An artificial intelligence-based construction safety monitoring method and system

Abstract

The application relates to the technical field of safety monitoring, in particular to a building construction safety monitoring method and system based on artificial intelligence, which comprises the following steps: acquiring multi-source heterogeneous data of a construction site, performing distributed feature extraction on the multi-source heterogeneous data by adopting a federal learning framework, and generating site state representation data associated with time and space; performing risk prediction on the site state representation data by using a preset dynamic risk prediction model, outputting a multi-level risk prediction result, and constructing the preset dynamic risk prediction model based on a construction safety knowledge graph and a space-time graph neural network; triggering an adaptive feedback mechanism according to a risk level corresponding to the prediction result, generating visual warning information and equipment control instructions, and linking the construction site control system to execute emergency response operations. The problems of traditional monitoring methods, such as complicated data processing, insufficient real-time performance, high cost and lack of prediction ability, are solved.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, specifically to a construction safety monitoring method and system based on artificial intelligence. Background Technology

[0002] In modern building and infrastructure construction, construction safety monitoring is a crucial step in ensuring project success, protecting worker safety, and reducing accident risks. Construction safety monitoring encompasses multiple aspects, including but not limited to environmental monitoring, structural health monitoring, and personnel behavior monitoring.

[0003] Traditional methods rely on manual inspections, sensor data collection, and manual data analysis. While effective, these methods also have many limitations:

[0004] Data processing is cumbersome: a large amount of sensor data needs to be manually screened and analyzed, which is time-consuming and prone to errors.

[0005] Insufficient real-time capability: Manual inspections cannot achieve 24 / 7 real-time monitoring, which may result in safety hazards not being detected in a timely manner.

[0006] High costs: Frequent manual inspections and complex data analysis increase the project's operating costs.

[0007] Lack of predictive ability: Traditional methods are unable to provide early warnings of potential risks, resulting in measures being taken only after an accident has occurred.

[0008] Therefore, the present invention provides a construction safety monitoring method and system based on artificial intelligence to solve the above problems. Summary of the Invention

[0009] In order to overcome the shortcomings of the existing technology, this invention provides a construction safety monitoring method and system based on artificial intelligence, which solves the problems of cumbersome data processing, insufficient real-time performance, high cost and lack of predictive ability of traditional monitoring methods.

[0010] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0011] Firstly, an artificial intelligence-based method for monitoring construction safety includes: acquiring multi-source heterogeneous data from the construction site; performing distributed feature extraction on the multi-source heterogeneous data using a federated learning framework to generate spatiotemporally correlated site status representation data, wherein the multi-source heterogeneous data includes video images, environmental monitoring data, equipment status data, and personnel positioning data; performing risk prediction using the site status representation data based on a preset dynamic risk prediction model, and outputting multi-level risk prediction results, wherein the preset dynamic risk prediction model is constructed based on a construction safety knowledge graph and a spatiotemporal graph neural network; triggering an adaptive feedback mechanism according to the risk level corresponding to the prediction results, generating visualized early warning information and equipment control commands, and linking the construction site control system to execute emergency response operations.

[0012] A further improvement in this application is that the distributed feature extraction of multi-source heterogeneous data using a federated learning framework to generate spatiotemporally correlated site status representation data includes: obtaining local feature model parameters of each data source, wherein video data is used to extract spatial-motion features through a 3D convolutional neural network, environmental monitoring data is used to extract temporal dependency features through an LSTM network, equipment status data is used to extract abnormal pattern features through an autoencoder, and personnel positioning data is used to extract group interaction features through a graph attention network; a spatiotemporal fusion module is deployed in the federated learning server, the spatiotemporal resolution of multi-source features is aligned through deformable convolution, and a cross-modal attention mechanism is used to establish dynamic correlations between features; and the feature parameters are encrypted and aggregated based on differential privacy technology to generate a global spatiotemporal correlation matrix as site status representation data.

[0013] A further improvement of this application is that the pre-defined dynamic risk prediction model based on a construction safety knowledge graph and a spatiotemporal graph neural network includes: constructing a construction safety knowledge graph, which contains the relationships between construction element entities, safety specification constraints, and accident cases, and mapping entity relationships into low-dimensional vectors through graph embedding technology; determining a spatiotemporal graph neural network, and fusing real-time construction site status data with semantic information of the knowledge graph through a gated spatiotemporal propagation mechanism; establishing a risk propagation dynamic equation, combining the causal analysis results of historical accident data, and establishing a diffusion model of risk probability in the spatiotemporal dimension; and integrating a cellular automata simulator to deduce the risk propagation path based on neighborhood weights and risk accumulation effects.

