A multi-source heterogeneous data fusion and decision method and system for a restricted space

Abstract

This invention relates to the field of security monitoring technology, and discloses a method and system for multi-source heterogeneous data fusion and decision-making in confined spaces. The method includes: generating task documents via a touchscreen, verifying identity and controlling personnel access using facial recognition; separating audio and video media data streams from low-bandwidth sensor control data using a switch, preventing the audio and video media data streams from entering the PLC module's processing flow to reduce the main control load; dynamically reconstructing the concentration upper limit threshold based on the task documents, combining gas concentration data and heart rate data, and comprehensively performing limit static judgment and first-order trend cross-coupling early warning judgment; triggering a multi-level scheduling process including local prompts, cloud reporting, and offline SMS forced alarms when an anomaly is detected. This invention improves the timeliness of early warnings through multi-dimensional data trend cross-analysis and ensures the transmission of emergency alarm information when the communication network is interrupted, thereby improving the reliability of confined space security monitoring.

Description

Technical Field

[0001] This invention relates to the field of security monitoring technology, specifically to a method and system for multi-source heterogeneous data fusion and decision-making in confined spaces. Background Technology

[0002] Confined space operations are complex environments, and real-time monitoring of internal gas concentrations, personnel status, and on-site visual data is fundamental to ensuring equipment operation and personnel safety. Existing confined space safety monitoring systems have technical deficiencies in data transmission architecture, early warning judgment logic, and communication reliability. In their data processing architecture, traditional monitoring systems typically transmit a mixture of high-bandwidth audio and video media data collected on-site and low-bandwidth underlying sensor and control data to the core control unit for centralized processing. The encoding and decoding tasks of audio and video data consume a significant amount of the control unit's hardware computing resources, leading to an overload and causing delays in the response of underlying control commands and critical sensor data.

[0003] At the level of early warning judgment logic, existing systems generally use preset fixed static thresholds for single-dimensional safety judgment, without dynamically adjusting the judgment criteria according to the specific construction task type. Furthermore, existing technologies treat environmental gas monitoring and personnel physiological characteristic monitoring as completely independent judgment branches, lacking cross-analysis of gas concentration change trends and personnel physiological indicator change trends. This results in the system only triggering alarms when the absolute values ​​of various environmental parameters exceed limits, leading to a lag in the early warning mechanism. In addition, the data transmission of existing monitoring systems heavily relies on conventional broadband networks such as Ethernet. When an accident occurs within the confined space causing a physical interruption of the main communication line or network paralysis, the system directly loses its ability to send data outwards. The lack of underlying network degradation and fault tolerance design and emergency communication backup links hinders the effective transmission of on-site alarm information. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for multi-source heterogeneous data fusion and decision-making in confined spaces. It solves the problems of existing systems in confined spaces that mix and process high-bandwidth video streams and low-bandwidth sensor data, leading to excessive processing load on the main control unit and potential control command delays; existing early warning mechanisms often use fixed thresholds without comprehensively analyzing the rate of change in the on-site environment and the physiological characteristics of workers, resulting in insufficient accuracy and timeliness of early warnings; furthermore, when the internal communication network in the confined space is interrupted, the system lacks an underlying network degradation and fault tolerance mechanism and an emergency communication link, failing to guarantee the effective transmission of alarm information.

[0005] To address the above problems, the present invention provides the following technical solution: The first aspect of this invention provides a method for multi-source heterogeneous data fusion and decision-making in confined spaces. The method is executed collaboratively by multiple terminals, including a touchscreen, a face recognition camera, a switch, and a PLC module connected via a communication network, and includes the following steps: The touchscreen generates electronic construction task documents based on the input work restriction parameters; The facial recognition camera acquires the facial feature data of the object to be verified, and performs identity verification in conjunction with the electronic construction task document. The PLC module controls the generation of an entry release command. The PLC module dynamically reconstructs the dynamic concentration upper limit threshold for the current task cycle based on the business code in the electronic construction task document. The switch acquires audio and video media data streams, gas concentration data, and individual heart rate and blood pressure data of workers within the confined space, and performs heterogeneous data splitting to separate the audio and video media data streams from low-bandwidth sensor control data, so that the audio and video media data streams do not enter the main control operation process of the PLC module. The PLC module, based on the gas concentration data and heart rate data in the low-bandwidth sensor control data, and combined with the dynamic concentration upper limit threshold, comprehensively performs limit static judgment and first-order trend cross-coupled early warning judgment. When the judgment result meets the abnormal conditions, the PLC module generates a global alarm judgment status quantity and triggers a multi-level alarm scheduling process.

