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

Abstract

The invention discloses a building construction safety monitoring and early warning system based on the Internet of Things, and relates to the technical field of construction safety monitoring, and the system comprises an intelligent sensing module which generates a multi-source heterogeneous safety monitoring data flow with a space-time label; the risk research and judgment module is used for performing deep feature extraction and cross-modal fusion analysis on the multi-source heterogeneous safety monitoring data stream, and outputting a refined regional risk level quantized value by using a trained construction safety risk assessment model; the dynamic early warning module is used for intelligently matching a dynamically updated early warning strategy library based on correlation analysis of a regional risk level quantized value and a construction process time sequence; and the security center module is used for storing non-tampered historical monitoring data and early warning records by adopting a distributed block chain storage architecture. According to the scheme, adaptive sampling, cross-modal risk assessment, hierarchical early warning linkage and block chain and knowledge graph management technologies are fused, and the whole-process intelligent and precise upgrading of building construction safety from data acquisition to decision support is realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of construction safety monitoring, in particular to a building construction safety monitoring and early warning system based on the Internet of Things. BACKGROUND

[0002] The building construction industry has the characteristics of high-altitude operation, frequent cross-operation, high equipment density and complex and changeable environment, and is a field with high safety accidents. Traditional construction safety management mainly relies on manual inspection, fixed-point monitoring and experience judgment, and has the following significant deficiencies: first, the monitoring range is limited, it is difficult to achieve full-time and space coverage of the construction area, and blind spots in supervision are prone to occur; second, data collection is lagging, manual recording and single device monitoring cannot capture personnel violations, abnormal equipment operation and environmental parameter changes in real time; third, risk assessment is extensive, there is a lack of deep fusion analysis of multi-source data, and it is difficult to accurately identify high-risk areas and potential hazards; fourth, early warning response is passive, and it is mostly post-alarm rather than pre-prevention, and the early warning strategy lacks pertinence, which easily leads to low disposal efficiency; fifth, data tracing is difficult, monitoring records are easy to tamper with and lose, and it is difficult to support accident tracing and management optimization. With the development of the Internet of Things, artificial intelligence and blockchain technology, intelligent monitoring systems have gradually been applied to construction safety management, but the existing systems still have problems such as unreasonable terminal deployment, fixed sampling frequency, insufficient feature fusion and static early warning strategy, and cannot meet the needs of dynamic safety management in complex construction scenarios. Therefore, it is urgent to build an intelligent monitoring and early warning system that integrates real-time sensing, intelligent judgment, dynamic early warning and data tracing, so as to improve the fine and active level of building construction safety management. SUMMARY

[0003] To solve the above technical problems, the building construction safety monitoring and early warning system based on the Internet of Things is provided, which solves the above problems.

[0004] To achieve the above purposes, the technical scheme adopted by the application is as follows: The building construction safety monitoring and early warning system based on the Internet of Things comprises an intelligent sensing module, a risk judgment module, a dynamic early warning module and a safety hub module. The intelligent sensing module is used to deploy an array of Internet of Things monitoring terminals with edge computing capability, dynamically adjusts the collection frequency through an adaptive sampling algorithm, and collects personnel three-dimensional dynamic trajectory data, equipment full life cycle operation parameters and environmental multi-dimensional index data in real time in the construction area, to generate multi-source heterogeneous safety monitoring data stream with time and space labels. The risk assessment module is communicatively connected to the intelligent sensing module. The risk assessment module is used to perform deep feature extraction and cross-modal fusion analysis on multi-source heterogeneous safety monitoring data streams. Using a trained construction safety risk assessment model, it focuses on the characteristics of high-risk areas through a spatiotemporal attention mechanism and outputs a refined quantitative value of the regional risk level. The dynamic early warning module is communicatively connected to the risk assessment module. The dynamic early warning module is used to intelligently match the dynamically updated early warning strategy library based on the correlation analysis between the quantitative value of regional risk level and the construction procedure sequence, generate hierarchical early warning instructions containing spatial positioning information, and execute them in a coordinated manner. The safety central module is communicatively connected to the intelligent sensing module, risk assessment module, and dynamic early warning module. The safety central module uses a distributed blockchain storage architecture to store tamper-proof historical monitoring data and early warning records, and constructs a construction safety decision support system through knowledge graph technology.

[0005] Preferably, the intelligent sensing module includes a terminal deployment unit, a dynamic acquisition unit, and an edge processing unit; The terminal deployment unit is used to plan the deployment locations of IoT monitoring terminals based on the three-dimensional modeling results of the construction site. The IoT monitoring terminals include UWB positioning base stations, millimeter-wave radar, vibration sensors, gas sensors, and infrared thermal imagers. The dynamic acquisition unit is used to control the terminal array to acquire data through an adaptive sampling algorithm. When the risk level of the area is higher than the preset threshold, the sampling interval is shortened to 1 to 3 seconds, and when the risk level of the area is lower than the preset threshold, the sampling interval is extended to 10 to 30 seconds. The edge processing unit is used to perform noise reduction, format conversion, and preliminary feature extraction on the original acquired data locally on the terminal, thereby reducing the amount of data transmission.

