Big data driven road construction safety risk identification method

Abstract

The invention discloses a big data-driven road construction safety risk identification method, and relates to the technical field of industrial data mining and analysis, and the method comprises the steps: constructing a multi-source data collection network to collect road construction region data, and carrying out the processing to generate a structured time-space correlation data set; a construction behavior mode is determined, and accident association is excavated to construct a dynamic risk assessment channel; and a safety early warning map is generated through channel risk quantitative grading, and a safety control strategy is formulated according to the map. The technical problems that in a traditional road construction safety risk identification means, multi-source data are isolated and split, risk assessment is short of depth and early warning management and control efficiency is low are solved, and accurate quantitative grading and real-time closed-loop management and control of road construction safety risks are achieved. And the accuracy of risk identification and the emergency response efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial data mining and analysis, and in particular to a method for identifying road construction safety risks driven by big data. Background Art

[0002] Road construction safety is crucial to human life, project progress and equipment operation, and accurate identification of construction safety risks is a key link in ensuring safety.

[0003] Traditional road construction safety risk identification relies primarily on single methods such as manual inspections and localized sensor monitoring. These methods have been effective in relatively stable construction environments, but with the increasing complexity of construction scenarios and the demand for intelligent systems, their limitations are becoming increasingly prominent. Firstly, data collection is fragmented, failing to cover all elements of a construction site—people, machinery, the environment, and historical accidents. Furthermore, multi-source data lacks spatiotemporal correlation, making it difficult to capture the dynamic evolution of risks through industrial data mining. Secondly, risk assessment relies on empirical judgment or static thresholds, lacking the support of industrial data analysis. This lack of deep exploration of the hidden connections between construction behavior and accidents makes it difficult to accurately quantify and rank risks such as landslides and mechanical injuries. Summary of the Invention

[0004] This application solves the technical problems of isolated and fragmented multi-source data, lack of depth in risk assessment, and low efficiency of early warning and control in traditional road construction safety risk identification methods. This application builds a multi-source data acquisition network by deploying IoT sensors, video surveillance equipment, and mobile terminals to obtain real-time personnel positioning, machinery status, environmental parameters, and historical accident data at the construction site. The Hado op architecture is used to clean and fuse heterogeneous data, build a dynamic risk assessment model, and realize the quantitative classification of risks such as landslides, mechanical injuries, and falls from heights by correlating and mining the implicit laws between construction behavior patterns and accident characteristics. Specifically, multi-source heterogeneous data sets are obtained and structured through a multi-source data acquisition network. Based on this, construction behavior patterns are identified, accident correlations are mined, and a dynamic risk assessment channel is constructed. Then, a safety warning map is generated for the quantitative risk classification, and finally a safety control strategy is formulated.

[0005] In response to the above technical problems, the present application proposes a big data-driven road construction safety risk identification method, wherein the method includes: constructing a multi-source data acquisition network, performing real-time data acquisition on the road construction area through the multi-source data acquisition network to obtain a multi-source heterogeneous data set, performing structured processing on the multi-source heterogeneous data set to generate a structured spatiotemporal correlation data set; performing construction identification based on the structured spatiotemporal correlation data set, determining the construction behavior pattern, performing accident correlation mining according to the construction behavior pattern, and constructing a dynamic risk assessment channel; performing risk quantitative classification of the road construction area through the dynamic risk assessment channel to generate a safety warning map, wherein the safety warning map contains multiple risk levels, and a safety management and control strategy for the road construction area is formulated according to the multiple risk levels.

[0006] This application proposes one or more technical solutions, which have at least the following technical effects:

[0007] This application builds a multi-source data acquisition network by deploying IoT sensors, video surveillance equipment and mobile terminals to obtain real-time personnel positioning, machinery status, environmental parameters and historical accident data at the construction site. Hadoop architecture is used to clean and fuse heterogeneous data, build a dynamic risk assessment model, and calculate quantitative grading parameters for risks such as landslides, mechanical injuries, and falls from heights by correlating and mining the implicit laws between construction behavior patterns and accident characteristics. Combined with the division of construction cycle stages, analysis of hot spots where personnel gather, and multi-dimensional accident feature vectors, the risk transmission rule set and the dynamic risk level matrix are adjusted to accurately identify the safety risk level of the road construction area, making the road construction safety risk identification results more accurate and reliable, achieving accurate quantitative grading and real-time closed-loop control of road construction safety risks, and improving the accuracy of risk identification and the efficiency of emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 This is a flow chart of the big data-driven road construction safety risk identification method provided in an embodiment of the present application.

[0010] Figure 2 It is a flow chart of constructing a dynamic risk assessment channel in the big data-driven road construction safety risk identification method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] This application provides a big data-driven road construction safety risk identification method to solve the technical problems existing in traditional road construction safety risk identification methods, such as isolated and fragmented multi-source data, lack of depth in risk assessment, and low efficiency in early warning and control.

[0012] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0013] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0014] like Figure 1 As shown, a big data-driven road construction safety risk identification method includes:

[0015] Step A100: construct a multi-source data acquisition network, perform real-time data acquisition on the road construction area through the multi-source data acquisition network to obtain a multi-source heterogeneous data set, perform structured processing on the multi-source heterogeneous data set, and generate a structured spatiotemporal correlation data set.

[0016] In this embodiment of the present application, the multi-source heterogeneous data set includes construction site personnel location data, machinery operation status data, environmental parameter data, and historical accident data sets. Structured processing uses the Hadoop distributed architecture to clean and fuse the heterogeneous data acquired from the multi-source data acquisition network.

[0017] Specifically, a multi-source data acquisition network is constructed to collect multi-source heterogeneous data such as the mechanical operating status, video surveillance streams, and environmental parameters in the road construction area in real time. The key data sets are cleaned and extracted through the Hadoop distributed architecture, and spatiotemporal mapping is performed to generate a structured spatiotemporal correlation data set. The specific steps are detailed in A110-A130.

[0018] Step A200: Perform construction identification based on the structured spatiotemporal correlation data set, determine the construction behavior pattern, perform accident association mining according to the construction behavior pattern, and build a dynamic risk assessment channel.

[0019] Optionally, based on a structured spatiotemporal correlation dataset, data subsets are divided according to the construction period and personnel gathering hotspots are analyzed to identify construction behavior patterns. The accident feature library is constructed by mining association rules based on historical accident feature vectors, and then a dynamic risk assessment channel is constructed based on accident feature indicators. The specific steps are described in detail in A210-A260.

[0020] Step A300: Quantify and grade the risks of the road construction area through the dynamic risk assessment channel to generate a safety warning map. The safety warning map includes multiple risk levels, and a safety management and control strategy for the road construction area is formulated according to the multiple risk levels.

[0021] In one embodiment of the present application, a first risk dimension is constructed by assessing slope displacement risk based on key environmental data, a second risk dimension is constructed through interactive analysis of personnel positioning and mechanical vibration spectrum, quantitative rules are formulated based on the two dimensions and coupled analysis is performed to construct a dynamic risk level matrix containing multiple risk levels. The specific steps are described in detail in A310-A340.