[0006] The innovative principle of the above-mentioned technical solution of the present invention lies in the following: the system performs heterogeneous data diversion through a switch, allowing high-bandwidth audio and video media data to bypass the PLC module's calculation process and be directly directed to the display or transmission terminal, reducing the computational load of the main control unit and ensuring the real-time processing capability of low-bandwidth sensor control data; the system dynamically reconstructs the upper limit threshold of concentration in conjunction with business code, making the safety judgment standard adaptable to the current construction task type; in the early warning logic, the system performs cross-coupling judgment by combining the rate of change of ambient gas concentration and the rate of change of personnel heart rate, and identifies anomalies and outputs early warning signals by changing trends before the absolute value reaches the limit static threshold, thereby improving the timeliness of early warning in complex environments of confined spaces.

[0007] Furthermore, the electronic construction task document includes a permit time parameter and the business code; the facial recognition camera acquires the facial feature data, performs identity verification in conjunction with the electronic construction task document, and the PLC module controls the generation of the entry release instruction, specifically including: The face recognition camera extracts the facial feature vector of the current object to be verified from the acquired facial feature data; it obtains the temporary authorized feature comparison whitelist library associated with the electronic construction task document and extracts the authorized feature vector in the temporary authorized feature comparison whitelist library; it calculates the cosine similarity between the facial feature vector of the current object to be verified and the authorized feature vector; when the cosine similarity is not less than the preset facial feature matching similarity judgment threshold, it determines that the facial feature matching is successful and generates a face matching status quantity with a value of 1; the PLC module obtains the access judgment time and judges... The PLC module determines whether the admission judgment time is between the start and end timestamps of the permission operation converted by the permission time parameter. When the face matching status value is 1 and the admission judgment time is between the start and end timestamps of the permission operation, the PLC module generates the entry release command and sets the admission status value to 1. When the face matching status value is not 1, or the admission judgment time is not between the start and end timestamps of the permission operation, the PLC module sets the admission status value to 0.

[0008] Furthermore, the PLC module dynamically reconstructs the dynamic concentration upper limit threshold for the current task cycle based on the business code in the electronic construction task document, specifically including: Extract the business code from the electronic construction task document, call the threshold adjustment coefficient function corresponding to the business code mapping inside the PLC module, and obtain the output adjustment coefficient; multiply the system preset standard allowable concentration limit with the adjustment coefficient to calculate the dynamic concentration upper limit threshold of the corresponding target gas that is actually effective in the current task cycle.

[0009] Furthermore, the execution of heterogeneous data splitting, separating the audio and video media data stream from the low-bandwidth sensor control data, specifically includes: The switch reads the MAC address and port mapping relationship of the Ethernet data packets; it directly forwards the audio and video media data stream with the touch screen or transparent transmission module as the target node to the target receiving port; it sends the gas concentration data, heart rate data, and blood pressure data as the low-bandwidth sensing control data to the PLC module for unified reception and register address allocation, forming a logical split from the audio and video media data stream on the processing link.

[0010] Furthermore, after receiving the low-bandwidth sensing control data, the PLC module also performs heterogeneous timestamp alignment processing, specifically including: Since the gas concentration data, heart rate data, and blood pressure data arrive at different frequencies, the PLC module constructs a circular data buffer based on the current sampling time in the register and caches the gas concentration data, heart rate data, and blood pressure data into the circular data buffer. The PLC module uses zero-order hold interpolation to timestamp-align the data in the circular data buffer and reconstructs a comprehensive feature matrix for subsequent execution of the first-order trend cross-coupling early warning judgment.

[0011] Furthermore, in the integrated execution of the static limit judgment and the first-order trend cross-coupling early warning judgment, the static limit judgment introduces a time-domain filtering anti-jitter mechanism, and the judgment logic includes: A preset confirmation time window is set; when the gas concentration data sampling value of any target gas is continuously higher than the corresponding dynamic concentration upper limit threshold within the preset confirmation time window, the PLC module sets an environmental abnormality status quantity.

[0012] Furthermore, the specific logic for the first-order trend cross-coupling early warning judgment includes: The system determines whether the heart rate or blood pressure data of any individual worker exceeds a preset physiological safety threshold. If it does, a physiological abnormality state quantity with a value of 1 is generated. The system extracts the gas concentration data and heart rate data within a preset time window, and calculates the concentration change rate of the target gas and the heart rate change rate of the individual worker within the most recent time window. When the concentration change rate exceeds a preset environmental trend gradient threshold, and the heart rate change rate exceeds a preset physiological trend gradient threshold, the system determines that the current environment has had an initial impact on the specific individual, and outputs a coupled early warning decision variable with a value of 1. The trigger condition for the global alarm decision state quantity is that any individual worker's physiological abnormality state quantity is 1, the environmental abnormality state quantity is 1, or the coupled early warning decision variable is 1.

[0013] Furthermore, the multi-level alarm scheduling process includes performing network degradation fault tolerance judgment and offline SMS forced alarm, specifically including: The background program of the PLC module continuously monitors the heartbeat status of the Ethernet port. When the PLC module fails to receive an Ethernet heartbeat for a preset number of consecutive periods or the received data suddenly becomes invalid or lost, and there are abnormal environmental conditions, it determines that an accident has occurred inside the confined space, causing network paralysis or personnel to be trapped, and triggers the network degradation fault tolerance mechanism. The PLC module stops relying on the Ethernet link and sends a set of offline alarm instructions to the SMS module with a serial communication physical connection, driving the SMS module to use the mobile cellular network to send an alarm SMS containing a disconnection diagnosis code and the last known geographical location coordinates to a preset terminal.

