the system clock to ensure that the collected data is comparable and relevant in the time series, providing an accurate time reference for subsequent data mapping and analysis. Map the Internet of Things data stream preprocessed by the edge computing layer, including data screening, statistics, cleaning, standardization, etc., to the corresponding spatial coordinates of the BIM three-dimensional model according to the time series and spatial position. The wind speed, temperature, and humidity values in the meteorological data can be mapped to the corresponding spatial position points in the BIM model according to their collection time and sensor location; the crane load and voltage values in the equipment status data can be mapped to the corresponding positions of the crane in the BIM model. Based on the BIM model and the initial sensor data, construct a digital twin scene consistent with the initial state of the construction site. In this scene, information such as the building structure, equipment layout, and personnel distribution at the construction site can be clearly displayed.With the continuous real-time data collection of various sensors for the Internet of Things data stream, the data and status information in the digital twin scenario are continuously updated to always be synchronized with the actual situation at the construction site, realizing dynamic monitoring and real-time reflection of the construction site. Refer to Table 1 for the conversion results of real-time data mapped to the corresponding spatial coordinates of the BIM model.

[0061] Table 1. Conversion of Corresponding Spatial Coordinates of BIM Model

[0062]

[0063] In this embodiment, using a rule engine, a safety threshold is preset for primary warning, including the following steps:

[0064] The preset safety thresholds include: lack of high-altitude operation protection, intrusion into dangerous areas, equipment overload, over-limit, and operation beyond the specified area. The rule engine receives the preprocessed Internet of Things data stream in real time and matches the Internet of Things data stream with predefined rules; when the Internet of Things data stream meets the conditions of the preset safety threshold, the rule engine triggers the warning mechanism for initial warning.

[0065] Specifically, corresponding safety thresholds are preset according to common safety risks at the construction site. Among them, the lack of high-altitude operation protection is judged by combining personnel positioning data and equipment status data to determine whether construction workers correctly wear protective equipment such as safety belts and safety ropes during high-altitude operations. If the sensor detects that a person is in the high-altitude operation area and not wearing protective equipment, an alarm is triggered. The intrusion into a dangerous area is judged based on personnel positioning data to determine whether construction workers enter an unauthorized dangerous area. By setting up a virtual fence in the dangerous area, when the personnel positioning information shows entry into this area, an alarm is triggered. For the overloading, over-limiting, and over-area operation of equipment, taking a crane as an example, its load, voltage, operating speed and other parameters are monitored through equipment status sensors; when the load of the crane exceeds 80% of the rated load, the voltage fluctuation exceeds ±10% of the normal range, the operating speed exceeds the specified limit or the operating area exceeds the preset range, an alarm is triggered. A rule engine is deployed at the cloud computing layer, and rules corresponding to the preset safety thresholds are configured. The rule engine receives in real time the preprocessed Internet of Things data stream from the edge computing layer, including meteorological data, equipment status data, personnel positioning data, and environmental monitoring data, etc. The rule engine matches the received Internet of Things data stream with the predefined rules. The crane load value in the equipment status data can be matched with the threshold rule for overloading operation of the equipment; the personnel positioning data can be matched with the rule for intrusion into a dangerous area, etc. Through the logical judgment and pattern matching functions of the rule engine, data situations that meet the alarm conditions can be quickly and accurately identified. When the Internet of Things data stream meets the conditions of the preset safety threshold, the rule engine immediately triggers the alarm mechanism. If it is detected that a construction worker is performing high-altitude operations without wearing protective equipment, the rule engine triggers an alarm for the lack of high-altitude operation protection; when the load of the crane exceeds 80% of the rated load, an alarm for overloading operation of the equipment is triggered. After the alarm is triggered, the system generates corresponding alarm information, including alarm type, alarm location, alarm time, etc. Then, the alarm information is pushed to the manager's APP through the application layer to ensure that the manager can receive the alarm notification in time. At the same time, the alarm information can also be visually displayed in the BIM model to intuitively show the safety risk status of the construction site.