[0006] Preferably, the step of controlling the terminal array to collect data through an adaptive sampling algorithm specifically includes: Initialize the sampling parameters, set the baseline sampling interval to 10 seconds, the risk level adjustment coefficient to 0.85, the preset high-risk threshold to 0.5, and the low-risk threshold to 0.3; The system receives the quantitative value of the regional risk level output from the risk assessment module in real time and dynamically adjusts the sampling frequency according to the sampling interval calculation formula, which is as follows: ; In the formula, This is the current sampling interval. As the reference sampling interval, This represents the risk level adjustment factor. This represents the quantitative value of the regional risk level, output by the risk assessment module. The value ranges from 0 to 1, reflecting the current level of safety risk in the construction area. When the calculated sampling interval is less than 1 second, the minimum sampling interval is automatically limited to 1 second; when the sampling interval is greater than 30 seconds, the maximum sampling interval is automatically limited to 30 seconds. According to the adjusted sampling interval, a synchronous acquisition command is sent to the IoT monitoring terminal array to ensure that each terminal completes data acquisition on the same time axis. The sampling interval is calibrated every 5 minutes, and the interval value is finely adjusted based on the risk level change trend in the past 5 minutes. If the risk level continues to rise, the sampling interval is shortened by an additional 20%; if it continues to fall, the sampling interval is extended by an additional 20%.

[0007] Preferably, the risk assessment module includes a feature extraction unit, a cross-modal fusion unit, and a risk calculation unit; The feature extraction unit is used to extract behavioral features from three-dimensional dynamic trajectory data of personnel using convolutional neural networks, extract spectral features from equipment operating parameters using wavelet transform, and perform dimensionality reduction processing on multi-dimensional environmental index data through principal component analysis. The cross-modal fusion unit is used to construct a feature association matrix and assign dynamic weights to different types of features through a spatiotemporal attention mechanism to achieve deep fusion of multi-source features. The risk calculation unit is used to input the fused features into the trained construction safety risk assessment model and output personnel risk values, equipment risk values, and environmental risk values.

[0008] Preferably, the step of inputting the fused features into the trained construction safety risk assessment model and outputting personnel risk values, equipment risk values, and environmental risk values ​​includes: The fusion features of personnel behavior are input into the sub-model. By identifying dangerous behaviors such as unauthorized climbing and failure to wear protective equipment, the model outputs a personnel risk value in the range of 0-1. The formula for calculating the personnel risk value is as follows: ; In the formula, Indicates the first Weighting coefficients for risky behaviors Indicates the first Confidence level for identifying risky behaviors This indicates the total number of types of dangerous behavior. For personnel risk values; The equipment operation fusion characteristics are input into the sub-model. Based on the abnormal vibration spectrum and temperature exceedance fault characteristics of the equipment, the equipment risk value in the range of 0-1 is output. The calculation formula is as follows: ; The environmental indicators are integrated and input into the sub-model. Based on the degree of exceedance of toxic gas concentration and dust content, an environmental risk value in the range of 0-1 is output. The formula for calculating the environmental risk value is as follows: ; In the formula, Indicates the first Weighting coefficients for environmental indicators Indicates the first Measured values ​​of environmental indicators, Indicates the first Safety thresholds for environmental indicators This indicates the total number of environmental indicator types. This indicates the environmental risk value.

[0009] Preferably, the training process of the construction safety risk assessment model includes: Collect samples of dangerous behaviors, equipment failure cases, and environmental exceedance events from historical monitoring data to construct a labeled training dataset; An improved random forest algorithm is used to construct the model framework, with the number of decision trees set to 50-100. The hyperparameters of maximum depth and minimum number of split samples are optimized using grid search. The formula for calculating feature importance weights is as follows: ; In the formula, Indicates the first The importance weights of each feature Indicates the number of decision trees. Indicates the first The first decision tree The change in Gini coefficients before and after each feature split; The model performance is evaluated using five-fold cross-validation to obtain the model's accuracy and recall. Training is completed when both accuracy and recall exceed 0.9, and an evaluation model file that can be directly used is generated.

[0010] Preferably, the dynamic early warning module includes a threshold comparison unit, a strategy matching unit, and an instruction execution unit; The threshold comparison unit is used to compare the quantitative value of the regional risk level with the preset three-level warning thresholds, which include a first-level warning threshold (0.3-0.5), a second-level warning threshold (0.5-0.7), and a third-level warning threshold (>0.7). The strategy matching unit is used to match the corresponding strategy from the early warning strategy library according to the comparison results. The first-level early warning corresponds to the sound and light prompts, the second-level early warning corresponds to the equipment load limit operation, and the third-level early warning corresponds to the equipment shutdown and area blockade. The instruction execution unit is used to send control instructions containing spatial positioning information to the on-site early warning device via the LoRa wireless communication protocol, and simultaneously trigger a pop-up alarm on the management platform.