[0014] Furthermore, the multi-level alarm scheduling process also includes performing local fusion display processing, specifically including: The graphics rendering engine inside the touchscreen performs hardware decoding operations on the audio and video media data stream to restore and output dynamic video frames; based on OSD screen character overlay technology, a transparent data overlay layer is created on top of the underlying video frames; the touchscreen reads the real-time values ​​of the gas concentration data, heart rate data, and blood pressure data uploaded by the PLC module, converts them into text or graphic controls, and draws a preset display area overlaid on the transparent data overlay layer.

[0015] A second aspect of the present invention provides a multi-source heterogeneous data fusion and decision-making system for confined spaces, applied to the multi-source heterogeneous data fusion and decision-making method described in the first aspect above, the system comprising: The touch screen box, located outside the confined space, contains a touch screen, a first switch, a second switch, a transparent transmission module, a face recognition camera, a first intercom module, a PLC module, an SMS module, and a GPS module. A camera box and a gas alarm are arranged inside a confined space. The camera box contains a third switch, a monitoring camera, a second intercom module, and a wristband radar for receiving wireless signals. And a monitoring wristband that collects heart rate and blood pressure data, and transmits the data wirelessly to the wristband radar; The modules and devices within the system are connected via Ethernet, industrial bus interface, or serial interface.

[0016] This invention provides a method and system for multi-source heterogeneous data fusion and decision-making in confined spaces. It offers the following advantages: 1. This invention performs heterogeneous data splitting through a switch, physically and logically separating high-bandwidth audio and video media data streams from low-bandwidth sensor control data. This structure prevents audio and video media data streams from entering the main control processing flow of the PLC module, directly forwarding them to the touch screen or transparent transmission module. This avoids video encoding and decoding tasks occupying core control resources, reduces the computational load of the PLC module, and ensures the system's real-time processing capability and response speed for underlying key sensor data such as gas concentration and physiological characteristics.

[0017] 2. This invention employs a first-order trend cross-coupling early warning judgment mechanism, combined with dynamic reconstruction of the concentration upper limit threshold using business code. Based on conventional static limit judgment, the PLC module simultaneously calculates the rate of change of ambient gas concentration and the rate of change of the worker's heart rate. When the rates of change of both environmental and physiological indicators simultaneously exceed preset gradients, the system outputs a coupled early warning signal. This method changes the traditional judgment mode that relies on a single fixed value. Through trend cross-analysis of multi-dimensional data, it achieves early identification before the absolute value of environmental parameters reaches its limit, improving the accuracy and timeliness of safety warnings.

[0018] 3. This invention establishes a network degradation fault tolerance judgment and offline SMS forced alarm mechanism. During operation, the PLC module continuously monitors the Ethernet port heartbeat packet status. When a partial network interruption is detected and environmental status variables are abnormal, the system automatically triggers the network degradation fault tolerance mechanism. The PLC module cuts off its dependence on the Ethernet link and drives the SMS module through the underlying serial interface to send a structured alarm SMS containing diagnostic codes and coordinates via the mobile cellular network. This mechanism ensures that even in the event of an accident within a confined space that paralyzes conventional broadband communication, emergency alarm information can still be effectively transmitted externally, enhancing the system's operational reliability. Attached Figure Description

[0019] Figure 1 The system structure schematic diagram provided for the embodiments of the present invention; Figure 2 A flowchart illustrating the online construction monitoring method for confined spaces provided in this embodiment of the invention; Figure 3 A flowchart illustrating the steps for creating and issuing electronic construction orders as provided in this embodiment of the invention; Figure 4 This is a flowchart illustrating the access verification and log linkage process provided in an embodiment of the present invention. Figure 5 This is a flowchart of heterogeneous data offloading, acquisition, and local aggregation provided in an embodiment of the present invention; Figure 6 This is a flowchart of multi-parameter joint early warning and parallel alarm distribution provided in an embodiment of the present invention; Figure 7 The following is a time series diagram of the cross-coupling early warning time series of target gas concentration and personnel heart rate change rate provided in the embodiments of the present invention, wherein (a) is a curve of the first derivative change rate of target gas concentration; and (b) is a curve of the first derivative change rate of personnel heart rate. Figure 8 A bar chart comparing the impact of heterogeneous data offloading on the main control CPU load in experimental examples of this invention; Figure 9 A scatter plot comparing the trigger times of the cross-coupling early warning algorithm and the traditional early warning algorithm provided as experimental examples of this invention.