[0022] The construction area is divided into grids according to multiple risk levels, and the spatial grid is dynamically colored to generate RGB color gamut blocks. The color gamut changes are monitored in real time and the risk level is determined to update the blocks and build a safety warning map. The specific steps are detailed in A350-A380.

[0023] Multiple risk levels are matched with construction areas to generate a hierarchical control instruction set (including machinery control, personnel evacuation, and environmental adjustment instructions). The instructions are executed and the feedback effects are measured and controlled. Incremental learning is performed based on the feedback to build a safety control strategy. The specific steps are detailed in A390-1-A390-5.

[0024] Furthermore, step A200 in the method provided in the embodiment of the present application includes:

[0025] A110: Build a multi-source data collection network by deploying machinery monitoring sensors, video surveillance equipment, and environmental monitoring equipment and integrating them with mobile terminals.

[0026] A120: The multi-source data collection network is used to collect the machine operation status data, video surveillance stream data, and environmental parameter data of the road construction area in real time to obtain a multi-source heterogeneous data set.

[0027] A130: Use the Hadoop distributed architecture to clean the machine operation status data, the video surveillance stream data, and the environmental parameter data, extract key data sets, perform spatiotemporal mapping according to the key data sets, and construct the structured spatiotemporal correlation data set.

[0028] In the embodiments of the present application, the Hadoop distributed architecture is the core technical framework for processing multi-source heterogeneous data in road construction safety risk identification.

[0029] Specifically, first, deploy IoT sensors, video surveillance equipment, and mobile terminals to build a multi-source data collection network covering all elements of machine, human, and environment. The details are as follows:

[0030] To collect data on machinery operating status, IoT-based piezoelectric vibration sensors (such as the PC B 352C03) are deployed to monitor machinery vibration frequency in real time. IoT-based magnetoelectric speed sensors (such as the Pepperl+Fuchs VS-012) are used to obtain machinery speed data, and IoT-based strain gauge load sensors (such as the HBM Z6FD1) are used to collect machinery load parameters. These three types of sensors, serving as IoT terminal devices, are embedded in the bearings, drive shafts, and load-bearing structures of construction machinery. These sensors are connected to edge computing gateways via IoT communication protocols such as 5G / Industrial Ethernet, enabling high-frequency (100 samples per second) precise monitoring and real-time data transmission of machinery operating status, forming the core components of the IoT perception layer.

[0031] The video surveillance equipment uses 4K high-definition network cameras (such as Hikvision DS-2CD864XYZ UV model) equipped with 12mm wide-angle lenses, which can provide panoramic coverage of the construction area at a frame rate of 25 frames per second. The monitoring range of a single camera can reach 500m. 2 Through multi-camera networking (deployed at intervals of 20-30 meters), it achieves no-dead-angle coverage of more than 90% of the working space in the construction area, and supports infrared night vision function to adapt to night construction scenes.

[0032] Environmental parameter collection uses a laser displacement meter (such as the Keyence IL-600, with an accuracy of ±1mm) to monitor slope displacement, and SHT30 temperature and humidity sensors (with an accuracy of ±0.5℃ / ±2%RH) are deployed at the four corners of the construction area and high-risk points. Ultrasonic anemometers (such as the NRG 40C, with an accuracy of ±0.1m / s) are used to obtain real-time environmental temperature, humidity, and wind speed data, forming a multi-dimensional environmental monitoring network.

[0033] At the mobile terminal level, construction workers are equipped with smart safety helmets (such as the Shenzhen Upos UP-600) with an integrated Beidou positioning module (Hexin Xingtong UB282, with a positioning accuracy of ±0.5m) to transmit the personnel's location coordinates in real time; industrial-grade touch terminals (such as Advantech TPC-1581H, supporting Bluetooth 5.0) are embedded in the machine operation panel. Operators can send operating instructions through the terminal, and the terminal automatically collects operating data such as machine start and stop, gear switching, etc., forming a human-machine interaction data chain with the personnel positioning data.

[0034] The above-mentioned sensors, equipment and terminals are connected to the edge computing gateway (the core hub device of the multi-source data acquisition network, used to realize data access and communication protocol conversion of terminal devices such as IoT sensors, video surveillance equipment, and mobile terminals) through wired (industrial Ethernet) or wireless (5G / Bluetooth) communication protocols, forming a multi-source heterogeneous original data set containing mechanical operating status (vibration frequency, speed, load and other parameters), video surveillance stream (generating about 2TB of data per day), environmental parameters (slope displacement, temperature and humidity, wind speed and other indicators) and personnel positioning and operation instructions, providing full-factor support for subsequent data processing.

[0035] Finally, the Hadoop distributed architecture is used to traverse the video surveillance stream to extract key frames to analyze personnel positioning and mechanical vibration spectrum. Based on the key frames, the time sequence node matching is determined to extract key environmental data. A three-dimensional coordinate system is constructed as a spatiotemporal index to unify the mapping of various data types and build a data association relationship table. This table is added to the spatiotemporal association dataset to generate a structured spatiotemporal association dataset. The specific steps are described in detail in A131-A135.

[0036] Through the above steps, with the help of the three-dimensional deployment of the multi-source data acquisition network, real-time perception of all elements of the construction scene is achieved, and fragmented data is converted into a structured data set with clear temporal and spatial logical relationships, providing a standardized and correlated data foundation for subsequent construction behavior pattern recognition, accident correlation mining and dynamic risk assessment.

[0037] Furthermore, step A130 in the method provided in the embodiment of the present application includes:

[0038] A131: Traverse the video surveillance stream data based on the Hadoop distributed architecture to extract multiple key frames, perform personnel positioning analysis based on the multiple key frames, and obtain personnel positioning data.

[0039] A132: Perform mechanical vibration analysis based on the multiple key frames and the mechanical operation status data to obtain mechanical vibration spectrum data.

[0040] A133: Determine multiple key timing nodes based on the multiple key frames, match and extract the environmental parameter data according to the multiple key timing nodes, and determine multiple key environmental data.

[0041] A134: Construct a three-dimensional coordinate system for the road construction area, use the three-dimensional coordinate system as a spatiotemporal index, map the personnel positioning data, the mechanical vibration spectrum data, and the multiple key environmental data to the three-dimensional coordinate system for unification, and construct a data association relationship table.

[0042] A135: Add the data association relationship table to the spatiotemporal association dataset.

[0043] In the embodiment of the present application, the time sequence node is a time node determined based on multiple key frames extracted from the video surveillance stream data. The spatiotemporal correlation dataset is a dataset generated by performing structured processing on multi-source heterogeneous datasets.

[0044] Optionally, the video surveillance stream data is first processed in a distributed manner: Using MapReduce tasks within the Hadoop distributed architecture, the continuous video stream is segmented into multiple data blocks (each 1GB in size) and distributed in parallel to cluster nodes for processing. On each node, the video frames are downsampled using the Gaussian Pyramid algorithm from the OpenCV library. By constructing an image pyramid structure, key frames are filtered at a rate of one every ten frames while preserving the key image features, reducing the data size to 10% of the original video. The non-maximum suppression (NMS) algorithm is then used to remove duplicate frames, improving subsequent processing efficiency.