[0066] In this embodiment, a machine learning model is used to analyze time series data and predict the collapse risk, including the following steps:

[0067] Using the LSTM long short-term memory network model in the machine learning model to analyze time series data includes: the deformation trend of the support structure and predicting the collapse risk;

[0068] Collecting historical time series data of support structure deformation includes: displacement, strain, inclination, cracks, temperature, humidity and rainfall; cleaning, normalizing or standardizing the collected time series data; labeling the time series data including: labels of whether collapse occurs;

[0069] The time - series data of the historical support structure deformation with labels is input into the LSTM model for training. The time - series data of the support structure deformation collected in real - time at the construction site is input into the trained LSTM model, and the predicted deformation trend of the support structure and the probability of collapse risk are output.

[0070] Specifically, collect the time - series data of the historical support structure deformation, including: displacement, strain, inclination, cracks, temperature, humidity, rainfall and other data. These data can be obtained from the records of previous construction projects and the monitored historical Internet of Things data streams to ensure the integrity and accuracy of the data. Clean the collected time - series data to remove invalid data such as outliers and missing values. Then perform normalization or standardization processing to convert data with different dimensions to the same numerical range to improve the training effect and convergence speed of the model. Label the historical time - series data by adding labels indicating whether collapse has occurred; for normal data without collapse, label it as 0; for abnormal data with collapse, label it as 1; ensure the accuracy and consistency of the labeling so that the model can correctly learn and identify the characteristics of collapse risk. Select the LSTM (Long Short - Term Memory) network model as the machine - learning model for analyzing time - series data. The LSTM model has the unique ability to process and predict time - series data and is suitable for analyzing the time - series data of support structure deformation. When building the LSTM model, determine the number of neurons in the input layer, hidden layer and output layer, and select appropriate parameters such as activation functions, loss functions and optimizers. Divide the time - series data of the historical support structure deformation with labels into a training set and a test set, generally in a ratio of 7:3 or 8:2. Input the training set into the LSTM model for training, and by continuously adjusting the weights and biases of the model, enable the model to accurately learn the relationship between support structure deformation and collapse risk. During the training process, monitor indicators such as the loss value and accuracy of the model to ensure good training effect of the model. Use the test set to verify the trained LSTM model and evaluate the prediction performance and generalization ability of the model. If the model performs poorly on the test set, the structure and parameters of the model can be adjusted and optimized, such as increasing the number of neurons in the hidden layer, adjusting the learning rate, adding regularization, etc., to improve the accuracy and stability of the model. Collect the time - series data of the support structure deformation in real - time at the construction site, including displacement, strain, inclination, cracks, temperature, humidity and rainfall, etc. Perform the same cleaning, normalization or standardization processing on the real - time collected data as on the historical data to ensure the consistency and comparability of the data. Input the processed time - series data of the real - time support structure deformation into the trained LSTM model, and the model will output the predicted deformation trend of the support structure and the probability of collapse risk. According to the prediction results, managers can timely understand the collapse risk status at the construction site and take corresponding preventive measures to ensure construction safety.

[0071] In this embodiment, identifying dangerous behaviors based on on-site image data includes the following steps:

[0072] Based on the on-site image data collected and preprocessed in real time at the construction site, use a convolutional neural network model to detect and identify dangerous behaviors;

[0073] Collect a historical on-site image data set and label the dangerous behavior type tags including: not wearing safety equipment and illegal stacking;

[0074] Input the labeled historical on-site image data set into the convolutional neural network model for training; input the on-site image data collected and preprocessed in real time into the trained convolutional neural network model, and the convolutional neural network model identifies whether a dangerous behavior occurs in the real-time on-site image data and outputs the category of the identified dangerous behavior.