[0020] Among them, 100 is the touch screen box; 101 is the touch screen; 102 is the first switch; 103 is the second switch; 104 is the pass-through module; 105 is the face recognition camera; 106 is the first intercom module; 107 is the PLC module; 108 is the SMS module; 109 is the GPS module; 200 is the camera box; 201 is the third switch; 202 is the monitoring camera; 203 is the wristband radar; 204 is the second intercom module; 300 is the gas alarm; and 400 is the monitoring wristband. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See attached document Figure 1 This invention provides a multi-source heterogeneous data fusion and decision-making system for confined spaces. Unlike existing technologies where monitoring modules operate in isolation, this system constructs an active defense architecture based on task-driven dynamic reconstruction of the safety baseline and an environment-physiology cross-coupling early warning mechanism. The confined spaces referred to in this invention include restricted spaces, semi-restricted spaces, poorly ventilated spaces, and construction spaces with restricted access and safety risks such as toxicity, oxygen deficiency, and flammability / explosiveness.

[0023] The system includes a touch screen box 100 located outside the confined space, a camera box 200 and a gas alarm 300 located inside the confined space, and a monitoring wristband 400 worn by construction personnel entering the confined space.

[0024] The touch screen box 100 is equipped with a touch screen 101, a first switch 102, a second switch 103, a pass-through module 104, a face recognition camera 105, a first intercom module 106, a PLC module 107, an SMS module 108, and a GPS module 109. The second switch 103 is connected to the first switch 102, the first intercom module 106, and the PLC module 107 via Ethernet. The pass-through module 104, the touch screen 101, and the face recognition camera 105 are connected to the first switch 102. The SMS module 108 and the GPS module 109 are connected to the PLC module 107. The SMS module 108 establishes a communication connection with the PLC module 107 through an RS-232 or RS-485 serial interface.

[0025] Among them, the SMS module 108 sends an alarm SMS to a preset terminal based on the alarm signal output by the PLC module 107; the GPS module 109 provides the PLC module 107 with the location information of the construction point where the touch screen box 100 is located, so that the alarm SMS or uploaded data carries the on-site location information.

[0026] The camera box 200 is arranged inside a confined space. Inside it are a third switch 201, a monitoring camera 202, a wristband radar 203, and a second intercom module 204. The monitoring camera 202, wristband radar 203, and second intercom module 204 are respectively connected to the third switch 201. The third switch 201 is connected to the second switch 103 inside the touch screen box 100 via an Ethernet cable. The first intercom module 106 and the second intercom module 204 form a two-way voice link through the second switch 103 and the third switch 201 for communication between people inside and outside the confined space.

[0027] The gas alarm 300 is placed in a confined space and connected to the serial communication interface of the PLC module 107 via an RS-485 industrial bus to upload gas concentration data to the PLC module 107.

[0028] The monitoring wristband 400 is worn on the wrist of construction workers entering confined spaces to collect their heart rate and blood pressure data. The collected heart rate and blood pressure data are then wirelessly transmitted to the wristband radar 203. After receiving the data, the wristband radar 203 forwards it to the PLC module 107 via the third switch 201 and the second switch 103.

[0029] This system adopts a heterogeneous data splitting transmission structure. For the video images captured by the monitoring camera 202 and the voice data generated by the second intercom module 204, the system frames them into a media stream and transmits them to the transparent transmission module 104 and the touch screen 101 via the third switch 201, the second switch 103 and the first switch 102 respectively. The transparent transmission module 104 pushes the media stream to the cloud platform and the mobile terminal, and the touch screen 101 performs local decoding and display of the media stream.

[0030] The system inputs the identity verification status output by the face recognition camera 105, the gas concentration data output by the gas alarm 300, and the heart rate and blood pressure data forwarded by the wristband radar 203 as monitoring status data into the PLC module 107 for processing. Specifically, the PLC module 107 receives the identity verification status through the first switch 102 and the second switch 103, receives the gas concentration data through the serial communication interface, and receives the heart rate and blood pressure data forwarded by the wristband radar 203 through the second switch 103.

[0031] PLC module 107 compares the gas concentration data, heart rate data, and blood pressure data with their respective preset thresholds. The preset thresholds for gas concentration are selected based on the permissible concentration limits of the corresponding gases specified in the national occupational health standards. The preset thresholds for heart rate data and blood pressure data are selected based on the resting physical signs standards of healthy adults, and are individually calibrated in conjunction with the pre-employment physical examination data of workers.

[0032] The setting of the aforementioned preset threshold directly affects the accuracy of the system's early warning: if the threshold is set too high, it will lead to the failure to report real dangers, threatening the lives of construction workers; if the threshold is set too low, it will frequently trigger false alarms due to the fact that it does not conform to the industrial background environment or the fluctuations in vital signs caused by workers' normal heavy physical labor in confined spaces.

[0033] When any data reaches or exceeds the corresponding threshold, the PLC module 107 generates an alarm signal and links the SMS module 108 to send an alarm SMS. The identity verification status output by the face recognition camera 105 is used to form a record of the construction personnel's entry verification. If the verification fails, the PLC module 107 generates an identity abnormality signal. The video image data is transmitted and displayed along the media stream channel. The PLC module 107 does not participate in the encoding and decoding calculation of the video image data.