[0045] When analyzing personnel location within the extracted keyframes, the YOLOv5 (You Only LookOnce version 5) object detection algorithm is used to identify the outlines of construction personnel within the keyframes. This algorithm, based on a deep convolutional neural network and pre-trained on the COCO dataset, can detect personnel targets in images in real time and output bounding box coordinates. After obtaining the pixel coordinates, a homography transformation algorithm is used to convert the two-dimensional pixel coordinates into actual geographic coordinates. The intrinsic and extrinsic parameter matrices are obtained using the Zhang Zhengyou calibration method, which was previously performed on the camera. A mapping relationship between the pixel coordinate system and the world coordinate system is established. Combined with the coordinates of the construction site benchmarks, the geographic coordinates of the personnel location data are calculated (with an accuracy of ±0.5 meters), enabling sub-meter real-time tracking of the spatial distribution of construction personnel.

[0046] For machine operating status data, a Pig script in Hadoop aligns keyframe timestamps with the sampling data from the mechanical vibration sensor (sampled 100 times per second) in a time window (50 milliseconds), extracting the raw mechanical vibration signal at the keyframe moment. The Hadoop streaming component uses the Python scientific computing library to perform a fast Fourier transform, converting the time domain signal into frequency domain data. This generates mechanical vibration spectrum data containing vibration frequency, amplitude, and phase. This can identify abnormal signals with frequency deviations exceeding ±10% of the normal operating range of the equipment.

[0047] In terms of environmental parameter processing, based on the timestamp of the key frame (accurate to the millisecond level), the time index function of the HBase database is used to match and extract data such as slope displacement, temperature and humidity, and wind speed at the same moment from the environmental parameter data pool to form a key environmental data set. For example, for a key frame time of 2025-05-26 14:30:00.000, the corresponding extracted slope displacement data is 0.8mm / h (laser displacement meter accuracy ±1mm), the ambient temperature is 32°C (temperature and humidity sensor accuracy ±0.5°C), and the wind speed is 2.1m / s (ultrasonic anemometer accuracy ±0.1m / s).

[0048] After extracting features from multi-source data, a three-dimensional coordinate system is constructed with the upper left corner of the construction area as the origin (0,0,0). The X and Y axes correspond to geographic coordinates (unit: meters), and the Z axis corresponds to altitude (unit: meters). Using Hadoop's Spark component, the longitude and latitude of the personnel location data are converted to planar coordinates. These are spatially mapped to the collection locations of the mechanical vibration spectrum data and the coordinates of the monitoring points for key environmental data, generating data points in the three-dimensional coordinate system. For example, personnel location data (longitude 116.3855°, latitude 39.9044°, altitude 50m) is converted to planar coordinates (X = 100m, Y = 200m, Z = 50m). The corresponding mechanical vibration spectrum data (frequency 55Hz, amplitude 0.3g) and environmental data (temperature 32°C) are annotated at these coordinate points, forming a data association table containing timestamp, spatial coordinates, and data parameters. (Each row of data contains fields such as timestamp, X / Y / Z coordinates, personnel density, mechanical vibration frequency, and ambient temperature.) The table is then added to the spatiotemporal correlation dataset to form a structured multi-source data set.

[0049] The structured spatiotemporal correlation dataset generated through the above steps contains deep correlation relationships between multi-source data, providing a standardized and spatiotemporally aligned data foundation for subsequent construction behavior pattern recognition, accident feature mining, and dynamic risk assessment, significantly improving the accuracy and timeliness of road construction safety risk analysis.

[0050] Further, if Figure 2 As shown, step A200 in the method provided in the embodiment of the present application includes:

[0051] A210: Divide the structured spatiotemporal correlation dataset into stages according to the construction period to determine multiple data subsets.

[0052] A220: Perform cluster analysis based on the personnel positioning data to identify hot spots where people gather.

[0053] A230: Based on the personnel gathering hotspot areas and combining multiple data subsets, behavior labeling is performed to identify and determine construction behavior patterns.

[0054] A240: Retrieve historical accident data record logs, perform accident feature analysis based on the historical accident data record logs, and obtain an accident feature vector.

[0055] A250: Perform association rule mining based on the accident feature vector according to the construction behavior pattern to construct an accident feature library, where the accident feature library includes multiple accident feature indicators.

[0056] A260: Construct the dynamic risk assessment channel based on the multiple accident characteristic indicators.

[0057] In the embodiment of the present application, the personnel gathering hotspot area is an area identified by cluster analysis based on the personnel positioning data. The construction behavior pattern is a pattern determined by behavior annotation identification based on the personnel gathering hotspot area combined with multiple data subsets.

[0058] Specifically, the structured spatiotemporal correlation dataset was first divided into stages according to the construction cycle (such as foundation construction, main structure construction, and decoration construction), forming multiple data subsets. Each stage corresponds to different construction activity characteristics. For example, the foundation construction stage is mainly characterized by mechanical excavation and pile foundation work, while the main structure stage is mainly characterized by rebar binding and formwork installation. Using the Spark component of the Hadoop distributed architecture, the dataset was divided into time windows (such as by week or construction node), and personnel location data, mechanical vibration spectrum data, and environmental parameter data for each stage were extracted to form stage-specific data subsets. For example, mechanical vibration data during the foundation construction stage is mainly concentrated in the low-frequency band (20-50Hz), corresponding to the operating characteristics of equipment such as excavators and pile drivers.

[0059] Then, based on the personnel positioning data (accuracy ±0.5m, from Beidou positioning or video analysis), the specific process of density cluster analysis (such as the DBSCAN algorithm) is as follows: First, the personnel positioning data is converted into a discrete point set (X, Y, Z coordinates) in a three-dimensional coordinate system and divided into data sets with equal time intervals according to the time series (for example, a time window of 10 minutes). Then, the DBSCAN algorithm parameters are set: the spatial neighborhood radius ε = 1m (considering the positioning accuracy of ±0.5m, to ensure that the neighborhood covers the actual gathering range) and the minimum sample number MinPts = 5 (corresponding to the spatial density threshold of 5 people / 10㎡) are calculated for the point set within each time window. For each point, the number of points in its ε neighborhood is counted. If the number of points is ≥ MinPts, it is marked as a core point. All reachable points in the ε neighborhood of the core point constitute a cluster, that is, the personnel gathering hotspot area. For example, within a 1m neighborhood of the coordinates (X=150m, Y=300m, Z=0m), 8, 9, and 7 people were detected in three consecutive time windows (a total of half an hour), all of which met the condition of MinPts=5. It was determined to be a hot spot for continuous gathering of people. The system automatically marked it and triggered a potential risk warning, indicating that there may be omissions in collective work management or illegal gatherings.