[0075] Specifically, collect the historical on-site image data set of the construction site, covering various construction scenarios and behaviors. Label the dangerous behaviors in the on-site images. The main types include: not wearing safety equipment such as not wearing a safety helmet, not fastening a seat belt, etc. and illegal stacking such as materials being stacked too high and unstably. Professional annotators can be invited to annotate the images one by one to ensure the accuracy and consistency of the annotations. Preprocess the collected image data, including operations such as cropping, resizing, and normalization, to enhance the generalization ability and training effect of the model. All images can be adjusted to the same size for unified processing when input into the convolutional neural network. Input the labeled historical on-site image data set into the convolutional neural network model for training. During the training process, use optimization methods such as backpropagation algorithm and gradient descent method to continuously adjust the weights and biases of the model so that the model can accurately learn the characteristics of dangerous behaviors in the images. At the same time, set appropriate training parameters, such as learning rate, number of iterations, batch size, etc., to improve the training effect and convergence speed of the model. Divide the image data set into a training set and a validation set, generally in a ratio of 7:3 or 8:2. Use the validation set to verify the trained model and evaluate indicators such as the accuracy, recall rate, and F1 value of the model. If the performance of the model on the validation set is not good, the structure and parameters of the model can be adjusted and optimized, such as adding regularization, data augmentation, etc., to improve the generalization ability and robustness of the model. Deploy intelligent cameras at the construction site to collect on-site image data in real time. Perform the same preprocessing operations on the collected images as the historical data to ensure the consistency and comparability of the image data. Input the preprocessed real-time on-site image data into the trained convolutional neural network model, and the model will automatically identify and classify the dangerous behaviors in the images and output the categories of the identified dangerous behaviors.

[0076] In this embodiment, a risk prediction model is constructed using historical accident data. Combining adaptive learning and transfer learning techniques, the risk prediction model is updated in real time, and the prediction results are integrated into the BIM model, including the following steps:

[0077] Collect a historical accident data set, and use an ensemble learning method to fuse the output results of the rule engine, the LSTM model, and the convolutional neural network model, and integrate them into the historical accident data set; label the risk level labels of each accident sample in the historical accident data set, including: red, orange, and yellow;

[0078] Input the labeled historical accident data set into the risk prediction model for training; input the preprocessed Internet of Things data stream and on-site image data collected in real time into the trained risk prediction model to output the risk event level;

[0079] Utilize the adaptive learning mechanism. According to the newly collected Internet of Things data stream and new on-site image data at the construction site in real time, use transfer learning technology to transfer the knowledge of the trained risk prediction model to the new construction site scenario;

[0080] Integrate the output results of the risk prediction model into the BIM model to generate a dynamic risk heat map that changes over time; by overlaying real-time risk level data in the BIM model, the risk distribution at the construction site is displayed in real time.