[0034] See attached document Figure 2 A method for multi-source heterogeneous data fusion and decision-making in constrained spaces, which can be executed by the aforementioned system, includes the following steps: S10, Create an electronic construction task document carrying operation attribute factors.

[0035] Operators input construction worker identification information, work content, and permitted time parameters through touchscreen 101, generating and storing electronic construction task documents. These electronic task documents are stored on touchscreen 101 and simultaneously uploaded to the cloud platform via transparent transmission module 104.

[0036] The construction worker identity information includes a construction worker identification identifier and a facial feature ID associated with that identifier. The facial feature ID is used to call the corresponding pre-stored facial feature template.

[0037] S20, perform on-site verification and dynamic reconstruction of environmental safety baseline.

[0038] Before construction workers enter the confined space, a facial recognition camera 105 captures their facial images and extracts facial feature data. The facial recognition camera 105 compares the extracted facial feature data with the pre-stored facial feature template associated with the construction worker's identity in the electronic construction task document, and sends the identity verification status to the PLC module 107.

[0039] PLC module 107 determines whether entry is permitted based on the identity verification status and the permission time parameter. When the facial features match and the current time is within the permission time parameter, PLC module 107 generates an entry release instruction and verification log record; the verification log record is stored in association with the electronic construction task document and uploaded to the cloud platform through the transparent transmission module 104.

[0040] When the facial features do not match, or the current time is not within the permitted time parameters, the PLC module 107 generates an entry rejection instruction and a verification anomaly record, and uploads the verification anomaly record to the cloud platform through the transparent transmission module 104.

[0041] S30, collects on-site monitoring data.

[0042] The gas alarm 300 detects the concentration of harmful gases in the confined space in real time. The monitoring wristband 400 collects the heart rate and blood pressure data of the workers and transmits the heart rate and blood pressure data wirelessly to the wristband radar 203; the wristband radar 203 then transmits the heart rate and blood pressure data to the PLC module 107 via the third switch 201 and the second switch 103.

[0043] GPS module 109 acquires the geographical coordinates of the construction site where the touch screen box 100 is located. PLC module 107 summarizes the gas concentration data, heart rate data, blood pressure data, and geographical coordinates, and sends the summarized monitoring data to the touch screen 101 for display.

[0044] S40 performs safety checks and alarm linkage.

[0045] PLC module 107 compares the gas concentration data, heart rate data, and blood pressure data with preset gas concentration safety thresholds, heart rate safety ranges, and blood pressure safety ranges, respectively.

[0046] When the gas concentration data exceeds the gas concentration safety threshold, or when the heart rate data or blood pressure data deviates from the corresponding safety range, the PLC module 107 outputs an alarm status signal to the touch screen 101, which then displays an alarm screen and issues an audible prompt.

[0047] Simultaneously, PLC module 107 generates an alarm event packet, which is then sent to transparent transmission module 104 via second switch 103 and first switch 102. Transparent transmission module 104 then uploads the packet to the cloud platform. The alarm event packet includes abnormal parameters, construction personnel identification information, electronic construction task document number, alarm time, and geographic location coordinates.

[0048] PLC module 107 also drives SMS module 108 to generate alarm SMS messages based on alarm event packets and send the alarm SMS messages to the management communication terminal.

[0049] See attached document Figure 3 In this embodiment, the operator enters the operation restriction parameters through the touch screen 101, which generates an electronic construction task document and sends the data related to personnel verification and entry judgment to the face recognition camera 105 and the PLC module 107 respectively.

[0050] S101, the operator triggers a task creation command on the display interface of the touch screen 101, the touch screen 101 runs the task management program, and provides corresponding input controls on the display interface to obtain the operation restriction parameters input by the operator.

[0051] The specific work restriction parameters include a list of personnel authorized to enter the confined space, the permitted start time of the work, the permitted end time of the work, and the business code for the current construction operation. Among them, business code This is used to characterize the environmental risk feature coefficients corresponding to different types of operations (such as hot work, anti-corrosion coating, and routine inspection) within confined spaces.

[0052] The operator identification information table includes the operator's name, identification number, and facial feature ID associated with the identification number.

[0053] The start and end times of the permitted work are manually set by the operator based on the actual planned cycle of the on-site construction work.

[0054] S102, the task management program running on the touch screen 101 performs structured processing on the job restriction parameters, generates electronic construction task documents in a unified format, and assigns a corresponding document number to the electronic construction task document.

[0055] This electronic construction task document is used to define the authorized personnel, permitted work time window, and task content for this construction operation. It is converted into construction task data tuples at the system software layer.