[0011] Preferably, the linkage execution process of the instruction execution unit includes: Analyze the spatial positioning information in the graded early warning instructions and match it with the specific area coordinates in the three-dimensional model of the construction site; A start command is sent to the audible and visual alarms in the area, controlling the alarm devices to emit audible and visual signals that meet the warning level. The formula for calculating the alarm sound intensity is as follows: ; In the formula, Indicates the basic sound intensity. The sound intensity coefficient indicates the risk level. This represents the quantitative value of the regional risk level; Send control commands to the associated equipment controller. In the event of a Level 1 warning, maintain equipment operation and issue a warning. In the event of a Level 2 warning, limit the equipment load to 70% of the rated value. In the event of a Level 3 warning, immediately trigger the equipment shutdown procedure. The warning information will be pushed to the management terminal simultaneously, including the coordinates of the risk area, the risk level, and the recommended handling measures.

[0012] Preferably, the security hub module includes a blockchain storage unit, a knowledge graph construction unit, and a decision support unit; The blockchain storage unit is used to store historical monitoring data and early warning records using a consortium blockchain architecture. Each block contains a data hash value, a timestamp, and a node signature. The knowledge graph construction unit is used to extract entities and relationships related to construction safety and construct a knowledge graph that includes personnel, equipment, environment and risk events. The decision support unit is used to provide functions such as risk event association query, historical similar case matching, and emergency response plan recommendation based on knowledge graph.

[0013] Preferably, the specific construction process of the knowledge graph construction unit includes: Extract core entities of personnel, equipment, environment, and risk events from historical monitoring data and early warning records, and define entity attributes such as personnel type, equipment model, environment type, and event level; Identify the relationships between entities and calculate the strength of these relationships. The formula for calculating the strength of entity relationships is as follows: ; in, Representing entities With entity The number of times they appear together Representing entities Total number of occurrences Representing entities Total number of occurrences express; A graph database is used to store entity and relation data, and a knowledge graph structure in the form of triples is constructed. New monitoring data and early warning events are regularly added, and the knowledge graph content is updated through entity linking and relation reasoning techniques.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By deploying an IoT terminal array with edge computing capabilities, real-time collection of personnel's three-dimensional dynamic trajectories, equipment's full lifecycle parameters, and multi-dimensional environmental indicators is achieved. Combined with multi-source heterogeneous data streams with spatiotemporal tags, a full-coverage monitoring network for the construction area is constructed, effectively eliminating the blind spots of traditional manual inspections. At the same time, an adaptive sampling algorithm based on regional risk levels is adopted to dynamically adjust the terminal collection frequency, reducing redundant data, lowering transmission and processing pressure, and improving the overall system operating efficiency while ensuring monitoring accuracy, thus providing comprehensive and efficient data support for safety management. 2. By leveraging technologies such as convolutional neural networks and wavelet transforms to extract features from multi-source data, and combining them with a spatiotemporal attention mechanism to achieve cross-modal feature fusion, the accuracy of risk level assessment is significantly improved through quantitative calculation and weight allocation mechanisms for personnel, equipment, and environmental risk values. Based on a three-level early warning threshold and construction procedure time sequence correlation analysis, an intelligent matching early warning strategy library is used to achieve graded handling such as audible and visual prompts, equipment load limits, and shutdown and blockade. Combined with a linkage execution process that includes spatial positioning information and a dynamic adjustment mechanism for alarm sound intensity, the early warning information is ensured to reach the target audience accurately and respond quickly, significantly reducing the probability of accidents and the extent of losses. 3. Adopting a distributed blockchain storage architecture ensures that monitoring data and early warning records are tamper-proof, providing a reliable basis for accident tracing and liability determination. By leveraging knowledge graph technology to build an entity relationship network, it supports risk event association queries, historical similar case matching, and emergency plan recommendations, promoting the transformation of safety management from passive response to proactive prevention. The combination of blockchain evidence storage and knowledge graph support enables the entire construction safety process to be traceable, analyzable, and optimizable, improving the scientific and efficient nature of safety decision-making and facilitating the upgrading of construction safety management models. Attached Figure Description

[0015] Figure 1 This is a system framework diagram of the present invention; Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0016] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0017] Reference Figure 1As shown, the IoT-based building construction safety monitoring and early warning system includes: an intelligent sensing module, a risk assessment module, a dynamic early warning module, and a safety central module; Reference Figure 2 As shown, the intelligent sensing module is used to deploy an array of IoT monitoring terminals with edge computing capabilities. It dynamically adjusts the acquisition frequency through an adaptive sampling algorithm, and collects real-time three-dimensional dynamic trajectory data of personnel in the construction area, equipment life cycle operation parameters, and multi-dimensional environmental indicator data to generate a multi-source heterogeneous safety monitoring data stream with spatiotemporal tags. The risk assessment module is communicatively connected to the intelligent sensing module. The risk assessment module is used to perform deep feature extraction and cross-modal fusion analysis on multi-source heterogeneous safety monitoring data streams. Using a trained construction safety risk assessment model, it focuses on the characteristics of high-risk areas through a spatiotemporal attention mechanism and outputs a refined quantitative value of the regional risk level. The dynamic early warning module is communicatively connected to the risk assessment module. The dynamic early warning module is used to intelligently match the dynamically updated early warning strategy library based on the correlation analysis between the quantitative value of regional risk level and the construction procedure sequence, generate hierarchical early warning instructions containing spatial positioning information, and execute them in a coordinated manner. The safety central module is communicatively connected to the intelligent sensing module, risk assessment module, and dynamic early warning module. The safety central module uses a distributed blockchain storage architecture to store tamper-proof historical monitoring data and early warning records, and constructs a construction safety decision support system through knowledge graph technology.