[0014] A further improvement of this application is that the expression of the dynamic risk prediction model is:

[0015] (1), in expression (1), This represents the risk probability value at the spatiotemporal location (t,s), and its value ranges from [0,1]. This represents a spatiotemporal graph neural network used to output the hidden states that fuse real-time data and knowledge graphs. This represents a cellular automaton simulator used to predict risk propagation paths. This indicates a multi-granularity risk perception module, used to comprehensively output situation prediction results at the device, regional, and global levels.

[0016] A further improvement of this application is that the adaptive feedback mechanism includes: constructing a mapping rule base for risk levels and response strategies, wherein the rule base contains three levels of risk thresholds; constructing a distributed control command verification mechanism, wherein after verifying the legality of the command through a blockchain smart contract, it is synchronously sent to the tower crane control system, the power safety module, and the personnel evacuation guidance system; and updating the emergency response log in real time, wherein the log includes the risk trigger time, command execution status, and operator identification, and the log data is stored in a distributed database.

[0017] A further improvement in this application is that the three-level risk thresholds of the rule base include: when the predicted risk value exceeds the first threshold, an augmented reality visualization warning is triggered, the risk area is highlighted in the smart helmet of the construction worker, and a voice prompt is pushed; when the predicted risk value exceeds the second threshold, an equipment speed limit command and an area isolation command are generated to limit the operating speed of the equipment in the dangerous area and to isolate high-risk workstations through an electronic fence; when the predicted risk value exceeds the third threshold, an equipment shutdown command and a personnel evacuation command are generated, the power supply to the dangerous equipment is cut off simultaneously, and the audible and visual alarm of the evacuation guidance system is activated.

[0018] A further improvement in this application is that the legality of control instructions is verified through blockchain smart contracts, including: generating smart contract code based on a predefined security policy template and deploying it to a private chain node; verifying the instruction signature using asymmetric encryption technology to ensure that the instruction source is trustworthy and has not been tampered with; and synchronously sending the verified instructions to the tower crane control system, the power safety module, and the personnel evacuation guidance system through a cross-chain protocol.

[0019] A further improvement of this application is that the federated learning framework adopts a dynamic weighted aggregation strategy, which calculates the dynamic fusion weight of each node parameter based on the spatiotemporal coverage completeness, feature update frequency and noise level of the data source.

[0020] Second aspect, an artificial intelligence-based construction safety monitoring system for implementing the above-mentioned method. The monitoring system includes: a multi-source data acquisition module for obtaining video images, environmental monitoring data, equipment status data, and personnel positioning data; a federated learning feature extraction module for distributively generating spatio-temporally associated construction site status characterization data; a dynamic risk prediction module integrating a construction safety knowledge graph, a spatio-temporal graph neural network, and a cellular automaton simulator to output multi-level risk prediction results; and an adaptive feedback control module for generating visual warning information and equipment control instructions and联动executing emergency response operations.

[0021] A further improvement of this application is that the dynamic risk prediction module includes: a multi-granularity risk perception sub-module including an equipment-level anomaly detection layer, a regional-level risk assessment layer, and a global-level situation prediction layer; a risk evolution deduction sub-module that dynamically simulates the risk propagation path based on the cellular automaton principle and displays the deduction result through a visual interface.

[0022] The beneficial effects of this invention are as follows: By integrating multi-source heterogeneous data and using the federated learning framework for distributed feature extraction, the efficiency and accuracy of data processing are effectively improved. The dynamic risk prediction model based on the construction safety knowledge graph and the spatio-temporal graph neural network not only strengthens the prediction function but also can timely warn of potential risks, significantly reducing the possibility of accidents. Introducing an adaptive feedback mechanism enables the system to automatically initiate corresponding emergency response measures according to the prediction results, ensuring the rapidity and effectiveness of emergency handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a schematic flowchart of an artificial intelligence-based construction safety monitoring method of this invention;

[0024] Figure 2 is a schematic structural diagram of an artificial intelligence-based construction safety monitoring system of this invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The embodiments of this invention will be described in detail below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of this invention and are not intended to limit the protection scope of this invention.