[0081] Specifically, collect the historical accident dataset of the construction site, including background information of accident occurrence, relevant Internet of Things data streams, including meteorological data, equipment status data, personnel positioning data, environmental monitoring data, etc., on-site image data, and detailed information such as accident type and severity. Integrate the output results of the rule engine, LSTM model, and convolutional neural network model, including early warning information, collapse risk prediction, dangerous behavior identification, etc., into the historical accident dataset. The output results of these models can be used as new features or variables to enrich the information dimension of the historical accident dataset. Label the risk level of each accident sample in the historical accident dataset, divided into three levels: red for high risk, orange for medium risk, and yellow for low risk. The labeling work can be completed by professional safety engineers or domain experts to ensure the accuracy and consistency of the labeling. Adopt an ensemble learning method to combine multiple models of logistic regression, decision tree, random forest, support vector machine, and neural network to construct a risk prediction model to improve the accuracy and stability of prediction. Divide the labeled historical accident dataset into a training set and a test set, generally in a ratio of 7:3 or 8:2. Use the training set to train the risk prediction model, and optimize the performance of the model by adjusting the parameters and hyperparameters of the model. During the training process, techniques such as cross-validation can be adopted to prevent overfitting and improve the generalization ability of the model. Use the test set to evaluate and validate the trained risk prediction model, calculate indicators such as the accuracy, recall rate, and F1 value of the model, and evaluate the prediction performance of the model. If the performance of the model is not ideal, the model can be further adjusted and optimized, such as adding feature engineering and adjusting the model structure. Real-time collect the Internet of Things data stream and on-site image data of the construction site. Preprocess the collected data, including data cleaning, normalization, standardization, etc., to ensure the quality and consistency of the data. Input the preprocessed real-time data into the trained risk prediction model, and the model will output the risk event level of the current construction site, including red, orange, or yellow, according to the characteristics of the input data and the patterns in the historical accident data. Managers can take corresponding measures in a timely manner according to the risk level to prevent and control the occurrence of safety accidents. Establish an adaptive learning mechanism to monitor the new Internet of Things data stream and new on-site image data of the construction site in real time. When new data arrives, the system can automatically judge the change situation of the data. If the data distribution or characteristics change significantly, trigger the model update process. In the new construction site scenario, use transfer learning technology to transfer the knowledge and parameters of the existing risk prediction model to the new scenario. First, collect a certain amount of local data on the new construction site, and then, based on these local data, fine-tune and optimize the original risk prediction model to make it adapt to the specific characteristics and environmental conditions of the new construction site and improve the applicability and prediction accuracy of the model.Integrate the output result of the risk prediction model, i.e., the risk event level, into the BIM model; generate a dynamic risk heat map that changes over time by overlaying real-time risk level data in the BIM model. Managers can intuitively understand the risk distribution of each area at the construction site through the visualization interface of the BIM model, and promptly discover high-risk areas and potential safety hazards. Combine with the risk heat map in the BIM model to set a risk warning threshold. When the risk level of a certain area exceeds the preset threshold, the system automatically triggers an early warning mechanism and sends a warning message to the managers. Managers can, according to the warning message, quickly organize personnel for on-site investigation and handling, and take corresponding safety measures, such as strengthening protection, adjusting the construction plan, etc., to ensure the safety of the construction site. Refer to Table 2 for the performance comparison results of the risk prediction model used in the present invention.

[0082] Table 2. Performance Comparison Table of Risk Prediction Model

[0083] Model type Accuracy / % Recall / % F1 value Logistic regression 78％ 75％ 0.76 Decision tree 82％ 80％ 0.81 Random forest 88％ 85％ 0.86 Support vector machine 84％ 82％ 0.83 Neural network 89％ 87％ 0.88 The present invention 92％ 90％ 0.91

[0084] In this embodiment, the following steps are further included:

[0085] Divide the risks at the construction site into three levels, including: red, orange, and yellow; generate corresponding warning messages according to the risk event level prediction results output by the risk prediction model, including: risk event, occurrence location, and risk level;

[0086] Push the generated warning messages through the mobile APP of the managers. Among them, different levels of warning messages are distinguished by colors and icons, including: red warning is displayed with a red background and an alarm icon, orange warning is displayed with an orange background and a warning icon, and yellow warning is displayed with a yellow background and a prompt icon;

[0087] In the BIM model, automatically mark and highlight the corresponding dangerous areas according to the occurrence location and influence range of the risk event; among them, the highlighting is carried out with colors corresponding to the risk levels, including: red areas represent high-risk areas, orange areas represent medium-risk areas, and yellow areas represent low-risk areas;

[0088] When the patrol personnel wear AR glasses with augmented reality function and conduct patrols at the construction site, fuse the warning messages with the display interface of the AR glasses in real time; when the patrol personnel are in a dangerous area, corresponding risk point markings automatically pop up on the display interface of the AR glasses, and the marking information includes: risk event, risk level, and risk location;

[0089] Automatically generate corresponding risk investigation work orders according to the warning messages, assign them to the responsible persons and track the progress of handling; the content of the work order includes: risk event, occurrence location, risk level, discovery time;

[0090] Feed the processing results back to the BIM model to automatically update the status of the corresponding risk points in the model; through the visual display of the BIM model, the closed-loop risk investigation and management process follows the PDCA cycle principle.