[0060] After identifying the hotspot area, the behavior annotation is performed by combining the corresponding multi-category data subsets (such as the mechanical operation status data and environmental parameter data of the area). For example, if the hotspot area has abnormal mechanical vibration frequency (such as exceeding the normal threshold of the equipment by 15%), abnormal environmental dust concentration (threshold> 10mg / m 3 ) and is in the foundation construction stage, it is marked as mechanical excavation and earthwork operation behavior mode; if the hotspot area is located on an aerial work platform (Z = 20m), personnel positioning data shows that multiple people have stayed there for more than 30 minutes, and mechanical vibration data shows that cranes are operating frequently, it is marked as high-altitude installation operation behavior mode.

[0061] Next, retrieve the historical accident data record log, perform time, space, equipment, and environment analysis in turn to obtain the corresponding data, generate accident precursor, process, and impact characteristics through multi-dimensional feature analysis, and construct the accident feature vector through feature fusion. The specific steps are described in detail in A241-A245.

[0062] Association rules are mined between construction behavior patterns and accident feature vectors to obtain implicit association parameters and build a risk transmission rule set. Accident risk labels and trigger conditions are integrated with accident feature vectors to construct accident risk labels. After integration, accident feature indicators are obtained and synchronized to the accident feature library. The specific steps are detailed in A251-A253.

[0063] Finally, the process of constructing a dynamic risk assessment channel based on the multiple accident characteristic indicators is as follows: multiple accident characteristic indicators extracted from the accident characteristic library are used as input, for example, the mechanical injury risk indicator includes trigger conditions such as mechanical vibration frequency > normal threshold 20% (such as the normal range 20-50Hz, the threshold is set to 60Hz), personnel gathering density > 5 people / 10㎡, and the high-altitude fall risk indicator includes wind speed > 5m / s, personnel stay time in high-altitude areas > 30 minutes and other multi-dimensional characteristic parameters.

[0064] These indicators are matched and analyzed with real-time structured spatiotemporal correlation data sets, such as real-time mechanical vibration spectrum data (e.g., the vibration frequency of a certain excavator is 65Hz, exceeding the threshold by 15%), personnel location data (e.g., 8 people gather / 10㎡ at coordinates X=150m, Y=200m, Z=5m), environmental parameters (e.g., wind speed 6m / s), and other real-time data. Through association rule algorithms (e.g., Apriori algorithm, with a minimum support of 20% and a confidence level of 80%) and risk transmission rule sets (e.g., IF mechanical vibration frequency > 60Hz AND personnel density > 5 people / 10m), the risk of the risk is analyzed. 2 THEN mechanical injury risk level is improved), quantitative assessment and classification of real-time risks in the construction area are carried out.

[0065] The output results include specific risk levels (such as low, medium, and high), risk types (such as mechanical injuries, falls from heights), and potential accident types. For example, when real-time data matches a vibration frequency of 65Hz and a human density of 8 people / 10m 2 When a high risk level for mechanical injury is triggered, a red warning signal is output, and historical mechanical collision accident cases with similar characteristics are linked. This process achieves a precise quantitative assessment of construction risks by dynamically matching standardized characteristic indicators with real-time data, providing core data support for the generation of safety warning maps and improving the timeliness and accuracy of risk identification.

[0066] By dividing the structured spatiotemporal correlation data set into stages, analyzing personnel clusters and labeling behaviors, we determine the construction behavior patterns, build an accident feature database based on historical accident feature analysis, and construct a dynamic risk assessment channel based on accident feature indicators, thus achieving the technical effect of accurate identification and dynamic assessment of construction risks.

[0067] Furthermore, step A240 in the method provided in the embodiment of the present application includes:

[0068] A241: Perform spatiotemporal analysis based on the historical accident data record log to obtain accident spatiotemporal data.

[0069] A242: Perform equipment analysis based on the historical accident data record log to obtain accident equipment status data.

[0070] A243: Perform environmental analysis based on the historical accident data record log to obtain accident environment monitoring data.

[0071] A244: Perform multi-dimensional feature analysis based on the accident spatiotemporal data, the accident equipment status data, and the accident environment monitoring data to generate accident precursor features, accident process features, and accident impact features.

[0072] A245: Fusing the accident precursor features, the accident process features, and the accident impact features to construct the accident feature vector.

[0073] In the embodiment of the present application, the historical accident data record log is a record log containing information such as spatiotemporal data of historical accidents, equipment status data, and environmental monitoring data.

[0074] Specifically, first, the historical accident data record log is subjected to spatiotemporal analysis: by extracting the timestamp (accurate to seconds) and Beidou positioning coordinates (accuracy ±0.5m) in the log, the spatiotemporal coordinate point of the accident is constructed, such as 14:20:00 on October 5, 2024, coordinates (X=200m, Y=300m, Z=15m), forming the accident spatiotemporal data and clarifying the specific time and spatial location of the accident.

[0075] Next, equipment analysis is performed, extracting operational status data of the machinery involved in the accident from the logs, such as the vibration spectrum of the tower crane (sampling frequency 100Hz) and the operating speed curve of the elevator. Vibration data is analyzed using Fourier transform. If the mechanical vibration frequency is consistently above the normal threshold by 20% (e.g., a normal range of 20-50Hz, with an abnormal value of 60Hz) for the five minutes before the accident, the equipment is flagged as abnormal, generating the accident equipment status data.

[0076] Environmental analysis is then performed, simultaneously extracting environmental parameters before and after the accident, such as temperature and humidity (accuracy of ±0.5°C / ±2%RH), wind speed (accuracy of ±0.1m / s), and slope displacement (accuracy of ±1mm). If the ambient temperature at the time of the accident reached 38°C (exceeding the high-temperature working threshold of 35°C) and the wind speed reached 8m / s (exceeding the high-altitude working safety threshold of 5m / s), these data are recorded as accident environmental monitoring data.

[0077] Next, based on the above three types of data, multi-dimensional feature analysis is performed:

[0078] Accident precursor features: Extract abnormal data within 1 hour before the accident, such as a sudden increase in the density of people gathering (from 2 people / 10m 2 Increased to 8 people / 10m 2 ), equipment vibration frequency continues to increase, environmental parameters exceed the limit, etc.

[0079] Accident process characteristics: Capture key data at the moment of the accident, such as the peak value of mechanical vibration (assuming it is 100g), the trajectory of the person falling (technical personnel in the field can analyze it through video key frames), and sudden changes in environmental parameters (such as a sudden increase in wind speed to 12m / s).

[0080] Accident impact characteristics: record the degree of equipment damage after the accident (such as the deformation of the tower crane steel structure), the scope of environmental damage (such as the slope displacement increased by 50mm), etc.

[0081] Finally, during the construction of the accident feature vector, principal component analysis (PCA) is used for dimensionality reduction and feature combination. The specific process of PCA is as follows: First, multi-source feature data, including the spatiotemporal accident data (such as the three-dimensional coordinates and timestamps of the accident), vibration frequency anomalies (such as mechanical vibration spectrum data), and the number of temperature and humidity threshold violations and wind speed peaks in the environmental dimension, are normalized to eliminate dimensional differences (for example, coordinate values ​​are normalized to the range [0, 1] and vibration frequencies are converted to standard fractions). Next, the covariance matrix of the normalized dataset is calculated, and eigenvalues ​​and eigenvectors are extracted through eigenvalue decomposition. Principal components are selected based on their cumulative variance contribution (for example, a threshold of 85%), retaining key feature combinations that explain the majority of the data variance. For example, the first five principal components extracted from the multi-dimensional original features contribute a cumulative 88% of the variance, thus reducing the high-dimensional feature space to 5 dimensions. Finally, the selected principal components are linearly combined according to the weights of the eigenvectors to generate an accident feature vector containing multidimensional comprehensive features, such as [principal component 1 (comprehensive time and space and vibration frequency), principal component 2 (comprehensive wind speed and personnel density)], so as to achieve efficient characterization of accident precursors, processes and impact characteristics.

