Method for assessing risks of workers based on Beidou positioning and physiological characteristics fusion

Abstract

The application relates to the technical field of state evaluation, in particular to a worker risk evaluation method based on Beidou positioning and physiological feature fusion, which firstly establishes a communication connection between an intelligent receiving end and a panoramic monitoring end, acquires real-time position and heart rate data of a worker; then maps the position data to a dynamically updated three-dimensional simulation model, calculates a spatial absolute distance from a work area boundary to determine a spatial risk state; simultaneously, a dynamic reference heart rate is predicted by using a BP neural network in combination with spatial and time sequence features, a heart rate deviation degree of a fluctuation extreme value and the reference heart rate is calculated to determine a physiological risk state; finally, a linkage early warning instruction is generated by comprehensively integrating a spatial risk state, a physiological risk state and a visual risk state extracted based on a video stream. The application realizes multi-dimensional accurate quantitative evaluation and early warning of personnel risk in a complex construction scene.

Description

Technical Field

[0001] This application relates to the field of condition assessment technology, and in particular to a risk assessment method for workers based on the fusion of BeiDou positioning and physiological characteristics. Background Technology

[0002] The scale of power industry engineering construction continues to expand, and production and maintenance tasks are becoming increasingly heavy. Power operation sites encompass various high-risk work scenarios, including working at heights and live-line operations, and involve densely populated personnel and complex operating procedures. Comprehensive and real-time monitoring of violations at work sites is essential to ensuring the personal safety of workers.

[0003] Currently, to meet the safety supervision needs of power construction sites, a combination of online video monitoring and offline manual inspections is typically used. For online monitoring, this mainly relies on surveillance cameras installed at fixed locations on the construction site, allowing back-end safety management personnel to observe the operational behavior of on-site personnel through screens. For monitoring personnel's physiological state, portable wristbands with basic heart rate monitoring functions are provided to workers. When the heart rate value collected by the wristband exceeds a pre-set single static threshold, an alarm is triggered, enabling monitoring and intervention for abnormal health conditions of the workers.

[0004] However, traditional video surveillance has serious blind spots, making it difficult to accurately determine the true three-dimensional spatial distance between personnel and the boundary of live objects or dangerous areas, which can easily lead to misjudgment or omission of dangerous distances. Secondly, the labor intensity of construction workers is constantly changing, and the use of a single static heart rate threshold is affected by normal strenuous physical labor, resulting in a high false alarm rate for physiological risk warnings, making it difficult to achieve accurate risk assessment in complex construction environments. Summary of the Invention

[0005] To address the technical problems of existing monitoring methods having visual blind spots and being prone to false alarms due to physical labor interference, this application provides a risk assessment method for workers based on the fusion of BeiDou positioning and physiological characteristics.

[0006] This application provides a method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics. The assessment method includes: establishing a communication connection between an intelligent receiver and a panoramic monitoring terminal; acquiring real-time location data and real-time heart rate data of workers via the intelligent receiver; mapping the real-time location data onto a pre-constructed three-dimensional simulation model; calculating the absolute spatial distance between the real-time location data and the boundary of a preset work area; determining the spatial risk status of workers when the absolute spatial distance is less than a distance threshold; calculating the extreme fluctuation value of the real-time heart rate data within a preset time window; calculating the difference between the extreme fluctuation value and the dynamic reference heart rate to obtain the heart rate deviation; determining the physiological risk status of workers when the heart rate deviation is greater than a deviation threshold; and generating a linkage early warning command by combining the spatial risk status and the physiological risk status.

[0007] Spatial risks are determined by calculating spatial distance in a 3D model, and physiological risks are determined by calculating heart rate deviation using a dynamic benchmark. In turn, a comprehensive early warning is generated, realizing cross-validation between spatial location and physiological stress. This effectively isolates heart rate interference caused by normal physical labor and significantly reduces the false alarm rate of safety warnings under complex working conditions.

[0008] Preferably, before acquiring the real-time location data and real-time heart rate data of the workers, an identity access step is also included: acquiring the facial image data of the workers; comparing the facial image data with the feature points of the preset system personnel database to obtain a verification result value; when the verification result value is a successful match, generating an access activation order; and in response to the access activation order, acquiring the real-time location data and the real-time heart rate data.