[0091] Specifically, the risks at the construction site are divided into three levels: red, orange, and yellow. According to the risk event level prediction results output by the risk prediction model, corresponding early warning information is generated, and the early warning information includes: risk event, occurrence location, and risk level. The generated early warning information is pushed through the mobile APP of the management personnel to ensure that the management personnel can obtain the risk information in a timely manner. In the APP, different levels of early warning information are distinguished by colors and icons. The red early warning is displayed with a red background and an alarm icon, the orange early warning is displayed with an orange background and a warning icon, and the yellow early warning is displayed with a yellow background and a prompt icon. In the BIM model, according to the occurrence location and influence range of the risk event, the corresponding dangerous area is automatically marked and highlighted. The highlighting is in the color corresponding to the risk level. The red area represents the high-risk area, the orange area represents the medium-risk area, and the yellow area represents the low-risk area. The inspection personnel wear AR glasses with augmented reality function to conduct inspections at the construction site. The early warning information is fused with the display interface of the AR glasses in real time. When the inspection personnel are in the dangerous area, the corresponding risk point markings automatically pop up on the display interface of the AR glasses, and the marking information includes: risk event, risk level, and risk location. According to the early warning information, corresponding risk investigation work orders are automatically generated, assigned to the responsible persons, and the processing progress is tracked. The work order content includes key information such as risk event, occurrence location, risk level, and discovery time. After the responsible person completes the risk investigation and processing, the processing results are fed back to the BIM model. The BIM model automatically updates the status of the corresponding risk points in the model, and through the visual display of the BIM model, the closed-loop of the risk investigation and management process is realized, following the PDCA cycle principle.

[0092] Please refer to Figure 2 As shown, the present invention is an intelligent early warning method for construction site safety risks based on BIM and big data analysis, including the following steps:

[0093] S1: Deploy a variety of sensors at the construction site to collect IoT data streams in real time, and collect on-site image data in real time through intelligent cameras; transmit the collected IoT data streams and on-site image data to the edge computing layer;

[0094] S2: The preprocessing of the received IoT data stream includes: data screening and cleaning to eliminate interference data and data standardization; the preprocessing of the received on-site image data includes: filtering and denoising and image enhancement; using spatio-temporal alignment technology to map the real-time data to the corresponding spatial coordinates of the BIM model to form a digital twin scenario;

[0095] S3: Use the rules engine to preset safety thresholds for primary early warning; use a machine learning model to analyze time series data and predict collapse risks; identify dangerous behaviors based on on-site image data; use historical accident data to build a risk prediction model, combine adaptive learning and transfer learning techniques, update the risk prediction model in real time, and integrate the prediction results into the BIM model;

[0096] S4: Push early warnings to the manager's APP according to the risk level, and perform visual display in combination with the BIM model and AR glasses; automatically generate inspection work orders, assign them to responsible persons and track the progress of handling, and feedback the handling results to the system to update the risk status of the BIM model, forming a PDCA cycle.

[0097] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent early warning system for construction site safety risks based on BIM and big data analysis, characterized in that, It includes a data acquisition layer, an edge computing layer, a cloud computing layer, and an application layer: The data acquisition layer: Deploy various sensors at the construction site to collect IoT data streams in real time, and collect on-site image data in real time through intelligent cameras; Transmit the collected IoT data streams and on-site image data to the edge computing layer; The edge computing layer: Preprocess the received IoT data streams, including: data screening and cleaning to eliminate interference data and data standardization; preprocess the received on-site image data, including: filtering and denoising, and image enhancement; use spatio-temporal alignment technology to map real-time data to the corresponding spatial coordinates of the BIM model to form a digital twin scenario; The cloud computing layer: Use a rule engine to preset a safety threshold for primary warning; use a machine learning model to analyze time series data and predict the collapse risk; identify dangerous behaviors based on on-site image data; use historical accident data to build a risk prediction model, combine adaptive learning and transfer learning technologies, update the risk prediction model in real time, and integrate the prediction results into the BIM model.

2. An intelligent early warning system for construction site safety risks based on BIM and big data analysis, characterized in that, It also includes: The application layer: Push warnings to the manager's APP according to the risk level, and perform visual display in combination with the BIM model and AR glasses; Automatically generate inspection work orders, assign them to responsible persons and track the processing progress, and after the processing results are fed back to the system, update the risk status of the BIM model to form a PDCA cycle.