[0082] Through the above steps, big data technology was used to achieve a comprehensive quantitative representation of accident characteristics, providing standardized input for subsequent correlation mining between construction behavior patterns and accident risks, significantly improving the accuracy of the risk assessment model and the comprehensiveness of accident cause analysis.

[0083] Furthermore, step A250 in the method provided in the embodiment of the present application includes:

[0084] A251: Mining the construction behavior pattern and the accident feature vector according to association rules to obtain implicit association parameters, and constructing a risk transmission rule set based on the implicit association parameters.

[0085] A252: Integrate the risk transmission rule set in combination with the accident feature vector to construct multiple accident risk labels, and calibrate multiple accident triggering conditions based on the multiple accident risk label tables.

[0086] A253: Integrate the multiple accident risk tags with the multiple accident trigger conditions to obtain multiple accident feature indicators, and synchronize the multiple accident feature indicators to the accident feature library.

[0087] In the embodiments of this application, implicit correlation parameters are potential correlation parameters obtained by mining construction behavior patterns and accident feature vectors according to association rules, reflecting the implicit regularity between construction behavior patterns and accident characteristics. The risk transmission rule set is a set of rules constructed based on implicit correlation parameters.

[0088] Specifically, first, the Apriori association rule algorithm is used to mine construction behavior patterns (such as high-altitude work and intensive mechanical operation) and accident feature vectors (including wind speed, mechanical vibration frequency, and personnel density). By setting the minimum support (such as 20%) and confidence (such as 80%), implicit association parameters are identified. For example, when the personnel density is greater than 5 people / 10m 2 When the mechanical vibration frequency is greater than 60 Hz, the confidence level of association with mechanical injury accidents reaches 85%. This type of parameter combination is extracted as a hidden correlation parameter.

[0089] Based on implicit correlation parameters, a risk transmission rule set is constructed. For example, for high-altitude work behavior patterns, if the frequency of wind speeds exceeding 5m / s and personnel not wearing safety belts in the accident feature vector reaches 30%, the rule is generated: High-altitude work + wind speed exceeding the limit + not wearing protective equipment → risk of falling from height. The rule set uses an IF-THEN logic structure and covers over 80% of historical accident scenarios.

[0090] Subsequently, the risk transmission rule set is integrated with the accident feature vector to construct an accident risk label. For example, the rule mechanical vibration frequency > normal threshold 20% + personnel gathering density > 5 people / 10m 2 A high-collision risk tag is assigned to the machinery area, with trigger conditions calibrated as follows: vibration frequency exceeding the limit for 30 consecutive minutes and a sudden increase in occupancy density. Each tag is associated with 3-5 key parameter thresholds, such as vibration frequency threshold and occupancy density change rate.

[0091] Finally, accident risk tags are integrated with trigger conditions to generate accident characteristic indicators. For example, the mechanical injury risk indicator combines tags such as abnormal mechanical vibration, crowding, and non-standard operation with corresponding trigger conditions to form a structured indicator with multiple characteristics. These indicators are synchronized to the accident characteristic database via the Hadoop distributed architecture, enabling standardized storage and rapid retrieval of historical accident characteristics. The annotated accident characteristic records in the database support real-time risk matching.

[0092] Through the above steps, association rule mining technology is used to transform fuzzy construction behaviors and accident characteristics into quantifiable risk rules and characteristic indicators, and an accident characteristic library covering multi-dimensional risk factors is constructed. This provides an accurate feature matching basis for the dynamic risk assessment channel and improves the intelligent level of road construction safety risk identification.

[0093] Furthermore, step A300 in the method provided in the embodiment of the present application includes:

[0094] A310: Based on the multiple key environmental data synchronized to the dynamic risk assessment channel, a slope displacement risk assessment is performed on the road construction area to construct a first risk dimension.

[0095] A320: Interactively analyze the personnel positioning data and the mechanical vibration spectrum data to construct a second risk dimension.

[0096] A330: Formulate risk quantification rules based on the first risk dimension and the second risk dimension.

[0097] A340: Perform risk coupling analysis on the road construction area based on the risk quantification rule, and construct a dynamic risk level matrix, where the dynamic risk level matrix includes the multiple risk levels.

[0098] In the embodiment of the present application, slope displacement refers to the positional movement of the slope rock and soil caused by natural or human factors, which is usually manifested as deformation or sliding of the slope surface or internal rock and soil in a certain direction.

[0099] In one embodiment, when constructing multiple risk levels, risk assessment dimensions are first constructed based on multi-source data:

[0100] The first risk dimension (landslide risk): Dynamic risk assessment channels synchronize key environmental data (such as slope displacement monitoring data, with an accuracy of ±1mm and a sampling frequency of 1 time per hour). When the slope displacement rate exceeds a threshold (such as 0.5mm / h), the slope displacement risk level is determined to be elevated. For example, if the slope displacement rate in a certain area reaches 1.2mm / h for three consecutive hours, a landslide risk warning will be triggered, corresponding to a high risk level in the first dimension.

[0101] The second risk dimension (mechanical injury risk): interactive analysis of personnel positioning data (accuracy ±0.5m) and mechanical vibration spectrum data (such as vibration frequency, amplitude). When there are hot spots where people gather (such as 5 people / 10m 2 ) overlaps with the abnormal mechanical vibration frequency (exceeding 20% ​​of the normal threshold), the mechanical injury risk level is determined to be increased. For example, the human density at the coordinate (X=150m, Y=200m, Z=0m) is 8 people / 10m 2, and the nearby mechanical vibration frequency is 65Hz, triggering a high risk level in the second dimension.

[0102] The height-fall risk dimension uses data from personnel altitude (using Beidou GPS's Z-axis coordinates) and ambient wind speed data (accuracy ±0.1 m / s). When a person is at high altitude (Z ≥ 10 m) and wind speeds > 5 m / s, combined with safety equipment detection data (such as a smart helmet failing to provide a wear signal), a height-fall risk dimension is constructed. For example, if a person spends 40 minutes at Z = 15 m, with a wind speed of 6 m / s and no safety belt signal detected, they are considered at high risk of falling from a height.

[0103] Subsequently, risk quantification rules were formulated: each dimension was divided into levels (low, medium, and high), such as slope displacement ≤0.5mm / h is low risk, 0.5-1.0mm / h is medium risk, and >1.0mm / h is high risk; mechanical vibration frequency abnormality ≤10% is low risk, 10%-20% is medium risk, and >20% is high risk; for the risk of falling from height, wind speed ≤3m / s and Z<10m is low risk, 3-5m / s or Z≥10m is medium risk, and >5m / s and Z≥10m is high risk.