[0009] Preferably, before generating a joint warning instruction by combining the spatial risk state and the physiological risk state, a visual recognition step is further included: acquiring a dynamic video stream from the site; inputting the dynamic video stream into a pre-trained visual recognition network to extract action feature values ​​from the dynamic video stream; comparing the action feature values ​​with a preset database of illegal actions, and determining the visual risk state of the worker when a match is found; and generating the joint warning instruction by combining the spatial risk state, the physiological risk state, and the visual risk state.

[0010] Preferably, inputting the dynamic video stream into a pre-trained visual recognition network to extract action feature values ​​from the dynamic video stream includes: segmenting the dynamic video stream into multiple single-frame images in chronological order; inputting each single-frame image into a two-dimensional convolutional layer of the visual recognition network to extract basic feature maps; inputting the basic feature maps into a time-shifted feature layer to extract features along the time dimension to obtain a temporal fusion map; inputting the temporal fusion map into a cross-channel attention layer to calculate the weight values ​​of each feature channel; and using the weight values ​​to perform matrix multiplication on the temporal fusion map to obtain action feature values.

[0011] Preferably, establishing a communication connection between the intelligent receiver and the panoramic monitoring end includes: acquiring multiple candidate protocols supported by the intelligent receiver and the panoramic monitoring end; monitoring the frequency band occupancy and network latency of the on-site environment; selecting a target protocol from the multiple candidate protocols using an intelligent scheduling algorithm based on the frequency band occupancy and the network latency; configuring the intelligent receiver and the panoramic monitoring end to perform handshake matching using the target protocol to construct the multi-protocol fusion communication connection.

[0012] By monitoring frequency band occupancy and network latency, the system can select target protocols from candidate protocols, enabling it to adaptively switch communication links according to the complex electromagnetic environment on site. This avoids disconnection due to interference with a single protocol, significantly enhancing the anti-interference capability of terminal data transmission and the stability of network connection.

[0013] Preferably, the configuration of the intelligent receiver and the panoramic monitoring end to perform handshake matching using the target protocol to construct the multi-protocol fusion communication connection includes: generating public key data and private key data based on the elliptic curve algorithm; exchanging the public key data between the intelligent receiver and the panoramic monitoring end; performing hash operations on the real-time location data and the real-time heart rate data using the private key data to generate a verification feature value; appending the verification feature value to the transmission data packet and transmitting it through the target protocol to construct the communication connection with cross-protocol security protection.

[0014] Preferably, before mapping the real-time location data to the pre-constructed three-dimensional simulation model, a model synchronization step is also included: real-time acquisition of the equipment operation values ​​of the physical equipment on site; conversion of the equipment operation values ​​into a digital drive stream; input of the digital drive stream into the initial basic model; adjustment of the size parameters and state parameters of the corresponding virtual equipment in the initial basic model according to the digital drive stream; and generation of the three-dimensional simulation model with synchronized virtual and real states.

[0015] By collecting physical equipment operating values ​​and converting them into digital drive flows, and dynamically adjusting the virtual equipment parameters in the initial basic model, a three-dimensional simulation environment with synchronized virtual and real states is constructed. This enables the assessment of personnel space risks based on high-fidelity dynamic boundaries, significantly improving the absolute accuracy of space collision avoidance warnings in complex moving mechanical scenarios. Preferably, the linkage warning command is generated by combining the spatial risk status and the physiological risk status, including: calculating the difference between the distance threshold and the absolute spatial distance to obtain the spatial urgency; extracting the heart rate deviation; obtaining a pre-set spatial weight value and a physiological weight value; multiplying the spatial urgency by the spatial weight value to obtain a first product value, and multiplying the heart rate deviation by the physiological weight value to obtain a second product value; adding the first product value and the second product value to obtain a comprehensive risk value; and generating a linkage warning command containing a danger identification code when the comprehensive risk value is greater than a preset alarm threshold.

[0016] Preferably, the calculation of the dynamic baseline heart rate includes: inputting the spatial amplitude parameter and the temporal change frequency within a preset time window into the heart rate prediction model, and using the output of the heart rate prediction model as the dynamic baseline heart rate, wherein the heart rate prediction model is trained based on the historical data of the operator.

[0017] Preferably, the heart rate prediction model is trained based on the historical data of the workers, including: the heart rate prediction model is a BP neural network; the historical data includes spatial amplitude parameters and temporal change frequency within any preset time window, as well as the real heart rate collected in real time; outlier detection is performed on the historical data, and after deleting the outlier data, the heart rate prediction model is trained using the gradient descent method.