[0026] Traditional methods rely on manual inspections, sensor data collection, and manual data analysis. Although these methods are effective, they also have many limitations:

[0027] Complicated data processing: A large amount of sensor data needs to be manually screened and analyzed, which is time-consuming and error-prone.

[0028] Insufficient real-time capability: Manual inspections cannot achieve 24 / 7 real-time monitoring, which may result in safety hazards not being detected in a timely manner.

[0029] High costs: Frequent manual inspections and complex data analysis increase the project's operating costs.

[0030] Lack of predictive ability: Traditional methods are unable to provide early warnings of potential risks, resulting in measures being taken only after an accident has occurred.

[0031] Therefore, the present invention provides a construction safety monitoring method and system based on artificial intelligence to solve the above problems.

[0032] The technical solution will be described in detail below with reference to specific embodiments.

[0033] Example

[0034] refer to Figure 1 A construction safety monitoring method based on artificial intelligence, the method comprising the following steps S100-S300:

[0035] S100. Acquire multi-source heterogeneous data from the construction site, and use a federated learning framework to perform distributed feature extraction on the multi-source heterogeneous data to generate spatiotemporally correlated site status representation data. The multi-source heterogeneous data includes video images, environmental monitoring data, equipment status data, and personnel positioning data.

[0036] S200. Based on the preset dynamic risk prediction model, risk prediction is performed using the site status characterization data, and multi-level risk prediction results are output. The preset dynamic risk prediction model is constructed based on the construction safety knowledge graph and spatiotemporal graph neural network.

[0037] S300. Based on the risk level corresponding to the prediction result, an adaptive feedback mechanism is triggered to generate visual early warning information and equipment control commands, and the construction site control system is linked to execute emergency response operations.

[0038] In step S100, the distributed feature extraction of multi-source heterogeneous data using a federated learning framework to generate spatiotemporally correlated site status representation data includes:

[0039] S101. Obtain the local feature model parameters of each data source. Among them, video data is extracted with spatial-motion features through a 3D convolutional neural network, environmental monitoring data is extracted with temporal dependency features through an LSTM network, equipment status data is extracted with abnormal pattern features through an autoencoder, and personnel positioning data is extracted with group interaction features through a graph attention network.

[0040] S102. Deploy a spatiotemporal fusion module in the federated learning server, align the spatiotemporal resolution of multi-source features through deformable convolution, and establish dynamic correlations between features using a cross-modal attention mechanism.

[0041] S103. Based on differential privacy technology, the feature parameters are encrypted and aggregated to generate a global spatiotemporal correlation matrix as the site status representation data.

[0042] In step S100, the federated learning framework employs a dynamic weighted aggregation strategy to calculate the dynamic fusion weights of each node parameter based on the spatiotemporal coverage completeness, feature update frequency, and noise level of the data source.

[0043] In step S200, the construction of the preset dynamic risk prediction model based on the construction safety knowledge graph and spatiotemporal graph neural network includes:

[0044] S201. Construct a construction safety knowledge graph, which includes the relationships between construction element entities, safety specification constraints, and accident cases, and maps entity relationships into low-dimensional vectors through graph embedding technology.

[0045] S202. Determine the spatiotemporal graph neural network and fuse real-time construction site status data with knowledge graph semantic information through a gated spatiotemporal propagation mechanism;

[0046] S203. Establish a risk propagation dynamics equation, combine the causal analysis results of historical accident data, and establish a risk probability diffusion model in the spatiotemporal dimension.

[0047] S204, an integrated cellular automaton simulator, deduces risk propagation paths based on neighborhood weights and risk accumulation effects.

[0048] In one embodiment of this application, the expression of the dynamic risk prediction model is:

[0049] (1),

[0050] In expression (1), This represents the risk probability value at the spatiotemporal location (t,s), and its value ranges from [0,1]. This represents a spatiotemporal graph neural network used to output the hidden states that fuse real-time data and knowledge graphs. This represents a cellular automaton simulator used to predict risk propagation paths. This indicates a multi-granularity risk perception module, used to comprehensively output situation prediction results at the device, regional, and global levels.