3. An intelligent early warning system for construction site safety risks based on BIM and big data analysis according to claim 1, characterized in that, Collect IoT data streams in real time, including: meteorological data, equipment status data, personnel positioning data, and environmental monitoring data. Among them, meteorological data includes: wind speed, temperature, and humidity; equipment status data includes: crane load and voltage; personnel positioning data includes: personnel position information based on UWB ultra-wideband or RFID radio frequency identification; environmental monitoring data includes: dust and noise.

4. An intelligent early warning system for construction site safety risks based on BIM and big data analysis according to claim 1, characterized in that, Use spatio-temporal alignment technology to map real-time data to the corresponding spatial coordinates of the BIM model to form a digital twin scenario, including the following steps: Build a BIM three-dimensional model including construction progress, equipment layout, and personnel movement lines, and associate design changes and construction logs; the BIM three-dimensional model is connected to IoT data streams in real time, including: meteorological data, equipment status, personnel positioning, and environmental monitoring; Establish a unified spatial coordinate system in the BIM three-dimensional model, use the entrance of the construction site or a corner of a building as the origin, and set the directions of the X, Y, and Z axes, which represent the east-west, north-south, and height directions respectively; Convert the actual position information of each sensor at the construction site, including: longitude, latitude, and elevation, to the spatial coordinate system of the BIM three-dimensional model using a three-dimensional coordinate conversion formula; by adopting the Network Time Protocol or a precise time synchronization mechanism, keep the time of each sensor and the system synchronized; map the preprocessed IoT data streams to the corresponding spatial coordinates of the BIM three-dimensional model according to the time series and spatial position. Based on the BIM model and initial sensor data, construct a digital twin scenario consistent with the initial state of the construction site; in the digital twin scenario, display the building structure, equipment layout, and personnel distribution information of the construction site; according to the real-time data collection of various sensors on the Internet of Things data stream, update the digital twin scenario in real time.

5. An intelligent early warning system for construction site safety risks based on BIM and big data analysis according to claim 1, characterized in that, Using a rule engine, preset safety thresholds for primary warnings, including the following steps: The preset safety thresholds include: lack of protection for high-altitude operations, intrusion into dangerous areas, equipment overloading, exceeding limits, and operating beyond the designated area. The rule engine receives the preprocessed Internet of Things data stream in real time and matches the Internet of Things data stream with predefined rules; when the Internet of Things data stream meets the conditions of the preset safety thresholds, the rule engine triggers the warning mechanism for initial warnings.

6. The intelligent early warning system for construction site safety risks based on BIM and big data analysis according to claim 1, characterized in that, Using a machine learning model to analyze time-series data and predict collapse risks, including the following steps: Using the LSTM long short-term memory network model in the machine learning model to analyze time-series data including: the deformation trend of the support structure, and predict collapse risks; Collect historical time-series data on the deformation of the support structure including: displacement, strain, inclination, cracks, temperature, humidity, and rainfall; clean, normalize, or standardize the collected time-series data; label the time-series data including: labels indicating whether a collapse has occurred; Input the labeled historical time-series data on the deformation of the support structure into the LSTM model for training, input the real-time collected time-series data on the deformation of the support structure at the construction site into the trained LSTM model, and output the predicted deformation trend of the support structure and the probability of collapse risks.

7. An intelligent early warning system for construction site safety risks based on BIM and big data analysis according to claim 1, characterized in that, Identify dangerous behaviors based on on-site image data, including the following steps: Based on the on-site image data collected and preprocessed in real time at the construction site, use a convolutional neural network model to detect and identify dangerous behaviors; Collect historical on-site image data sets and label dangerous behavior type tags including: not wearing safety equipment and illegal stacking; Input the labeled historical on-site image data sets into the convolutional neural network model for training; input the on-site image data collected and preprocessed in real time into the trained convolutional neural network model, and the convolutional neural network model identifies whether dangerous behaviors occur in the real-time on-site image data and outputs the categories of the identified dangerous behaviors.