[0018] The intelligent sensing module includes a terminal deployment unit, a dynamic acquisition unit, and an edge processing unit; The terminal deployment unit is used to plan the deployment locations of IoT monitoring terminals based on the 3D modeling results of the construction site. The IoT monitoring terminals include UWB positioning base stations, millimeter-wave radar, vibration sensors, gas sensors, and infrared thermal imagers. The dynamic acquisition unit is used to control the terminal array to acquire data through an adaptive sampling algorithm. When the regional risk level is higher than a preset threshold, the sampling interval is shortened to 1-3 seconds, and when the regional risk level is lower than the preset threshold, the sampling interval is extended to 10-30 seconds. The edge processing unit is used to perform noise reduction, format conversion, and preliminary feature extraction on the raw acquired data locally on the terminal to reduce the amount of data transmission. The intelligent sensing module innovatively integrates multiple types of IoT monitoring terminals, accurately planning and deploying monitoring points based on 3D modeling of the construction site. Compared to traditional random point deployment methods, this achieves comprehensive, blind-spot-free monitoring of the construction area. The adaptive sampling algorithm, referencing personalized principal component analysis, dynamically adjusts the sampling frequency according to the risk level. This protects data privacy while reducing data transmission and storage costs, a groundbreaking approach in the data acquisition stage of construction safety monitoring. It makes data collection more closely reflect actual risk conditions, improving data quality and utilization efficiency.

[0019] The specific steps of controlling the terminal array to collect data using an adaptive sampling algorithm include: Initialize the sampling parameters, set the baseline sampling interval to 10 seconds, the risk level adjustment coefficient to 0.85, the preset high-risk threshold to 0.5, and the low-risk threshold to 0.3; The system receives the quantitative value of the regional risk level output from the risk assessment module in real time and dynamically adjusts the sampling frequency according to the sampling interval calculation formula, which is as follows: ; In the formula, This is the current sampling interval. As the reference sampling interval, This represents the risk level adjustment factor. This represents the quantitative value of the regional risk level, output by the risk assessment module. The value ranges from 0 to 1, reflecting the current level of safety risk in the construction area. When the calculated sampling interval is less than 1 second, the minimum sampling interval is automatically limited to 1 second; when the sampling interval is greater than 30 seconds, the maximum sampling interval is automatically limited to 30 seconds. Synchronous acquisition instructions are sent to the IoT monitoring terminal array according to the adjusted sampling interval to ensure that each terminal completes data acquisition on a unified time axis. The sampling interval is calibrated every 5 minutes, and the interval value is finely adjusted based on the risk level change trend in the past 5 minutes. If the risk level continues to rise, the sampling interval is shortened by an additional 20%; if it continues to fall, the sampling interval is extended by an additional 20%. This algorithm achieves highly dynamic and precise sampling interval control for the first time in the construction industry. By introducing a risk level adjustment coefficient and combining real-time risk quantification and trend analysis, it not only flexibly adjusts the sampling frequency but also sets reasonable upper and lower limits to prevent extreme sampling intervals. Furthermore, the 5-minute calibration mechanism, similar to the dynamic updates of time-series data, ensures data continuity while promptly capturing risk changes, providing a more accurate and real-time data foundation for subsequent analysis. This adaptive calibration mechanism is innovative and leading among similar monitoring systems.

[0020] The risk assessment module includes a feature extraction unit, a cross-modal fusion unit, and a risk calculation unit; The feature extraction unit is used to extract behavioral features from the three-dimensional dynamic trajectory data of personnel using a convolutional neural network, extract spectral features from the equipment operating parameters using wavelet transform, and perform dimensionality reduction processing on the multi-dimensional environmental indicator data through principal component analysis; the cross-modal fusion unit is used to construct a feature correlation matrix and assign dynamic weights to different types of features through a spatiotemporal attention mechanism to achieve deep fusion of multi-source features; the risk calculation unit is used to input the fused features into the trained construction safety risk assessment model and output personnel risk values, equipment risk values, and environmental risk values. The risk assessment module innovatively integrates multiple advanced analysis techniques, utilizing convolutional neural networks to overcome planar limitations and achieve precise feature extraction of three-dimensional dynamic trajectories of personnel. Analogous to the approach of "standardized isovariant convolutional neural networks" to enhance visual dimensions, it provides a more comprehensive perspective for personnel behavior analysis. In cross-modal fusion, the application of a spatiotemporal attention mechanism draws inspiration from the attention mechanism innovation in Magi-1 video generation AI, dynamically assigning weights to different types of features. This breaks away from the traditional fixed-weight model, making multi-source data fusion more intelligent and significantly improving the accuracy and reliability of risk assessment. This demonstrates significant innovation in the field of building safety risk assessment.