[0051] Specifically, the dynamic inference layer of the spatiotemporal graph neural network consists of gated spatiotemporal convolution and multi-head spatiotemporal Transformer, and its expression is:

[0052] (2),

[0053] In expression (2), The site state representation at time t (from the spatiotemporal correlation matrix of federated learning). This represents the knowledge graph embedding vector (generated using the TransE algorithm). Represents gated spatiotemporal convolution, the formula is: , where ⊙ represents element-wise multiplication, used to dynamically adjust feature importance; This represents a multi-head spatiotemporal attention encoder, whose attention weights are calculated as follows: ,in, They are generated from spatiotemporal location codes respectively; This represents the activation function (such as ReLU). The weight matrix is ​​trainable.

[0054] The state transition rules for cellular automata are defined as follows:

[0055] (3),

[0056] In expression (3), This represents the risk value of cell s at time t. This represents the neighborhood of cell s (such as adjacent equipment or workstations). The neighborhood weight is dynamically calculated based on the entity association strength in the knowledge graph. Where Sim is the cosine similarity. This represents a bias term used to adjust for sensitivity to risk accumulation.

[0057] Multi-granularity risk perception module Risk results are output in three levels:

[0058] (4),

[0059] In expression (4), device-level anomaly detection is:

[0060] ;

[0061] The regional risk assessment is as follows:

[0062] ;

[0063] Global situational prediction is as follows:

[0064] ;

[0065] Based on the quantification of risk diffusion using partial differential equations, the expression for the risk propagation dynamics equation is as follows:

[0066] (5),

[0067] In expression (5), This represents the spatial diffusion coefficient (determined by the layout of the construction area). This indicates a risk self-reinforcing factor (derived from historical accident chain reaction analysis). This indicates the environmental carrying capacity threshold (matching safety specification constraints). This represents the risk attenuation coefficient (reflecting emergency response efficiency).

[0068] In step S300, the adaptive feedback mechanism includes:

[0069] S301. Construct a rule base for mapping risk levels and response strategies, wherein the rule base contains three levels of risk thresholds;

[0070] S302. Construct a distributed control command verification mechanism. After verifying the legality of the command through a blockchain smart contract, the command is simultaneously sent to the tower crane control system, the power safety module, and the personnel evacuation guidance system.

[0071] S303 updates the emergency response log in real time. The log includes the risk trigger time, instruction execution status and operator identification, and stores the log data through a distributed database.

[0072] In step S301, the three-level risk thresholds of the rule base include:

[0073] When the predicted risk value exceeds the first threshold, an augmented reality visualization warning is triggered, highlighting the risk area in the construction worker's smart helmet and pushing a voice prompt.

[0074] When the predicted risk value exceeds the second threshold, an equipment speed limit command and an area isolation command are generated to limit the operating speed of equipment in the dangerous area and isolate high-risk workstations through electronic fences.

[0075] When the predicted risk value exceeds the third threshold, an equipment shutdown command and a personnel evacuation command are generated, the power supply to the dangerous equipment is cut off simultaneously, and the audible and visual alarms of the evacuation guidance system are activated.

[0076] In step 302, the legality of the control instructions is verified through a blockchain smart contract, including:

[0077] Based on a predefined security policy template, smart contract code is generated and deployed to a private chain node;

[0078] Asymmetric encryption technology is used to verify the instruction signature to ensure that the instruction source is trustworthy and has not been tampered with;

[0079] The verified instructions are synchronously sent to the tower crane control system, the electrical safety module, and the personnel evacuation guidance system via the cross-chain protocol.

[0080] To facilitate understanding of the above method, the following example is provided:

[0081] At a construction site, in the tower crane's operating area (labeled as cell s0), sensors detected abnormal vibration of the tower crane boom, with the vibration amplitude exceeding a preset safety threshold. The system acquires this equipment status data in real time through a multi-source data acquisition module and initiates a dynamic risk prediction process.

[0082] I. Data Flow and Preprocessing

[0083] Data collection:

[0084] Equipment status data: The tower crane vibration sensor recorded a vibration frequency of 25 Hz (the safety threshold is 20 Hz), and the health score dropped to 0.6 (out of 1).