8. An intelligent early warning system for construction site safety risks based on BIM and big data analysis according to claim 1, characterized in that, Construct a risk prediction model using historical accident data, combine adaptive learning and transfer learning techniques, update the risk prediction model in real time, and integrate the prediction results into the BIM model, including the following steps: Collect historical accident data sets and use an ensemble learning method to fuse the output results of the rule engine, LSTM model, and convolutional neural network model and integrate them into the historical accident data sets; label the risk level tags for each accident sample in the historical accident data sets including: red, orange, and yellow; Input the labeled historical accident data sets into the risk prediction model for training; input the preprocessed Internet of Things data stream and on-site image data collected in real time into the trained risk prediction model to output the risk event level; Using an adaptive learning mechanism, according to the newly collected IoT data stream and new on-site image data at the construction site in real time, and using transfer learning technology, transfer the knowledge of the trained risk prediction model to the new construction site scenario; Integrate the output results of the risk prediction model into the BIM model to generate a dynamic risk heat map that changes over time; by overlaying real-time risk level data in the BIM model, display the risk distribution at the construction site in real time.

9. An intelligent early warning system for construction site safety risks based on BIM and big data analysis according to claim 2, characterized in that, It also includes the following steps: Divide the risks at the construction site into three levels, including: red, orange, and yellow; according to the predicted results of the risk event levels output by the risk prediction model, generate corresponding early warning information, including: risk events, occurrence locations, and risk levels; Push the generated early warning information through the mobile APP of the management personnel. Among them, different levels of early warning information are distinguished by colors and icons, including: red early warning is displayed with a red background and an alarm icon, orange early warning is displayed with an orange background and a warning icon, and yellow early warning is displayed with a yellow background and a prompt icon; In the BIM model, automatically mark and highlight the corresponding dangerous areas according to the occurrence location and influence range of the risk event; among them, the highlighting uses colors corresponding to the risk levels, including: red areas represent high-risk areas, orange areas represent medium-risk areas, and yellow areas represent low-risk areas; The patrol personnel wear AR glasses with augmented reality function. When patrolling at the construction site, fuse the early warning information with the display interface of the AR glasses in real time; when the patrol personnel are in the dangerous area, corresponding risk point markings will automatically pop up on the display interface of the AR glasses, and the marking information includes: risk events, risk levels, and risk locations; Automatically generate corresponding risk investigation work orders according to the early warning information, assign them to the responsible persons and track the progress of handling; the content of the work order includes: risk events, occurrence locations, risk levels, and discovery times; Feed back the processing results to the BIM model, and automatically update the status of the corresponding risk points in the model; through the visual display of the BIM model, the closed-loop risk investigation management process follows the PDCA cycle principle.

10. An intelligent early warning method for construction site safety risks based on BIM and big data analysis, using an intelligent early warning system for construction site safety risks based on BIM and big data analysis according to any one of claims 1-9, characterized in that, It includes the following steps: S1: Deploy a variety of sensors at the construction site to collect IoT data streams in real time, and collect on-site image data in real time through intelligent cameras; Transmit the collected IoT data stream and on-site image data to the edge computing layer; S2: Preprocess the received IoT data stream, including: data screening and cleaning to eliminate interference data and data standardization; preprocess the received on-site image data, including: filtering and denoising and image enhancement; use spatio-temporal alignment technology to map real-time data to the corresponding spatial coordinates of the BIM model to form a digital twin scenario; S3: Use a rule engine to preset safety thresholds for primary early warning; use a machine learning model to analyze time series data and predict collapse risks; identify dangerous behaviors based on on-site image data; build a risk prediction model using historical accident data, combine adaptive learning and transfer learning technologies, update the risk prediction model in real time, and integrate the prediction results into the BIM model; S4: Push the warning to the manager's APP according to the risk level, and perform visual display in combination with the BIM model and AR glasses; automatically generate inspection work orders, assign them to the responsible persons and track the processing progress, and feedback the processing results to the system to update the risk status of the BIM model, forming a PDCA cycle.

Images