[0018] Compared with the prior art, the technical solution of this application has the following beneficial technical effects: By mapping real-time location to a 3D model to calculate spatial distance, and combining dynamic heart rate deviation for physiological assessment and generating linked early warnings, this approach overcomes the blind spots and misjudgment defects of traditional single monitoring, and realizes multimodal quantitative assessment of the risk status of complex power workers, thereby improving the accuracy of anti-violation early warning and the effectiveness of on-site safety management. Attached Figure Description

[0019] Figure 1 This is a flowchart of a risk assessment method for workers based on the fusion of BeiDou positioning and physiological characteristics, according to an embodiment of this application. Detailed Implementation

[0020] The following describes in detail the worker risk assessment method based on the fusion of BeiDou positioning and physiological characteristics provided in this application embodiment, using a specific application scenario. For example, this application scenario can be a power construction site, such as a substation construction area or a high-voltage line maintenance site. In this scenario, workers are equipped with intelligent receivers, and panoramic monitoring devices are installed at preset locations within the construction site to provide real-time and comprehensive safety monitoring of the workers' behavior, status, and location. Figure 1 This is a flowchart illustrating a method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics, according to an embodiment of this application. Figure 1 As shown, the risk assessment method for workers based on the fusion of BeiDou positioning and physiological characteristics includes steps S101 to S104, which are described in detail below.

[0021] S101, establish a communication connection between the intelligent receiver and the panoramic monitoring terminal, and obtain the real-time location data and real-time heart rate data of the operator through the intelligent receiver.

[0022] In one embodiment, to ensure that only authorized personnel can participate in power operations and enhance the standardization of on-site operations, an identity verification step is performed before acquiring personnel status data. Specifically, this includes: collecting facial image data of the personnel; comparing the facial image data with feature points in a preset system personnel database to obtain a verification result value; generating an access activation order when the verification result value is a successful match; and acquiring the real-time location data and real-time heart rate data in response to the access activation order.

[0023] Understandably, the panoramic monitoring terminal captures facial image data of workers entering the site passageway. The system's personnel database pre-stores the biometric information of legitimate construction personnel. By comparing the facial image data with key facial feature points in the system's personnel database, the similarity between the two is calculated. When the similarity is greater than 90%, the verification result is considered a successful match, triggering an access activation order. The access activation order wakes up the BeiDou positioning module and heart rate sensor inside the intelligent receiver, thus beginning the continuous collection of real-time location data and real-time heart rate data.

[0024] To improve communication anti-interference capabilities in complex construction electromagnetic environments and overcome the limitations of single communication protocols in terms of coverage or power consumption, dynamic scheduling of multiple protocols is performed when establishing a communication connection between the intelligent receiver and the panoramic monitoring end. Specifically, this includes: acquiring multiple candidate protocols supported by both the intelligent receiver and the panoramic monitoring end; monitoring the frequency band occupancy and network latency of the site environment; using an intelligent scheduling algorithm to select a target protocol from the candidate protocols based on the frequency band occupancy and network latency; configuring the intelligent receiver and the panoramic monitoring end to perform a handshake matching using the target protocol to construct a multi-protocol fusion communication connection.

[0025] The alternative protocols include Bluetooth Low Energy (BLE), Wi-Fi, and Wireless Fidelity. The panoramic monitoring device collects real-time information on channel congestion in the space to determine frequency band occupancy and calculates round-trip time by sending probe data packets to determine network latency. For example, priorities are set according to task requirements. When the Wi-Fi frequency band occupancy is detected to be higher than 80% and the network latency is greater than 100 milliseconds, it is determined that the current broadband channel is restricted, and the target protocol is automatically switched to the more interference-resistant and low-power Wi-Fi protocol, thereby ensuring the continuity of the communication connection.

[0026] To prevent data from being eavesdropped on or tampered with during transmission and to meet the high-security requirements of cross-protocol transmission, further encryption protection is implemented when constructing a multi-protocol integrated communication connection. Specifically, this includes generating public and private key data based on an elliptic curve algorithm; exchanging the public key data between the intelligent receiving end and the panoramic monitoring end; using the private key data to perform a hash operation on the real-time location data and the real-time heart rate data to generate a verification feature value; appending the verification feature value to the transmission data packet and transmitting it through the target protocol to construct the communication connection with cross-protocol security protection.