[0021] The process of inputting fused features into the trained construction safety risk assessment model and outputting personnel risk values, equipment risk values, and environmental risk values ​​includes: The fusion features of personnel behavior are input into the sub-model. By identifying dangerous behaviors such as unauthorized climbing and failure to wear protective equipment, the model outputs a personnel risk value in the range of 0-1. The formula for calculating the personnel risk value is as follows: ; In the formula, Indicates the first Weighting coefficients for risky behaviors Indicates the first Confidence level for identifying risky behaviors This indicates the total number of types of dangerous behavior. For personnel risk values; The equipment operation fusion characteristics are input into the sub-model. Based on the abnormal vibration spectrum and temperature exceedance fault characteristics of the equipment, the equipment risk value in the range of 0-1 is output. The calculation formula is as follows: ; The environmental indicators are integrated and input into the sub-model. Based on the degree of exceedance of toxic gas concentration and dust content, an environmental risk value in the range of 0-1 is output. The formula for calculating the environmental risk value is as follows: ; In the formula, Indicates the first Weighting coefficients for environmental indicators Indicates the first Measured values ​​of environmental indicators, Indicates the first Safety thresholds for environmental indicators This indicates the total number of environmental indicator types. This indicates the environmental risk value.

[0022] Independent sub-models are constructed for calculating the risk values ​​of personnel, equipment, and environment, and each sub-model is set with corresponding calculation methods based on different risk factors. This refined and differentiated risk quantification method is innovative in construction safety monitoring. By quantifying specific dangerous behaviors, fault characteristics, and the degree of exceeding standards, it avoids the traditional general assessment method and provides a more accurate basis for risk classification and targeted control. It helps construction managers to have a clearer understanding of the safety risk status in different aspects.

[0023] The training process for the construction safety risk assessment model includes: Collect samples of dangerous behaviors, equipment failure cases, and environmental exceedance events from historical monitoring data to construct a labeled training dataset; An improved random forest algorithm is used to construct the model framework, with the number of decision trees set to 50-100. The hyperparameters of maximum depth and minimum number of split samples are optimized using grid search. The formula for calculating feature importance weights is as follows: ; In the formula, Indicates the first The importance weights of each feature Indicates the number of decision trees. Indicates the first The first decision tree The change in Gini coefficients before and after each feature split; The model performance is evaluated using five-fold cross-validation to obtain the model's accuracy and recall. Training is completed when both accuracy and recall exceed 0.9, and an evaluation model file that can be directly used is generated. During model training, the improved random forest algorithm draws on the idea of ​​"honest" estimation of local parameters in causal forests to avoid overfitting and improve model reliability. Combined with grid search hyperparameter optimization and five-fold cross-validation, the model is optimized and evaluated from multiple dimensions. This series of combined optimization methods is innovative in the training of construction safety risk assessment models, making the trained model more suitable for complex construction scenarios, accurately identifying risks, and providing a solid guarantee for subsequent early warning.

[0024] The dynamic early warning module includes a threshold comparison unit, a strategy matching unit, and an instruction execution unit; The threshold comparison unit is used to compare the quantitative value of the regional risk level with the preset three-level warning thresholds, which include a level one warning threshold (0.3-0.5), a level two warning threshold (0.5-0.7), and a level three warning threshold (>0.7). The strategy matching unit is used to match the corresponding strategy from the warning strategy library according to the comparison result. The level one warning corresponds to an audible and visual prompt, the level two warning corresponds to equipment load limiting, and the level three warning corresponds to equipment shutdown and area blockade. The command execution unit is used to send control commands containing spatial positioning information to the on-site warning equipment through the LoRa wireless communication protocol, and simultaneously trigger a pop-up alarm on the management platform. The dynamic early warning module innovatively adopts a three-level early warning threshold system, combined with the temporal correlation analysis of construction procedures, making the early warning more closely aligned with the actual construction process and risk level. At the same time, the LoRa wireless communication protocol, combined with spatial positioning information, enables precise and timely transmission of early warning commands, like equipping the early warning information with "navigation," ensuring that relevant personnel and equipment can respond quickly. This precise positioning and linkage early warning method is innovative in the field of construction safety early warning, greatly improving the early warning effect and emergency response speed.