[0056] Let the first The generated electronic construction task document is defined as follows: Its construction logic satisfies the following mathematical expression: ; In the formula, This is a set of facial feature IDs of authorized personnel extracted after querying personnel files based on their identification numbers in the personnel identification information table; This indicates that the touchscreen 101 converts the permitted operation start time into a timestamp based on the permitted operation start time set by the operator. This indicates that the touchscreen 101 converts the permitted operation termination timestamp according to the permitted operation termination time set by the operator. This indicates the business code for the current construction operation. The business code corresponds to the specific construction requirements and safety level.

[0057] The document number serves as the index field for the electronic construction task document, and is linked to the construction task data tuple. Associated storage.

[0058] Touchscreen 101 serializes and encapsulates construction task data tuples into a JSON data exchange format file, and uses this JSON data exchange format file as the document configuration file for distribution to relevant devices. The serialization, parsing, and encapsulation of JSON format data can be implemented through conventional software application programming interfaces, which will not be elaborated here.

[0059] S103, the touch screen 101 encapsulates the document configuration file into data packets based on the Ethernet communication standard and TCP / IP transmission protocol, and sends them to the first switch 102 via a physical network cable.

[0060] The first switch 102 reads the MAC address of the data packet, forwards the data packet used for face whitelist configuration to the receiving port of the face recognition camera 105, and sends the data packet containing the document number and the authorized operation start timestamp. Permitted operation termination timestamp and construction operation business codes The data packets are sent to the PLC module 107 via the first switch 102 and the second switch 103.

[0061] S104, The face recognition camera 105 receives and unpacks the document configuration file, and extracts the set of authorized personnel's facial feature IDs. It calls the locally stored human facial model library for matching and verification.

[0062] When the local facial recognition model library needs to be updated, the face recognition camera 105 accesses the facial recognition model library in the cloud platform via the first switch 102 and the pass-through module 104 to obtain the updated facial recognition model. The facial recognition model is generated by the system after capturing facial images and extracting features during the personnel registration stage.

[0063] If the face recognition camera 105 discovers an unregistered facial feature ID in the Ureq during the matching and verification process, it will return parameter setting error information to the touch screen 101 and interrupt the current document issuance process.

[0064] If the verification is successful, a temporary authorized feature comparison whitelist library for this task will be generated in the local random access memory of the face recognition camera 105.

[0065] The facial recognition camera 105 does not generate a final entry release command. Instead, it performs subsequent facial comparison based on a temporary authorized feature comparison whitelist and sends the identity verification status to the PLC module 107. The PLC module 107 receives and stores the document number and the authorized operation start timestamp. Permitted operation termination timestamp and construction operation business codes and according to The corresponding environmental risk weight matrix is ​​extracted from the internal memory to provide benchmark parameters for the subsequent dynamic threshold reconstruction of the underlying sensors.

[0066] After the configuration is completed, the face recognition camera 105 and the PLC module 107 respectively return confirmation receipt messages to the touch screen 101. Upon receiving the confirmation receipt message, the touch screen 101 determines that the electronic construction task document has been issued; if the touch screen 101 does not receive the confirmation receipt message within a preset time, it determines that the issuance has failed and prompts for re-issuance on the interactive interface.

[0067] After the electronic construction task order is issued, the face recognition camera 105 enters the face comparison standby state, and the PLC module 107 enters the entry permission judgment standby state.

[0068] See attached document Figure 4 This embodiment further explains the process of facial feature comparison, access judgment, and verification log upload for construction workers before they enter a confined space.

[0069] S201, Collect and extract the facial features of the current person.

[0070] When construction workers prepare to enter the site, the optical sensor of the face recognition camera 105 captures video frame images containing facial information. The data processing chip built into the face recognition camera 105 performs grayscale conversion, face alignment, target facial region cropping, and scale normalization on the video frame images, generating an input image matrix with a size of 112×112 pixels.

[0071] In one embodiment of the present invention, the face recognition camera 105 internally loads a deep convolutional neural network model, preferably using the ResNet-50 network architecture. Specifically, the structure of the ResNet-50 model includes: a 7×7 initial convolutional layer for shallow feature extraction, a 3×3 max pooling layer for dimensionality reduction; followed by a cascaded four-stage residual convolutional module, each residual block employing a 1×1, 3×3, 1×1 bottleneck convolutional structure to reduce computational parameters, and using skip connections to directly add the input and output to solve the gradient vanishing problem during backpropagation in deep networks; finally, a global average pooling layer flattens the features, and a fully connected layer outputs a 512-dimensional facial feature vector of the current object to be verified. The facial feature vector of the object to be verified Used to characterize the feature distribution of the current subject's facial image in a high-dimensional feature space.

[0072] To improve the stability of on-site recognition, the deep convolutional neural network model can be pre-trained on a publicly available face dataset, and then fine-tuned using the on-site personnel face image dataset generated during the system's initial archiving. The sample labels in the on-site personnel face image dataset are the personnel's identity category ID codes, and the training process uses the ArcFace loss function based on angle margin penalty to guide the iterative update of network parameters.