[0027] It should be noted that the elliptic curve algorithm is a standard asymmetric encryption algorithm. By exchanging public key data, the intelligent receiver and the panoramic monitoring end establish a secure encrypted channel. Subsequently, the intelligent receiver performs a standard secure hash operation on the extracted real-time location data and real-time heart rate data to generate a 256-bit verification feature value. After receiving the transmitted data packet, the panoramic monitoring end uses the public key data to decrypt the verification feature value and perform a reverse comparison. If the hash comparison matches, it confirms that the data has not been tampered with.

[0028] S102, map the real-time location data to a pre-built three-dimensional simulation model, calculate the absolute spatial distance between the real-time location data and the boundary of the preset work area, and determine the spatial risk status of the worker when the absolute spatial distance is less than a distance threshold.

[0029] In one embodiment, to reproduce the dynamic changes of the real power operation environment, model synchronization is first performed before spatial mapping. Specifically, this includes: real-time acquisition of equipment operating values ​​from on-site physical devices; conversion of these operating values ​​into digital drive streams; inputting the digital drive streams into an initial base model; adjusting the size and state parameters of the corresponding virtual devices in the initial base model based on the digital drive streams; and generating a three-dimensional simulation model that synchronizes virtual and real states.

[0030] It should be noted that the physical equipment on site can include generators, transformers, and cranes. Sensors deployed on the equipment acquire real-time data such as current, voltage, and robotic arm deflection angle as operating values. The collected data is cleaned and standardized to transform it into a unified format digital drive stream. The initial basic model is a pre-established static 3D scene. After inputting the digital drive stream, digital twin simulation technology is used to dynamically update the energized state parameters of the virtual transformer and the mechanical cantilever size parameters of the virtual crane online, thereby constructing a 3D simulation model that converges at the same rate as the real physical world.

[0031] Based on a 3D simulation model that synchronizes virtual and real states, real-time position data from an intelligent receiver is mapped to the global coordinate system of this 3D simulation model to calculate the absolute spatial distance between the real-time position data and the preset work area boundary. Due to the complex environment of power construction sites, the boundary of the work area is usually an irregular 3D curved surface or polygonal boundary. To accurately measure the approach of the workers, the absolute spatial distance satisfies the following relationship: In the formula, Represents absolute spatial distance; , , These represent the three-dimensional coordinate components of the real-time position data after being mapped to the three-dimensional simulation model; This indicates the total number of discrete sampling points on the boundary of the preset work area. The total number can be set to 1000 to ensure the accuracy of the dangerous boundary depiction. , , They represent the first The formula calculates the three-dimensional coordinate components of each discrete sampling point. It understandably works by iterating through all sampling points on the boundary of the work area and finding the minimum Euclidean distance between the worker's current position and all discrete sampling points. This minimum distance is used to reflect the shortest straight-line distance between the worker and the hazardous boundary. When the calculated absolute spatial distance is less than a distance threshold, it is determined that the worker is about to cross the safety fence or enter a high-voltage hazardous area, thus determining the worker's spatial risk status.

[0032] For example, in the case of 110 kV transformer maintenance, the distance threshold is preferably set to 1.5 meters. When the real-time location data of the operator is less than 1.5 meters away from the boundary of the live body, the spatial risk status is immediately triggered and determined.

[0033] In this way, the real-time collected equipment operating values ​​are transformed into digital drive streams and used to drive the initial basic model, thus constructing a high-precision and virtual-real synchronized three-dimensional simulation model. Combined with spatial coordinate mapping and the shortest spatial absolute distance, the visual blind spots and distance misjudgments existing in two-dimensional video surveillance are effectively eliminated, and the accuracy and timeliness of preventing personnel from crossing boundaries in complex three-dimensional construction sites are improved.

[0034] S103, calculate the extreme value of the fluctuation of the real-time heart rate data within a preset time window, calculate the difference between the extreme value of the fluctuation and the dynamic benchmark heart rate to obtain the heart rate deviation, and determine the physiological risk status of the worker when the heart rate deviation is greater than the deviation threshold.