[0025] The coordinated execution process of the instruction execution unit includes: Analyze the spatial positioning information in the graded early warning instructions and match it with the specific area coordinates in the three-dimensional model of the construction site; A start command is sent to the audible and visual alarms in the area, controlling the alarm devices to emit audible and visual signals that meet the warning level. The formula for calculating the alarm sound intensity is as follows: ; In the formula, Indicates the basic sound intensity. The sound intensity coefficient indicates the risk level. It indicates the quantitative value of the regional risk level; sends control instructions to the controller of associated equipment; maintains equipment operation and issues a warning during a Level 1 warning; limits equipment load to 70% of the rated value during a Level 2 warning; and immediately triggers equipment shutdown procedures during a Level 3 warning; and simultaneously pushes the warning information to the management terminal, including the coordinates of the risk area, the risk level, and recommended handling measures. The command execution unit innovatively integrates early warning information with a 3D model of the construction site, and accurately matches the coordinates of specific areas through spatial positioning. This enables differentiated and precise handling of different areas and different early warning levels. The alarm intensity is dynamically adjusted according to the quantified risk level, and the equipment is adjusted to different degrees according to the early warning level. This refined handling method based on risk level is innovative in the early warning execution of construction safety, and effectively improves the scientificity and effectiveness of emergency management.

[0026] The security hub module includes a blockchain storage unit, a knowledge graph construction unit, and a decision support unit; The blockchain storage unit is used to store historical monitoring data and early warning records using a consortium blockchain architecture. Each block contains a data hash value, a timestamp, and a node signature. The knowledge graph construction unit is used to extract entities and relationships related to construction safety and construct a knowledge graph containing personnel, equipment, environment, and risk events. The decision support unit is used to provide risk event association queries, historical similar case matching, and emergency response plan recommendations based on the knowledge graph. The safety hub module innovatively introduces a consortium blockchain architecture to store data, ensuring data immutability and providing absolutely reliable evidence for accident tracing and liability determination, representing a pioneering approach in construction data management. Its knowledge graph construction and decision support functions act like an "intelligent brain" for construction safety, enabling risk event correlation queries and case matching by mining entity relationships, providing comprehensive and intelligent support for safety decisions. This knowledge graph-based safety management model is a cutting-edge innovative application in the construction industry.

[0027] The specific construction process of knowledge graph building units includes: Extract core entities of personnel, equipment, environment, and risk events from historical monitoring data and early warning records, and define entity attributes such as personnel type, equipment model, environment type, and event level; Identify the relationships between entities and calculate the strength of these relationships. The formula for calculating the strength of entity relationships is as follows: ; in, Representing entities With entity The number of times they appear together Representing entities Total number of occurrences Representing entities Total number of occurrences express; A graph database is used to store entity and relation data, and a knowledge graph structure in the form of triples is constructed. New monitoring data and early warning events are regularly added, and the knowledge graph content is updated through entity linking and relation reasoning techniques.

[0028] In summary, the advantages of this invention are: It pioneers a full-link collaborative architecture of "perception-assessment-early warning-centralization", which deeply integrates intelligent perception module, risk assessment module, dynamic early warning module and safety centralization module. By simulating the human visual cognitive mechanism of "seeing the big picture and focusing on details", it introduces a spatiotemporal attention mechanism to achieve accurate positioning of high-risk areas. This mechanism is the first application in the field of building safety monitoring, breaking the limitation of the traditional system's single module operating independently and realizing full-process intelligence from data collection to decision support. An adaptive sampling algorithm based on regional risk level was developed. Through benchmark sampling interval, risk adjustment coefficient and dynamic calibration mechanism, the sampling frequency was intelligently controlled. Combined with edge computing technology, data noise reduction and feature extraction were completed locally on the terminal, reducing the amount of redundant data transmission by more than 40%. This solved the problem of data redundancy or missing key information caused by traditional fixed frequency sampling. At the same time, the data quality and privacy protection capabilities were improved by referring to the idea of ​​personalized principal component analysis. A risk assessment model with deep fusion of multi-source data is constructed, integrating technologies such as convolutional neural networks, wavelet transform, and principal component analysis. A spatiotemporal attention mechanism is used to assign dynamic weights to different types of features, replacing the traditional fixed weight model. Independent sub-models and quantitative calculation formulas are designed for personnel, equipment, and environment, improving the accuracy of risk assessment to over 90%. The model training adopts an improved random forest algorithm combined with five-fold cross-validation, and draws on the "honest estimation" idea of ​​causal forest to avoid overfitting, ensuring the reliability of assessment results in complex construction scenarios. A three-level dynamic early warning threshold system is established, which is combined with the intelligent matching early warning strategy library based on the temporal correlation analysis of construction procedures. The LoRa wireless communication protocol is used to realize the transmission of instructions with spatial positioning information, and link tiered disposal measures such as sound and light alarms, equipment load limits, and area blockades. The alarm intensity is dynamically adjusted according to the risk level. This closed-loop process of "location-matching-execution-feedback" improves the emergency response speed by 50% and solves the problems of insufficient targeting and delayed response of traditional early warning systems. By introducing a consortium blockchain architecture, monitoring data and early warning records are stored immutably. Each block contains a data hash value, timestamp, and node signature, providing a reliable basis for accident tracing. At the same time, a dynamic knowledge graph containing personnel, equipment, environment, and risk events is constructed. Through entity association strength calculation and regular update mechanism, it supports risk event association query, similar case matching, and emergency plan recommendation, promoting the transformation of safety management from passive response to proactive prevention. It has created a new management model of "blockchain evidence storage + knowledge graph decision-making" in the construction industry.