[0085] Video data: The camera captured an abnormally large increase in the swing amplitude of the tower crane hook. Spatial-motion features were extracted using a 3D convolutional neural network to generate the event label "lifting imbalance".

[0086] Environmental monitoring data: The wind speed sensor shows an instantaneous wind speed of 8 m / s (the safe upper limit is 6 m / s).

[0087] Preprocessing:

[0088] Spatiotemporal alignment: All data are aligned by timestamps and features are extracted using a local feature model within a federated learning framework (e.g., abnormal pattern features of vibration data are extracted via an autoencoder).

[0089] Standardized input: generating the spatiotemporal correlation matrix It includes a fusion representation of device status, environmental data, and video features.

[0090] II. Operation of the Dynamic Risk Prediction Model

[0091] (a) Spatiotemporal graph neural network inference

[0092] enter:

[0093] Real-time construction site status representation (From Federated Learning).

[0094] Knowledge graph embedding : Query the entities associated with "tower crane overload accident" in the knowledge graph, including the causal chain of historical accidents (such as "overload → structural fatigue → fracture"), and map them as low-dimensional vectors.

[0095] Calculation process:

[0096] - Gated spatiotemporal convolution: Integrates vibration data with the "overload risk" pattern in a knowledge graph to output the hidden state. .

[0097] - Multi-head spatiotemporal Transformer: Captures the long-range dependencies between the tower crane area and surrounding workstations (such as the impact of wind speed on hoisting stability) and generates an attention weight matrix.

[0098] -Comprehensive Output: STGN outputs the current risk hidden state. =0.8 (equipment-level risk probability).

[0099] (b) Cellular Automata (CA) Risk Propagation Simulation

[0100] - Neighborhood definition: The neighborhood of cell s0 This includes the adjacent rebar processing area (s1) and personnel passage (s2).

[0101] -Weight calculation:

[0102] - Calculate neighborhood weights based on the entity association strength in the knowledge graph:

[0103] (The tower crane is linked to the equipment in the steel bar processing area) (Relevance of personnel access safety)

[0104] - Bias term b = 0.1 (reflects the additional risk of the current wind speed).

[0105] -State transition:

[0106]

[0107] This indicates that the risk value for the tower crane area has decreased slightly from 0.8 to 0.75, but it is still in the high-risk range.

[0108] (c) Risk propagation dynamics equation

[0109] -Parameter settings:

[0110] =0.2 (The tower crane area is an open space, and the risk spreads quickly).

[0111] =0.5 (Historical data shows that overloading accidents are prone to triggering chain reactions);

[0112] =0.8 (Safety regulations limit the maximum risk bearing capacity of the tower crane area);

[0113] =0.3 (Medium emergency response efficiency).

[0114] -Diffusion prediction:

[0115] By solving partial differential equations The risk is predicted to spread to the rebar processing area within 10 minutes (the S1 risk value rises to 0.65).

[0116] III. Multi-granularity risk perception and early warning triggering

[0117] - Device-level anomaly detection:

[0118] 0.8 (abnormal vibration) triggers a device-level alarm.

[0119] -Regional risk assessment:

[0120] Extract the highest risk value from the cellular automaton output using max pooling operations. A score of 0.75 indicates that the tower crane area is classified as high-risk.

[0121] -Global-level situation prediction:

[0122] The LSTM model integrates device and regional risks to predict the global risk probability within the next 15 minutes. 0.7.

[0123] Warning triggering logic:

[0124] - First threshold (0.6): The AR helmet highlights the tower crane area and pushes a voice prompt "Pay attention to hoisting safety".

[0125] -Second threshold (0.7):

[0126] -Issued a speed limit command to reduce the tower crane's operating speed to 50%.

[0127] - Activate the electronic fence to isolate the rebar processing area (s1).

[0128] -Third threshold (0.8): Not triggered, but the system continues to monitor.

[0129] IV. Adaptive Feedback and Instruction Execution

[0130] - Blockchain smart contract verification:

[0131] - After the control instructions (rate limit, isolation) are generated, the legality of the instruction format is verified through a predefined smart contract template.

[0132] - Use RSA asymmetric encryption to sign the instructions, ensuring that the instructions originate from a trusted server.

[0133] - Once the verification is successful, the instructions are synchronized to the tower crane control system and the electronic fence module via the cross-chain protocol.