[0073] Due to the complex lighting conditions in confined spaces and the fact that workers often wear safety helmets, traditional loss functions often fail to adequately distinguish between class features. The ArcFace loss function maximizes the inter-class classification boundary by introducing an additional angle penalty term into the angle space between the feature vector and the weight vector. Specifically, during network training, the system calculates the angle between the facial feature vector and the true class weight vector, directly adds a preset angle margin penalty term to this angle, and then calculates the loss.

[0074] The angular margin penalty term is obtained by performing grid search cross-validation on a face dataset related to the confined space to determine the optimal value; in this embodiment, 0.50 radians is preferred. The mechanism of this setting is to force the neural network to compress facial features belonging to the same person more tightly in the feature space and push features of different people further apart. This ensures that even if a worker's facial features shift due to changes in lighting within a confined space, their features will still stably fall within the boundaries of the corresponding high-dimensional feature clusters, significantly improving the robustness of recognition in harsh environments.

[0075] The above model training and parameter optimization can be achieved based on conventional deep learning frameworks.

[0076] S202, perform facial feature comparison and generate admission status.

[0077] The facial recognition camera 105 extracts the facial feature vector of the object to be verified. Then, the facial feature vector of the current object to be verified is... Normalization is performed, and the facial feature vector of the current object to be verified is calculated sequentially. Compare the temporary authorization features stored in local random access memory with the first one in the whitelist. One authorized feature vector Cosine similarity between The calculation formula satisfies: ; In the formula, This represents the facial feature vector of the object to be verified. With the One authorized feature vector Cosine similarity between them , This represents the number of feature vectors in the temporary authorized feature comparison whitelist; the numerator represents the dot product of the two feature vectors, and the denominator represents the L2 norm product of the two feature vectors.

[0078] To simultaneously restrict personnel identity and working hours, this embodiment adopts an access judgment method that combines face matching status with permission time window.

[0079] Among them, the face recognition camera 105 generates a face matching state quantity based on the cosine similarity calculation result. ,exist When the value is 1, the matched identity category ID code is determined, and the face matching state is recorded. The matched identity category ID code and comparison time are sent to PLC module 107; PLC module 107 obtains the admission judgment time according to its own system clock. And combined with the start timestamp of the permitted operation and the end timestamp of the licensed operation Generate admission state variables .

[0080] Face matching state quantity The determination and calculation rules are as follows: ; The rules for determining the admission state variable Vaccess are as follows: ; In the formula, This represents the system's preset facial feature matching similarity threshold, which is preferably set between 0.75 and 0.85. The selection of this threshold is based on the special lighting conditions in the confined space and the facial distortion caused by the camera's installation angle, combined with the system's previous cross-test data on false recognition rate (FAR) and false rejection rate (FRR) in real confined space scenarios, to select the optimal value near the equal error rate (EER) point that balances safety and passage efficiency.

[0081] The impact of this setting on the results is as follows: if the threshold is set too high (e.g., greater than 0.9), the false rejection rate of the system will increase sharply, and legitimate construction personnel will be frequently denied entry due to slight deviations in features caused by the shadow of wearing a safety helmet or sweat on their face, affecting construction efficiency; if the threshold is set too low (e.g., less than 0.6), the false recognition rate will increase significantly, and there is a risk that unauthorized personnel will be mistakenly allowed to enter the system due to similar facial contours, undermining the access security of the confined space.

[0082] When the temporary authorization feature comparison whitelist contains at least one authorization feature vector This makes it match the facial feature vector of the current object to be verified. Cosine similarity between Not less than the judgment threshold At that time, the face matching state quantity Output a value of 1; otherwise, the face matching state variable... The output value is 0. PLC module 107 is in... =1 and admission judgment time When within the permitted operation time window, the access status quantity will be... Set to 1; =0 or admission judgment time When the permitted operation time window is exceeded, the access status will be changed. Set it to 0 and output a rejection signal to the touch screen 101.

[0083] To prevent the same person or unauthorized personnel from repeatedly triggering the identification process, the system sets a threshold for the number of retry attempts. When the number of consecutive outputs of the value 0 exceeds the retry threshold, the face recognition camera 105 suspends the current recognition process and sends the continuous comparison failure status to the PLC module 107; the PLC module 107 generates an abnormal access event and uploads it to the cloud platform via the second switch 103, the first switch 102 and the transparent transmission module 104.

[0084] S203, generate release results and perform environmental safety baseline reconstruction.

[0085] When PLC module 107 determines the access status quantity When the value is 1, the PLC module 107 generates an entry release command and sends the entry release command to the touch screen 101, which then displays an entry permission prompt.

[0086] Meanwhile, PLC module 107 determines the access time based on the successfully matched identity category ID code and the access judgment time. Admission status quantity The release event code and electronic construction task document number are used to generate a structured verification log data package.

[0087] In addition, the main control program inside PLC module 107 is triggered synchronously, based on the business code in the current electronic construction task document. The system dynamically reconstructs and distributes the safety threshold of the underlying gas alarm 300. Assume the system targets the first... The preset permissible concentration limit for the target gas is The PLC module 107 internally maps a set of threshold adjustment coefficient functions. .