[0104] Finally, a risk coupling analysis is conducted: Based on the risk quantification rules, the risk levels of the three dimensions are matrix-combined to construct a dynamic risk level matrix containing 9 combinations (such as low + low + low = low risk, high + high + high = extremely high risk).

[0105] Through multi-dimensional data fusion, quantitative rule formulation and coupled analysis, multi-dimensional quantitative grading of risks such as landslides, mechanical injuries, and falls from heights has been achieved. The constructed dynamic risk level matrix can reflect the comprehensive risk status of the construction area in real time, provide a scientific basis for the generation of safety warning maps, and significantly improve the comprehensiveness and accuracy of road construction safety risk assessments.

[0106] Furthermore, step A300 in the method provided in the embodiment of the present application includes:

[0107] A350: Divide the road construction area into grids according to the multiple risk levels to obtain multiple spatial grids.

[0108] A360: Dynamically color the multiple spatial grids based on the multiple risk levels to obtain multiple RGB color gamut grid blocks.

[0109] A370: Real-time monitoring is performed on the multiple RGB color gamut grid blocks. When color gamut changes occur in the multiple RGB color gamut grid blocks, risk assessment is performed on the multiple spatial grids to generate risk assessment levels.

[0110] A380: Update the multiple RGB color gamut blocks based on the risk determination level to construct the safety warning map.

[0111] Optionally, when generating a safety warning map, the road construction area is first gridded: the construction area is divided into 10m×10m spatial grids based on geographic coordinates (technical personnel in this field can adjust the accuracy based on the scale of the construction). Each grid corresponds to a unique area in the three-dimensional coordinate system (X, Y, Z). Grids with a Z axis ≥ 10m are automatically marked as high-altitude work areas for targeted assessment of high-altitude fall risks. For example, the construction area of ​​a certain viaduct is divided into 500 spatial grids, of which high-altitude grids with a Z axis ≥ 10m account for more than 30%, requiring key monitoring of personnel positioning and wind speed data.

[0112] Next, perform dynamic coloring: assign RGB color values ​​to each grid based on the three risk levels (low, medium, and high) of landslide, mechanical injury, and height fall. For example:

[0113] Landslide risk: low risk (RGB: 255, 200, 150), medium risk (RGB: 255, 150, 100), high risk (RGB: 255, 50, 50); Mechanical injury risk: low risk (RGB: 255, 255, 150), medium risk (RGB: 255, 255, 50), high risk (RGB: 255, 200, 0); Height fall risk: low risk (RGB: 150, 200, 255), medium risk (RGB: 100, 150, 255), high risk (RGB: 50, 100, 255).

[0114] By superimposing the color values ​​of different risk dimensions, a comprehensive risk color gamut grid is generated. For example, if a grid has both a medium risk of landslide (red component 255, 150, 100) and a high risk of falling from height (blue component 50, 100, 255), the mixed color value is (255, 150, 255), corresponding to the purple system, which intuitively reflects the superposition status of multi-dimensional risks.

[0115] During the real-time monitoring and risk assessment stage, when a color gamut change is detected in the RGB color gamut grid, the target change grid is determined and its spatial grid is traversed to determine the risk change parameters. Combined with the RGB color gamut-risk list, it is determined whether there is a risk jump. If so, a risk warning signal is generated and added to the risk assessment level. The specific steps are described in detail in A371-A374.

[0116] Finally, the safety warning map is updated based on the risk assessment level results: the system traverses all grids every 5 minutes and detects changes in the color gamut. For example, due to a sudden increase in wind speed to 7m / s and the presence of personnel in a high-altitude grid for more than 40 minutes, the color gamut changes from light blue to dark blue, and the risk assessment level is updated to a high risk of falling from a height, and the grid is highlighted in the map. Technicians in this field can view the risk distribution of each area in real time through the map, such as red areas (high risk of landslides), yellow areas (high risk of mechanical injuries), blue areas (high risk of falling from a height), and mixed color areas (multi-dimensional risk superposition).

[0117] Through the above steps, the safety warning map integrates multi-dimensional risk data in a visual manner, realizes real-time dynamic display of construction area risks, converts abstract risk levels into intuitive color signals, solves the problem of lack of spatial visualization in traditional risk assessment, and significantly improves the response speed and decision-making efficiency of construction safety management.

[0118] Furthermore, step A370 in the method provided in the embodiment of the present application includes:

[0119] A371: When color gamut changes occur among the plurality of RGB color gamut grid blocks, determine a target change grid block.

[0120] A372: Traverse the multiple spatial grids according to the target change grid block to determine the risk change parameter.

[0121] A373: Retrieve the RGB color gamut-risk list, and determine whether there is a risk jump in the target change grid block based on the risk change parameter and the RGB color gamut-risk list. When there is a risk jump in the target change grid block, generate a risk warning signal.

[0122] A374: Add the risk warning signal to the risk determination level.

[0123] In the embodiment of the present application, the RGB color gamut-risk list is a pre-generated mapping relationship table that stores the correspondence between RGB color values ​​and risk levels, risk types, and trigger conditions.

[0124] In one embodiment, when monitoring RGB color gamut grid blocks in real time, the RGB color values ​​of each grid are continuously compared with historical data through an image recognition algorithm (such as OpenCV's color space conversion technology). When it is detected that the color value of a certain color gamut grid block deviates from a preset range, for example, the RGB value changes from (255, 200, 150) to (255, 150, 100), the grid block is determined to be a target change grid block, triggering a risk analysis process.

[0125] For the target change grid, the system automatically traverses its corresponding spatial grid and extracts real-time data as risk change parameters. For example, if the grid corresponds to a high-altitude work area (Z ≥ 10m), it retrieves personnel positioning data (such as a person staying at Z = 15m for 45 minutes), wind speed data (current wind speed 6.2m / s), mechanical vibration spectrum data (vibration frequency of a nearby crane 65Hz), etc. At the same time, it queries the historical risk level of the grid (such as the medium risk of mechanical injury in the previous period) and calculates the parameter change amplitude (such as the wind speed increased by 24% compared to the previous hour).

[0126] The system then retrieves a pre-built RGB color space-risk list, which stores the mapping between colors, risk levels, and parameter thresholds:

[0127] The blue line corresponds to the risk of falling from height, with a risk threshold of wind speed ≤ 3m / s and Z < 10m for low risk, 3-5m / s or Z ≥ 10m for medium risk, and > 5m / s and Z ≥ 10m for high risk; the red line corresponds to the risk of landslide, with a risk threshold of slope displacement ≤ 0.5mm / h for low risk, 0.5-1.0mm / h for medium risk, and > 1.0mm / h for high risk; the yellow line corresponds to the risk of mechanical injury, with a risk threshold of personnel density > 5 people / 10m 2 An abnormal mechanical vibration frequency of ≤10% is considered low risk, 10%-20% is considered medium risk, and >20% is considered high risk. The high, medium, and low risks correspond to dark, standard, and light colors, respectively.