[0134] - Emergency response log update:

[0135] - Record the event time and command execution status (e.g., "Tower crane speed has dropped to 50%").

[0136] - Log data is stored in a distributed database and marked as incremental learning samples.

[0137] V. Model Optimization and Closed-Loop Feedback

[0138] Incremental learning:

[0139] Store the data of this event (abnormal vibration, excessive wind speed, and early warning results) into the memory playback buffer.

[0140] The model parameters are updated using the Elastic Weight Consolidation (EWC) algorithm, preserving key weights for historical accident patterns (such as overload risk).

[0141] -Dynamic adjustment of risk threshold:

[0142] Based on the emergency response delay time in the logs (such as 2 seconds for issuing instructions), the second threshold is automatically optimized to 0.68 to trigger isolation measures in advance.

[0143] Compared to existing technologies, this application presents an AI-based construction safety monitoring method that automates risk prediction and emergency response, significantly improving the safety management level of construction sites. By fusing real-time data and knowledge graphs through a spatiotemporal graph neural network, the system can accurately capture changes in the site's status and predict potential risks. Cellular automata simulate risk propagation paths, further enhancing the accuracy and foresight of early warnings. A multi-granularity risk perception module comprehensively outputs device-level, regional-level, and global-level situation prediction results, providing managers with a comprehensive risk view. An adaptive feedback mechanism triggers different levels of response strategies based on predicted risk values, and verifies the legality of instructions through blockchain smart contracts, ensuring reliable execution of instructions. This method not only improves the efficiency and accuracy of safety monitoring but also promotes the digital transformation and intelligent upgrading of the construction industry.

[0144] refer to Figure 2 An artificial intelligence-based construction safety monitoring system is used to implement the above-described method, the monitoring system comprising:

[0145] The multi-source data acquisition module is used to acquire video images, environmental monitoring data, equipment status data, and personnel positioning data;

[0146] The federated learning feature extraction module is used to generate spatiotemporally correlated site status representation data in a distributed manner.

[0147] The dynamic risk prediction module integrates a construction safety knowledge graph, a spatiotemporal graph neural network, and a cellular automata simulator to output multi-level risk prediction results.

[0148] The adaptive feedback control module is used to generate visual early warning information and equipment control commands, and to coordinate and execute emergency response operations.

[0149] In one embodiment of this application, the dynamic risk prediction module includes:

[0150] The multi-granularity risk perception submodule includes an equipment-level anomaly detection layer, a regional-level risk assessment layer, and a global-level situation prediction layer.

[0151] The risk evolution simulation submodule dynamically simulates risk propagation paths based on the principle of cellular automata and displays the simulation results through a visual interface.

[0152] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0154] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0156] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0157] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0158] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0159] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A construction safety monitoring method based on artificial intelligence, characterized in that, The method includes: Acquire multi-source heterogeneous data from the construction site, and use a federated learning framework to perform distributed feature extraction on the multi-source heterogeneous data to generate spatiotemporally correlated site status representation data. The multi-source heterogeneous data includes video images, environmental monitoring data, equipment status data, and personnel positioning data. Based on a preset dynamic risk prediction model, risk prediction is performed using the site status characterization data, and multi-level risk prediction results are output. The preset dynamic risk prediction model is constructed based on a construction safety knowledge graph and a spatiotemporal graph neural network. Based on the risk level corresponding to the prediction results, an adaptive feedback mechanism is triggered to generate visual early warning information and equipment control commands, and the construction site control system is linked to execute emergency response operations. A federated learning framework is used to perform distributed feature extraction on multi-source heterogeneous data, generating spatiotemporally correlated site status representation data, including: The local feature model parameters of each data source are obtained. Among them, the video data is extracted with spatial-motion features through a 3D convolutional neural network, the environmental monitoring data is extracted with temporal dependency features through an LSTM network, the equipment status data is extracted with abnormal pattern features through an autoencoder, and the personnel positioning data is extracted with group interaction features through a graph attention network. Deploy a spatiotemporal fusion module in the federated learning server to align the spatiotemporal resolution of multi-source features through deformable convolution, and establish dynamic correlations between features using a cross-modal attention mechanism; Based on differential privacy technology, feature parameters are encrypted and aggregated to generate a global spatiotemporal correlation matrix as site status representation data; The expression for the dynamic risk prediction model is: （1）， In expression (1), This represents the risk probability value at the spatiotemporal location (t,s), and its value ranges from [0,1]. This represents a spatiotemporal graph neural network used to output the hidden states that fuse real-time data and knowledge graphs. This represents a cellular automaton simulator used to predict risk propagation paths. This indicates a multi-granularity risk perception module, used to comprehensively output situation prediction results at the device, regional, and global levels.