[0035] In one embodiment, because the labor intensity of construction site workers is dynamically changing—for example, the resting state during climbing ladders is vastly different from that during inspections on flat ground—using a fixed baseline heart rate is prone to false alarms. Therefore, it is necessary to predict a dynamic heart rate baseline by combining the worker's activity status. Specifically, this involves inputting spatial amplitude parameters and temporal variation frequencies within a preset time window into a heart rate prediction model, using the output of the heart rate prediction model as the dynamic baseline heart rate, and training the heart rate prediction model based on the worker's historical data.

[0036] The heart rate prediction model is trained based on the historical data of the workers, including: the heart rate prediction model is a BP neural network; the historical data includes spatial amplitude parameters and temporal change frequency within any preset time window, as well as the real heart rate collected in real time; outlier detection is performed on the historical data, and after deleting the outlier data, the heart rate prediction model is trained using the gradient descent method.

[0037] It should be noted that within a preset time window, the accelerometer collects acceleration data in real time along three axes and calculates the resultant vector magnitude of the three-axis acceleration. The difference between the maximum and minimum values ​​of the resultant vector magnitude within the preset time window is extracted as a spatial amplitude parameter to characterize the intensity of the worker's limb movements. Simultaneously, a barometer collects the fluctuation frequency of air pressure values ​​within the same preset time window, using this as the temporal variation frequency to characterize the frequency of movement of the worker in vertical space, such as during tower climbing operations. The preferred preset time window value is 60 seconds.

[0038] During the model training phase, historical real heart rates collected synchronously by the intelligent receiver during past operations are collected as a sample set. The standard isolated forest algorithm is used for outlier detection, eliminating abnormal heart rate data caused by poor sensor contact or signal loss. Subsequently, the standard backpropagation algorithm of a backpropagation neural network is employed, continuously updating the network node weights using gradient descent until the loss function is less than the minimum allowable value or the number of iterations exceeds the maximum number. This yields a trained heart rate prediction model capable of fitting the physical characteristics of a specific worker. The minimum allowable value is 0.01, and the maximum number of iterations is 100. Each worker corresponds to a unique heart rate prediction model.

[0039] After inputting the spatial amplitude parameter and temporal variation frequency within the current preset time window into the heart rate prediction model to obtain the dynamic baseline heart rate, it is necessary to evaluate the extreme values ​​of the current real-time heartbeat. Extreme fluctuation values. Satisfying the relation: In the formula, Indicates the preset time window; This represents the discrete sampling moments within a preset time window; Indicates in Real-time heart rate data is constantly acquired by the intelligent receiver.

[0040] Subsequently, the heart rate deviation was further calculated, and the heart rate deviation satisfies the following relationship: In the formula, Indicates the degree of heart rate deviation; Indicates extreme values ​​of fluctuation; This represents the dynamic baseline heart rate output by the heart rate prediction model.

[0041] Understandably, since the dynamic baseline heart rate already includes the expected model of the worker's heart rate acceleration under the current labor intensity, the heart rate deviation calculated by this difference formula effectively isolates the physiological fluctuations brought about by normal physical labor and purely reflects the physiological stress response of the worker when overly tense or encountering danger.

[0042] Thus, by introducing personalized adaptive prediction of dynamic baseline heart rate through BP neural network and combining it with the cleaning and elimination of historical abnormal data, the defect of static baseline heart rate being easily affected by construction intensity is overcome, and accurate assessment of physiological risk status is achieved, reducing the false alarm rate of early warning caused by strenuous labor in complex power construction environment.

[0043] S104, Based on the spatial risk status and the physiological risk status, generate a linkage early warning command.

[0044] In one embodiment, to more comprehensively capture violations at the work site, a visual recognition step is included before generating a coordinated early warning command by combining the spatial risk status and the physiological risk status. Specifically, this includes: acquiring a dynamic video stream from the site; inputting the dynamic video stream into a pre-trained visual recognition network to extract action feature values ​​from the dynamic video stream; comparing the action feature values ​​with a preset database of violations, and determining the visual risk status of the worker when a match is found; and generating the coordinated early warning command by combining the spatial risk status, the physiological risk status, and the visual risk status.

[0045] Further, in the above visual recognition step, the dynamic video stream is input into a pre-trained visual recognition network to extract action feature values ​​from the dynamic video stream, including: dividing the dynamic video stream into multiple single-frame images in chronological order; inputting each single-frame image into a two-dimensional convolutional layer of the visual recognition network to extract basic feature maps; inputting the basic feature maps into a time-shifted feature layer to extract features along the time dimension to obtain a temporal fusion map; inputting the temporal fusion map into a cross-channel attention layer to calculate the weight values ​​of each feature channel; and using the weight values ​​to perform matrix multiplication on the temporal fusion map to obtain action feature values.