[0128] The extracted risk change parameters are matched with the thresholds in the list. If the parameters meet the conditions for a higher risk level (such as the current wind speed of 6.2m / s exceeds the high risk threshold of 5m / s for falling from height), it is determined to be a risk jump, and a corresponding risk warning signal is generated (such as a red flashing alarm). The change in risk level before and after the jump is recorded (such as the risk of falling from height is upgraded to high risk).

[0129] Finally, risk warning signals are integrated into the risk assessment level and displayed in real time on a safety warning map. For example, the color of a target grid block is updated from blue (medium risk) to dark blue (high risk) and highlighted on the map. Construction managers can instantly access this risk change information through the map and initiate appropriate control measures (such as speed limits in high-altitude work areas and increased safety inspection frequency).

[0130] By converting abstract risk levels into visual color signals and combining them with real-time analysis of multi-source data, the lag problem of traditional risk assessment has been solved, the response speed and accuracy of construction safety management have been significantly improved, and data support has been provided for timely intervention in risks.

[0131] Furthermore, step A300 in the method provided in the embodiment of the present application includes:

[0132] A390-1: Match the multiple risk levels with the road construction area to generate a hierarchical management and control instruction set, which includes machinery control instructions, personnel evacuation instructions, and environmental adjustment instructions.

[0133] A390-2: Execute the mechanical control instructions to perform measurement and control feedback to obtain a mechanical control effect.

[0134] A390-3: Execute the personnel evacuation instructions to conduct measurement and control feedback to obtain the personnel evacuation effect.

[0135] A390-4: Execute the environmental adjustment instructions to perform measurement and control feedback to obtain the environmental adjustment effect.

[0136] A390-5: Conduct incremental learning based on the mechanical control effect, the personnel evacuation effect, and the environmental adjustment effect to construct the safety management and control strategy.

[0137] In one embodiment, the multiple risk levels output by the dynamic risk assessment channel (such as low, medium, and high levels for landslides, mechanical injuries, and falls from height) are first accurately matched to the spatial grids of the construction area. For example, when a grid is determined to have a high risk of landslide (RGB color values ​​255, 50, 50, corresponding to a slope displacement rate > 1.0 mm / h), the system automatically generates a hierarchical control instruction set for that area:

[0138] Mechanical control instructions: suspend the operation of pile foundation machinery in the area, trigger the equipment braking program, and send a forced stop signal to the operator; personnel evacuation instructions: send an evacuation alarm to personnel in the area through mobile terminals, requiring them to evacuate to a safe assembly point (≥50m away from the risk area) within 5 minutes; environmental adjustment instructions: start slope reinforcement equipment (such as anchor bolters) and increase the deployment monitoring frequency from 1 time / hour to 1 time / 10 minutes.

[0139] During the execution of the command, the system collects measurement and control feedback data in real time:

[0140] Mechanical control effect: Monitoring mechanical vibration spectrum data confirmed that the vibration frequency dropped from 65Hz to the normal range (20-50Hz) after the equipment was shut down; Personnel evacuation effect: Through personnel positioning data statistics, personnel in the area completed evacuation within 4 minutes and 30 seconds, with an evacuation efficiency of 100%; Environmental adjustment effect: The slope displacement rate dropped from 1.2mm / h to 0.6mm / h, and the risk level dropped from high to medium.

[0141] Based on this feedback data, we utilize incremental learning algorithms to optimize control strategies. If a certain risk triggers the same command multiple times with significant results, the priority of that command is increased. If the results are poor, we automatically adjust parameters (such as extending the evacuation time threshold to 8 minutes) or combine them with other commands (such as simultaneously initiating mechanical control and environmental adjustments). Through continuous iteration, we build an adaptive library of safety control strategies. For example, for high-risk areas for falls (RGB: 50, 100, 255), we develop a combined strategy that includes forced cessation of operations when wind speeds exceed 8m / s, automatic locking of high-altitude equipment, and drone inspections.

[0142] Through multi-dimensional data fusion and dynamic strategy adjustment, the accuracy and effectiveness of construction safety management have been significantly improved. Ultimately, through the dynamic coupling of quantified risk levels and instruction execution, the technical effect of reducing accident rates and ensuring the safety of construction workers has been achieved.

[0143] In summary, the big data-driven road construction safety risk identification method provided in the embodiments of the present application has the following technical effects:

[0144] This application builds a multi-source data acquisition network by deploying machinery monitoring sensors, video monitoring equipment, environmental monitoring equipment and mobile terminals in road construction areas, and collects multi-source heterogeneous data such as machinery operation status, video monitoring streams, and environmental parameters in real time. The Hadoop distributed architecture performs data cleaning, key data set extraction, and time-space mapping to generate a structured spatiotemporal correlation data set. Based on this data set, the construction cycle is divided into stages, and behavioral patterns are identified through the aggregation analysis of personnel positioning data. An accident feature library is constructed by combining the feature analysis of historical accident data, and then the implicit laws of construction behavior and accident characteristics are mined to build a dynamic risk assessment channel. Through multi-dimensional risk quantification, the application also builds a multi-source data acquisition network by deploying machinery monitoring sensors, video monitoring equipment, environmental monitoring equipment and mobile terminals in road construction areas, and collects multi-source heterogeneous data such as machinery operation status, video monitoring streams, and environmental parameters in real time. The application also performs data cleaning, key data set extraction, and time-space mapping to generate a structured spatiotemporal correlation data set. The application also divides the construction cycle into stages, and identifies behavioral patterns through the aggregation analysis of personnel positioning data. The application also builds an accident feature library by combining the feature analysis of historical accident data. Through the coupling analysis, a dynamic level matrix including risks such as landslide, mechanical injury and falling from height is generated, which is synchronized to the safety warning map for grid coloring and real-time monitoring. The risk jump is determined based on the change of color domain and the warning is updated. Finally, based on the risk level matching, a hierarchical control instruction set such as mechanical control, personnel evacuation and environmental adjustment is generated. The safety control strategy is optimized through measurement and control feedback and incremental learning, so as to realize the accurate identification, quantitative classification and dynamic control of road construction safety risks, improve the scientificity and effectiveness of construction safety management, and achieve the technical effect of accurate quantitative classification and real-time closed-loop control of road construction safety risks, improving the accuracy of risk identification and the efficiency of emergency response.

[0145] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0146] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A big data-driven road construction safety risk identification method, characterized by: The method comprises: Constructing a multi-source data acquisition network, performing real-time data acquisition on the road construction area through the multi-source data acquisition network to obtain a multi-source heterogeneous data set, and performing structured processing on the multi-source heterogeneous data set to generate a structured spatiotemporal correlation data set; Perform construction identification based on the structured spatiotemporal correlation data set, determine the construction behavior pattern, conduct accident correlation mining according to the construction behavior pattern, and build a dynamic risk assessment channel; The dynamic risk assessment channel is used to quantitatively grade the risks of the road construction area and generate a safety warning map. The safety warning map includes multiple risk levels, and a safety management and control strategy for the road construction area is formulated according to the multiple risk levels.