2. The construction safety monitoring method based on artificial intelligence according to claim 1, characterized in that, The pre-defined dynamic risk prediction model constructed based on construction safety knowledge graph and spatiotemporal graph neural network includes: A construction safety knowledge graph is constructed, which includes the relationships between construction element entities, safety specification constraints, and accident cases, and the entity relationships are mapped into low-dimensional vectors through graph embedding technology. A spatiotemporal graph neural network is determined, and real-time construction site status data is fused with semantic information from a knowledge graph through a gated spatiotemporal propagation mechanism. Establish a risk propagation dynamics equation, combine the causal analysis results of historical accident data, and establish a diffusion model of risk probability in the spatiotemporal dimension; An integrated cellular automaton simulator is used to deduce risk propagation paths based on neighborhood weights and risk accumulation effects.

3. The construction safety monitoring method based on artificial intelligence according to claim 1, characterized in that, The adaptive feedback mechanism includes: Construct a rule base for mapping risk levels to response strategies, wherein the rule base contains three levels of risk thresholds; A distributed control command verification mechanism is constructed. After verifying the legality of the command through a blockchain smart contract, it is synchronously sent to the tower crane control system, the power safety module, and the personnel evacuation guidance system. The emergency response log is updated in real time. The log includes the risk trigger time, instruction execution status and operator identification, and the log data is stored in a distributed database.

4. The construction safety monitoring method based on artificial intelligence according to claim 3, characterized in that, The three-level risk thresholds of the rule base include: When the predicted risk value exceeds the first threshold, an augmented reality visualization warning is triggered, highlighting the risk area in the construction worker's smart helmet and pushing a voice prompt. When the predicted risk value exceeds the second threshold, an equipment speed limit command and an area isolation command are generated to limit the operating speed of equipment in the dangerous area and isolate high-risk workstations through electronic fences. When the predicted risk value exceeds the third threshold, an equipment shutdown command and a personnel evacuation command are generated, the power supply to the dangerous equipment is cut off simultaneously, and the audible and visual alarms of the evacuation guidance system are activated.

5. The construction safety monitoring method based on artificial intelligence according to claim 3, characterized in that, Verifying the legitimacy of control commands through blockchain smart contracts includes: Based on a predefined security policy template, smart contract code is generated and deployed to a private chain node; Asymmetric encryption technology is used to verify the instruction signature to ensure that the instruction source is trustworthy and has not been tampered with; The verified instructions are synchronously sent to the tower crane control system, the electrical safety module, and the personnel evacuation guidance system via the cross-chain protocol.

6. The construction safety monitoring method based on artificial intelligence according to claim 1, characterized in that, The federated learning framework employs a dynamic weighted aggregation strategy, which calculates the dynamic fusion weights of each node's parameters based on the spatiotemporal coverage completeness, feature update frequency, and noise level of the data source.

7. A construction safety monitoring system based on artificial intelligence, used to implement the method described in any one of claims 1-6, characterized in that, The monitoring system includes: The multi-source data acquisition module is used to acquire video images, environmental monitoring data, equipment status data, and personnel positioning data; The federated learning feature extraction module is used to generate spatiotemporally correlated site status representation data in a distributed manner. The dynamic risk prediction module integrates a construction safety knowledge graph, a spatiotemporal graph neural network, and a cellular automata simulator to output multi-level risk prediction results. The adaptive feedback control module is used to generate visual early warning information and equipment control commands, and to coordinate and execute emergency response operations.

8. The construction safety monitoring system based on artificial intelligence according to claim 7, characterized in that, The dynamic risk prediction module includes: The multi-granularity risk perception submodule includes an equipment-level anomaly detection layer, a regional-level risk assessment layer, and a global-level situation prediction layer. The risk evolution simulation submodule dynamically simulates risk propagation paths based on the principle of cellular automata and displays the simulation results through a visual interface.

Images