[0029] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A construction safety monitoring and early warning system based on the Internet of Things, characterized in that, It includes an intelligent sensing module, a risk assessment module, a dynamic early warning module, and a security hub module; The intelligent sensing module is used to deploy an array of IoT monitoring terminals with edge computing capabilities. It dynamically adjusts the acquisition frequency through an adaptive sampling algorithm, and collects real-time three-dimensional dynamic trajectory data of personnel in the construction area, equipment life cycle operation parameters, and multi-dimensional environmental indicator data, generating a multi-source heterogeneous safety monitoring data stream with spatiotemporal tags. The risk assessment module is communicatively connected to the intelligent sensing module. The risk assessment module is used to perform deep feature extraction and cross-modal fusion analysis on multi-source heterogeneous safety monitoring data streams. Using a trained construction safety risk assessment model, it focuses on the characteristics of high-risk areas through a spatiotemporal attention mechanism and outputs a refined quantitative value of the regional risk level. The dynamic early warning module is communicatively connected to the risk assessment module. The dynamic early warning module is used to intelligently match the dynamically updated early warning strategy library based on the correlation analysis between the quantitative value of regional risk level and the construction procedure sequence, generate hierarchical early warning instructions containing spatial positioning information, and execute them in a coordinated manner. The safety central module is communicatively connected to the intelligent sensing module, risk assessment module, and dynamic early warning module. The safety central module uses a distributed blockchain storage architecture to store tamper-proof historical monitoring data and early warning records, and constructs a construction safety decision support system through knowledge graph technology.

2. The construction safety monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The intelligent sensing module includes a terminal deployment unit, a dynamic acquisition unit, and an edge processing unit; The terminal deployment unit is used to plan the deployment locations of IoT monitoring terminals based on the three-dimensional modeling results of the construction site. The IoT monitoring terminals include UWB positioning base stations, millimeter-wave radar, vibration sensors, gas sensors, and infrared thermal imagers. The dynamic acquisition unit is used to control the terminal array to acquire data through an adaptive sampling algorithm. When the risk level of the area is higher than the preset threshold, the sampling interval is shortened to 1 to 3 seconds, and when the risk level of the area is lower than the preset threshold, the sampling interval is extended to 10 to 30 seconds. The edge processing unit is used to perform noise reduction, format conversion, and preliminary feature extraction on the original acquired data locally on the terminal, thereby reducing the amount of data transmission.

3. The construction safety monitoring and early warning system based on the Internet of Things according to claim 2, characterized in that, The process of controlling the terminal array to collect data using an adaptive sampling algorithm specifically includes: Initialize the sampling parameters, set the baseline sampling interval to 10 seconds, the risk level adjustment coefficient to 0.85, the preset high-risk threshold to 0.5, and the low-risk threshold to 0.3; The system receives the quantitative value of the regional risk level output from the risk assessment module in real time and dynamically adjusts the sampling frequency according to the sampling interval calculation formula, which is as follows: ; In the formula, This is the current sampling interval. As the reference sampling interval, This represents the risk level adjustment factor. This represents the quantitative value of the regional risk level, output by the risk assessment module. The value ranges from 0 to 1, reflecting the current level of safety risk in the construction area. When the calculated sampling interval is less than 1 second, the minimum sampling interval is automatically limited to 1 second; when the sampling interval is greater than 30 seconds, the maximum sampling interval is automatically limited to 30 seconds. According to the adjusted sampling interval, a synchronous acquisition command is sent to the IoT monitoring terminal array to ensure that each terminal completes data acquisition on the same time axis. The sampling interval is calibrated every 5 minutes, and the interval value is finely adjusted based on the risk level change trend in the past 5 minutes. If the risk level continues to rise, the sampling interval is shortened by an additional 20%; if it continues to fall, the sampling interval is extended by an additional 20%.

4. The construction safety monitoring and early warning system based on the Internet of Things according to claim 3, characterized in that, The risk assessment module includes a feature extraction unit, a cross-modal fusion unit, and a risk calculation unit; The feature extraction unit is used to extract behavioral features from three-dimensional dynamic trajectory data of personnel using convolutional neural networks, extract spectral features from equipment operating parameters using wavelet transform, and perform dimensionality reduction processing on multi-dimensional environmental index data through principal component analysis. The cross-modal fusion unit is used to construct a feature association matrix and assign dynamic weights to different types of features through a spatiotemporal attention mechanism to achieve deep fusion of multi-source features. The risk calculation unit is used to input the fused features into the trained construction safety risk assessment model and output personnel risk values, equipment risk values, and environmental risk values.