2. The big data driven road construction safety risk identification method according to claim 1, characterized in that: A multi-source data acquisition network is constructed, and real-time data acquisition is performed on the road construction area through the multi-source data acquisition network to obtain a multi-source heterogeneous data set. The multi-source heterogeneous data set is structured and processed to generate a structured spatiotemporal correlation data set. The method includes: Build a multi-source data acquisition network by deploying machinery monitoring sensors, video surveillance equipment, and environmental monitoring equipment for mobile terminal integration; The multi-source data collection network collects in real time the machine operation status data, video surveillance stream data, and environmental parameter data of the road construction area to obtain a multi-source heterogeneous data set; The Hadoop distributed architecture is used to clean the machine operation status data, the video surveillance stream data, and the environmental parameter data, extract key data sets, perform spatiotemporal mapping according to the key data sets, and construct the structured spatiotemporal correlation data set.

3. The big data driven road construction safety risk identification method according to claim 2, characterized in that: The Hadoop distributed architecture is used to clean the machine operation status data, the video surveillance stream data, and the environmental parameter data, extract key data sets, perform spatiotemporal mapping according to the key data sets, and construct the structured spatiotemporal correlation data set. The method includes: Traversing the video surveillance stream data based on the Hadoop distributed architecture to extract multiple key frames, performing personnel positioning analysis based on the multiple key frames to obtain personnel positioning data; Performing mechanical vibration analysis based on the multiple key frames in combination with the mechanical operation status data to obtain mechanical vibration spectrum data; Determine a plurality of key timing nodes based on the plurality of key frames, match and extract the environmental parameter data according to the plurality of key timing nodes, and determine a plurality of key environmental data; Constructing a three-dimensional coordinate system for the road construction area, using the three-dimensional coordinate system as a spatiotemporal index, mapping the personnel positioning data, the mechanical vibration spectrum data, and the multiple key environmental data to the three-dimensional coordinate system for unification, and constructing a data association relationship table; Add the data association relationship table to the spatiotemporal association dataset.

4. The big data driven road construction safety risk identification method according to claim 3, characterized in that: Based on the structured spatiotemporal correlation dataset, construction identification is performed to determine the construction behavior pattern, accident association mining is performed according to the construction behavior pattern, and a dynamic risk assessment channel is constructed. The method includes: Dividing the structured spatiotemporal correlation dataset into stages according to the construction period to determine multiple data subsets; Performing cluster analysis based on the personnel location data to identify hotspots where people gather; Based on the personnel gathering hotspots and combining multiple data subsets, behavior annotation is performed to identify and determine construction behavior patterns; Retrieving historical accident data record logs, performing accident feature analysis based on the historical accident data record logs, and obtaining an accident feature vector; Performing association rule mining based on the accident feature vector according to the construction behavior pattern to construct an accident feature library, wherein the accident feature library includes multiple accident feature indicators; The dynamic risk assessment channel is constructed based on the multiple accident characteristic indicators.

5. The big data driven road construction safety risk identification method according to claim 4, characterized in that: Retrieving historical accident data record logs, performing accident feature analysis based on the historical accident data record logs, and obtaining an accident feature vector, the method includes: Performing spatiotemporal analysis based on the historical accident data record log to obtain accident spatiotemporal data; Performing equipment analysis based on the historical accident data record log to obtain accident equipment status data; Perform environmental analysis based on the historical accident data record log to obtain accident environment monitoring data; Perform multi-dimensional feature analysis based on the accident spatiotemporal data, the accident equipment status data, and the accident environment monitoring data to generate accident precursor features, accident process features, and accident impact features; The accident precursor features, the accident process features, and the accident impact features are fused to construct the accident feature vector.

6. The big data driven road construction safety risk identification method according to claim 4, characterized in that: According to the construction behavior pattern, association rules are mined based on the accident feature vector to construct an accident feature library, wherein the accident feature library includes multiple accident feature indicators. The method includes: Mining the construction behavior pattern and the accident feature vector according to association rules to obtain implicit association parameters, and constructing a risk transmission rule set based on the implicit association parameters; Integrate the risk transmission rule set with the accident feature vector to construct multiple accident risk labels, and calibrate multiple accident trigger conditions according to the multiple accident risk label tables; The multiple accident risk tags are integrated with the multiple accident triggering conditions to obtain multiple accident feature indicators, and the multiple accident feature indicators are synchronized to the accident feature library.

7. The big data driven road construction safety risk identification method according to claim 3, characterized in that: The process of constructing multiple risk levels includes: Performing a slope displacement risk assessment on the road construction area based on the plurality of key environmental data synchronized to the dynamic risk assessment channel to construct a first risk dimension; Interactively analyzing the personnel location data and the mechanical vibration spectrum data to construct a second risk dimension; Formulate risk quantification rules based on the first risk dimension and the second risk dimension; Based on the risk quantification rules, a risk coupling analysis is performed on the road construction area to construct a dynamic risk level matrix, which includes the multiple risk levels.

8. The big data driven road construction safety risk identification method according to claim 7, characterized in that: The dynamic risk assessment channel is used to quantitatively grade the risks in the road construction area and generate a safety warning map. The method includes: Dividing the road construction area into grids according to the multiple risk levels to obtain multiple spatial grids; Dynamically coloring the plurality of spatial grids based on the plurality of risk levels to obtain a plurality of RGB color gamut grid blocks; Performing real-time monitoring on the multiple RGB color gamut grid blocks, and when the multiple RGB color gamut grid blocks have color gamut changes, performing risk assessment on the multiple spatial grids and generating risk assessment levels; The plurality of RGB color gamut blocks are updated based on the risk determination level to construct the safety warning map.

9. The big data driven road construction safety risk identification method according to claim 8, characterized in that: The multiple RGB color gamut grid blocks are monitored in real time. When the multiple RGB color gamut grid blocks have color gamut changes, risk assessment is performed on the multiple spatial grids to generate risk assessment levels. The method includes: When the color gamut of the plurality of RGB color gamut grid blocks changes, determining a target change grid block; Traversing the plurality of spatial grids according to the target change grid block to determine a risk change parameter; Retrieving an RGB color gamut-risk list, determining whether the target change grid block has a risk jump based on the risk change parameter in combination with the RGB color gamut-risk list, and generating a risk warning signal when the target change grid block has a risk jump; The risk warning signal is added to the risk determination level.

10. The big data driven road construction safety risk identification method according to claim 1, characterized in that: Develop a safety management strategy for road construction areas based on the multiple risk levels described, including: Matching the multiple risk levels with the road construction area to generate a hierarchical control instruction set, the hierarchical control instruction set including machinery control instructions, personnel evacuation instructions, and environmental adjustment instructions; Executing the mechanical control instructions to perform measurement and control feedback to obtain a mechanical control effect; Execute the personnel evacuation command to perform measurement and control feedback to obtain the personnel evacuation effect; Executing the environmental adjustment instructions to perform measurement and control feedback to obtain environmental adjustment effects; Incremental learning is performed based on the mechanical control effect, the personnel evacuation effect, and the environmental adjustment effect to construct the safety management and control strategy.

Images