[0046] It should be noted that the panoramic monitoring terminal acquires dynamic video streams from the site in real time. The two-dimensional convolutional layer uses a standard residual network structure to extract spatial static features from single-frame images. The time-shifting feature layer can use recurrent neural networks such as LSTM to fuse the basic feature maps of each single-frame image into a temporal fusion map. The cross-channel attention layer achieves local cross-channel information interaction through a one-dimensional convolutional structure, autonomously learning and calculating the importance ratio of each feature channel, i.e., the weight value; using this weight value to perform matrix multiplication on the temporal fusion map effectively reduces background interference information and highlights core pixel features strongly correlated with the violation, thereby outputting high-precision action feature values. The preset violation action library contains feature vectors of standard violations such as not wearing a safety helmet, crossing a fence, and climbing an unmanned escalator. Once a match is found, the visual risk state of the person is immediately locked.

[0047] After acquiring the spatial risk status, physiological risk status, and visual risk status, a quantitative comprehensive risk assessment is performed to quantify and fuse the risks across multiple modalities, accurately output the final warning signal, and conduct the following steps: Calculate the difference between the distance threshold and the absolute spatial distance to obtain the spatial urgency; extract the heart rate deviation; obtain pre-set spatial weight values ​​and physiological weight values; multiply the spatial urgency by the spatial weight value to obtain a first product value, and multiply the heart rate deviation by the physiological weight value to obtain a second product value; add the first product value and the second product value to obtain a comprehensive risk value; when the comprehensive risk value exceeds a preset alarm threshold, generate a linked warning command containing a danger identification code.

[0048] The overall risk value satisfies the following relationship: In the formula, This represents the overall risk value; This represents a pre-set spatial weight value, with a preferred value of 0.4; Indicates the distance threshold; Represents absolute spatial distance; This represents a pre-set physiological weight value, with a preferred value of 0.6; Indicates the degree of heart rate deviation. This is the dynamic baseline heart rate.

[0049] Understandably, a linear weighted matrix is ​​used to quantify and aggregate the approach risk in the spatial dimension and the stress risk in the physiological dimension. The spatial urgency is calculated by subtracting the absolute spatial distance from the distance threshold, which intuitively reflects the urgency of a person's approach to the danger boundary. The risk fusion of cross-modal data is completed by multiplying the spatial urgency and heart rate deviation by their respective weights and then summing them.

[0050] For example, the preferred value for the alarm threshold is set to 0.8. When the overall risk value is greater than 0.8 and a visual risk state is triggered simultaneously, the system determines that the current risk level is extremely high and then generates a linkage warning command containing a danger identification code. The danger identification code contains the unique identification number of the violator, three-dimensional spatial coordinates, and violation type code, which is received by the intelligent receiver and triggers a high-frequency vibration and voice alarm.

[0051] In this way, by deeply integrating the action features output by the visual recognition network, the spatial urgency extracted by Beidou positioning, and the physiological stress features extracted by heart rate monitoring, the bottleneck of missed and false alarms caused by a single technical means is effectively avoided, and the reliability and timeliness of anti-violation early warning at the power operation site are improved.

[0052] It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application shall be determined by the appended claims.

Claims

1. A risk assessment method for workers based on the fusion of BeiDou positioning and physiological characteristics, characterized in that, The evaluation method includes: Establish a communication connection between the intelligent receiver and the panoramic monitoring terminal, and obtain the real-time location data and real-time heart rate data of the workers through the intelligent receiver; The real-time location data is mapped onto a pre-built three-dimensional simulation model, and the absolute spatial distance between the real-time location data and the boundary of the preset work area is calculated. When the absolute spatial distance is less than a distance threshold, the spatial risk status of the worker is determined. The extreme fluctuation value of the real-time heart rate data within a preset time window is calculated, and the difference between the extreme fluctuation value and the dynamic baseline heart rate is calculated to obtain the heart rate deviation. When the heart rate deviation is greater than the deviation threshold, the physiological risk status of the worker is determined. Based on the combined spatial risk status and physiological risk status, a coordinated early warning command is generated.

2. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 1, characterized in that, Before acquiring the real-time location and heart rate data of the workers, an identity verification process is also included: Collect facial image data of the workers; The facial image data is compared with the feature points of the preset system personnel database to obtain the verification result value; When the verification result is a successful match, an access activation order is generated; In response to the access activation command, the real-time location data and the real-time heart rate data are acquired.

3. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 1, characterized in that, Before generating a coordinated early warning command by comprehensively considering the spatial risk status and the physiological risk status, a visual recognition step is also included: The dynamic video stream captured on-site; The dynamic video stream is input into a pre-trained visual recognition network to extract action feature values ​​from the dynamic video stream; The action feature value is compared with a preset database of illegal actions. When a match is found, the visual risk state of the worker is determined. The linkage warning command is generated by combining the spatial risk status, the physiological risk status, and the visual risk status.

4. The worker risk assessment method based on the fusion of BeiDou positioning and physiological characteristics according to claim 3, characterized in that, The dynamic video stream is input into a pre-trained visual recognition network to extract action feature values ​​from the dynamic video stream, including: The dynamic video stream is divided into multiple single-frame images in chronological order; Each of the single-frame images is input into the two-dimensional convolutional layer of the visual recognition network to extract basic feature maps; The basic feature map is input into the time-shifted feature layer, and features are extracted from the basic feature map along the time dimension to obtain a time-series fusion map; The temporal fusion graph is input into the cross-channel attention layer, and the weight values ​​of each feature channel are calculated. The weight values ​​are then used to perform matrix multiplication on the temporal fusion graph to obtain action feature values.

5. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 1, characterized in that, Establishing a communication connection between the intelligent receiver and the panoramic monitoring terminal includes: Obtain multiple alternative protocols supported by the intelligent receiver and the panoramic monitoring terminal; Monitor the frequency band occupancy and network latency of the site environment; Based on the frequency band occupancy rate and the network latency, a target protocol is selected from multiple candidate protocols using an intelligent scheduling algorithm; The intelligent receiver and the panoramic monitoring terminal are configured to perform handshake matching using the target protocol to establish a multi-protocol fusion communication connection.

6. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 5, characterized in that, The configuration of the intelligent receiver and the panoramic monitoring end to perform handshake matching using the target protocol and establish a multi-protocol fusion communication connection includes: Generate public and private key data based on elliptic curve algorithm; exchange the public key data between the smart receiver and the panoramic monitoring terminal; use the private key data to perform hash operation on the real-time location data and the real-time heart rate data to generate a verification feature value; The verification feature value is appended to the transmission data packet and transmitted through the target protocol to construct the communication connection with cross-protocol security protection.

7. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 1, characterized in that, Before mapping the real-time location data to the pre-built 3D simulation model, a model synchronization step is also included: Real-time acquisition of equipment operating values ​​from on-site physical equipment; Convert the device operating values ​​into a digital drive stream; The digital drive stream is input into the initial base model, and the size and state parameters of the corresponding virtual device in the initial base model are adjusted according to the digital drive stream to generate the three-dimensional simulation model with synchronized virtual and real states.

8. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 1, characterized in that, Based on the aforementioned spatial risk status and physiological risk status, a coordinated early warning instruction is generated, including: Calculate the difference between the distance threshold and the absolute spatial distance to obtain the spatial urgency; extract the heart rate deviation; obtain the pre-set spatial weight value and physiological weight value; Multiply the spatial urgency by the spatial weight value to obtain a first product value, and multiply the heart rate deviation by the physiological weight value to obtain a second product value; The first product value and the second product value are added together to obtain the comprehensive risk value; when the comprehensive risk value is greater than the preset alarm threshold, a linkage warning instruction containing a danger identification code is generated.

9. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 1, characterized in that, The dynamic baseline heart rate is obtained through the following methods: The spatial amplitude parameter and temporal change frequency within a preset time window are input into the heart rate prediction model, and the output of the heart rate prediction model is used as the dynamic baseline heart rate. The heart rate prediction model is trained based on the historical data of the workers.

10. The method for risk assessment of workers based on the fusion of BeiDou positioning and physiological characteristics according to claim 9, characterized in that, The heart rate prediction model is trained based on the historical data of the workers, including: the heart rate prediction model is a BP neural network; the historical data includes spatial amplitude parameters and temporal change frequency within any preset time window, as well as the real heart rate collected in real time; outlier detection is performed on the historical data, and after deleting the outlier data, the heart rate prediction model is trained using the gradient descent method.

Images