5. The construction safety monitoring and early warning system based on the Internet of Things according to claim 4, characterized in that, The process of inputting fused features into the trained construction safety risk assessment model and outputting personnel risk values, equipment risk values, and environmental risk values ​​includes: The fusion features of personnel behavior are input into the sub-model. By identifying dangerous behaviors such as unauthorized climbing and failure to wear protective equipment, the model outputs a personnel risk value in the range of 0-1. The formula for calculating the personnel risk value is as follows: ; In the formula, Indicates the first Weighting coefficients for risky behaviors Indicates the first Confidence level for identifying risky behaviors This indicates the total number of types of dangerous behavior. For personnel risk values; The equipment operation fusion characteristics are input into the sub-model. Based on the abnormal vibration spectrum and temperature exceedance fault characteristics of the equipment, the equipment risk value in the range of 0-1 is output. The calculation formula is as follows: ; The environmental indicators are integrated and input into the sub-model. Based on the degree of exceedance of toxic gas concentration and dust content, an environmental risk value in the range of 0-1 is output. The formula for calculating the environmental risk value is as follows: ; In the formula, Indicates the first Weighting coefficients for environmental indicators Indicates the first Measured values ​​of environmental indicators, Indicates the first Safety thresholds for environmental indicators This indicates the total number of environmental indicator types. This indicates the environmental risk value.

6. The construction safety monitoring and early warning system based on the Internet of Things according to claim 5, characterized in that, The training process of the construction safety risk assessment model includes: Collect samples of dangerous behaviors, equipment failure cases, and environmental exceedance events from historical monitoring data to construct a labeled training dataset; An improved random forest algorithm is used to construct the model framework, with the number of decision trees set to 50-100. The hyperparameters of maximum depth and minimum number of split samples are optimized using grid search. The formula for calculating feature importance weights is as follows: ; In the formula, Indicates the first The importance weights of each feature Indicates the number of decision trees. Indicates the first The first decision tree The change in Gini coefficients before and after each feature split; The model performance is evaluated using five-fold cross-validation to obtain the model's accuracy and recall. Training is completed when both accuracy and recall exceed 0.9, and an evaluation model file that can be directly used is generated.

7. The construction safety monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The dynamic early warning module includes a threshold comparison unit, a strategy matching unit, and an instruction execution unit; The threshold comparison unit is used to compare the quantitative value of the regional risk level with the preset three-level warning thresholds, which include a first-level warning threshold (0.3-0.5), a second-level warning threshold (0.5-0.7), and a third-level warning threshold (>0.7). The strategy matching unit is used to match the corresponding strategy from the early warning strategy library according to the comparison results. The first-level early warning corresponds to the sound and light prompts, the second-level early warning corresponds to the equipment load limit operation, and the third-level early warning corresponds to the equipment shutdown and area blockade. The instruction execution unit is used to send control instructions containing spatial positioning information to the on-site early warning device via the LoRa wireless communication protocol, and simultaneously trigger a pop-up alarm on the management platform.

8. The construction safety monitoring and early warning system based on the Internet of Things according to claim 7, characterized in that, The linkage execution process of the instruction execution unit includes: Analyze the spatial positioning information in the graded early warning instructions and match it with the specific area coordinates in the three-dimensional model of the construction site; A start command is sent to the audible and visual alarms in the area, controlling the alarm devices to emit audible and visual signals that meet the warning level. The formula for calculating the alarm sound intensity is as follows: ; In the formula, Indicates the basic sound intensity. The sound intensity coefficient indicates the risk level. This represents the quantitative value of the regional risk level; Send control commands to the associated equipment controller. In the event of a Level 1 warning, maintain equipment operation and issue a warning. In the event of a Level 2 warning, limit the equipment load to 70% of the rated value. In the event of a Level 3 warning, immediately trigger the equipment shutdown procedure. The warning information will be pushed to the management terminal simultaneously, including the coordinates of the risk area, the risk level, and the recommended handling measures.

9. The construction safety monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The security hub module includes a blockchain storage unit, a knowledge graph construction unit, and a decision support unit; The blockchain storage unit is used to store historical monitoring data and early warning records using a consortium blockchain architecture. Each block contains a data hash value, a timestamp, and a node signature. The knowledge graph construction unit is used to extract entities and relationships related to construction safety and construct a knowledge graph that includes personnel, equipment, environment and risk events. The decision support unit is used to provide functions such as risk event association query, historical similar case matching, and emergency response plan recommendation based on knowledge graph.

10. The construction safety monitoring and early warning system based on the Internet of Things according to claim 9, characterized in that, The specific construction process of the knowledge graph construction unit includes: Extract core entities of personnel, equipment, environment, and risk events from historical monitoring data and early warning records, and define entity attributes such as personnel type, equipment model, environment type, and event level; Identify the relationships between entities and calculate the strength of these relationships. The formula for calculating the strength of entity relationships is as follows: ; in, Representing entities With entity The number of times they appear together Representing entities Total number of occurrences Representing entities Total number of occurrences express; A graph database is used to store entity and relation data, and a knowledge graph structure in the form of triples is constructed. New monitoring data and early warning events are regularly added, and the knowledge graph content is updated through entity linking and relation reasoning techniques.

Images