[0088] This threshold adjustment coefficient function is essentially a discrete piecewise mapping rule based on the risk level of the construction operation environment. Since different confined space operations have different sensitive media that could induce safety accidents, different tightening coefficients need to be assigned to different gas baselines.

[0089] PLC module 107 according to Recalculate and issue the actual effective dynamic concentration upper limit threshold for the current task cycle. The calculation formula satisfies: ; Specifically, the mapping rules for this function are based on the following criteria and examples: when When hot work (such as electric welding) is identified as having a high risk of explosion, this function outputs an adjustment coefficient of 0.5 for combustible gases (reducing the alarm threshold by 50%), an upper limit output of 0.8 for oxygen concentration (preventing oxygen-enriched combustion), and an output of 1.0 for other gases. when When the coating is applied for corrosion protection, the concentration of toxic solvents is likely to exceed the standard. This function has an output adjustment coefficient of 0.6 for volatile organic compounds and hydrogen sulfide, and 0.8 for combustible gases. when When the environmental risk is stable during routine inspections, the function has a default adjustment coefficient of 1.0 for all target gas outputs.

[0090] Through this dynamic reconfiguration mechanism, the system enables the security detection sensitivity of the underlying hardware modules to adapt to specific construction tasks, overcoming the shortcomings of fixed and rigid alarm thresholds in existing technologies.

[0091] See attached document Figure 5 This embodiment further explains the process of collecting, distributing, transmitting, and displaying gas concentration data, personnel vital signs data, geographical location data, and audio and video media data.

[0092] S301 collects low-bandwidth sensor control data.

[0093] The gas alarm 300 is installed inside a confined space, and its built-in electrochemical sensor array detects the concentration of various target gases in the space environment in real time. The gas alarm 300 converts analog electrical signals into digital signals and transmits the concentration variables of each component to the serial communication port of the PLC module 107 via an RS-485 industrial bus interface.

[0094] At the system logic level, the current sampling time Multichannel gas concentration feature vector Defined as: ; In the formula, Indicates the first The target gas at the current sampling time Real-time concentration sampling value, The total number of types of gases that can be detected by the system can be configured to 5, 6 or 8 types in specific embodiments of the present invention, depending on the hazard level of different space environments.

[0095] The monitoring wristband 400 is worn on the wrist of the worker and uses a photoplethysmography (PPG) sensor to detect changes in blood flow in real time, collecting heart rate and blood pressure data. Heart rate data can be directly calculated by the PPG sensor; blood pressure data can be collected by the built-in pressure sensing unit of the monitoring wristband 400, or calculated by the built-in blood pressure estimation processing unit based on pulse wave conduction time, pulse wave amplitude, heart rate, and preset individual calibration parameters.

[0096] The monitoring wristband 400 uses low-power Bluetooth or LoRa short-range wireless communication technology to send raw physiological status messages to the wristband radar 203 deployed inside the space. The communication processor inside the wristband radar 203 unpacks and converts the received wireless messages, generates Ethernet TCP / IP data packets, and sends them to the PLC module 107 via the third switch 201 and the second switch 103.

[0097] Corresponding individual workers At the current sampling time Physiological feature vector Represented as: ; In the formula, For individual workers At the current sampling time Heart rate value; For individual workers At the current sampling time The systolic blood pressure value; For individual workers At the current sampling time The diastolic blood pressure value.

[0098] The GPS module 109, located inside the touchscreen box 100, continuously receives satellite positioning signals. In one embodiment, the positioning antenna of the GPS module 109 extends to the outside of the touchscreen box 100 to enhance satellite signal reception. The GPS module 109 analyzes and obtains the geographical latitude and longitude coordinates of the construction site where the touchscreen box 100 is located. : ; In the formula, The current sampling time The acquired latitude data, and The current sampling time The obtained longitude data.

[0099] The aforementioned gas concentration feature vector, physiological feature vector, and geographic coordinate data together constitute a low-bandwidth sensing and control data set. This data is uniformly received and register address allocated by the central processing unit of PLC module 107. Based on the wristband identifier or personnel identification in the physiological status message, PLC module 107 then assigns the physiological feature vector... Individual corresponding workers The association is then established. Specifically, the physiological status message sent by the monitoring wristband 400 carries the identification of the monitoring wristband 400. The PLC module 107 reads the binding relationship between the identity number pre-stored in the electronic construction task document and the identification of the monitoring wristband 400, and determines the identity of the worker corresponding to the physiological feature vector accordingly.

[0100] For the specific conversion code and message verification mechanism of communication protocols for different types of underlying sensors, those skilled in the art can use communication protocol libraries such as Modbus, which are commonly used in the field of industrial automation control, to implement them through programming.

[0101] S302, splits and transmits audio and video media data.

[0102] The surveillance camera 202 continuously captures dynamic images of the confined space and encodes the video stream internally to generate a high-bandwidth media data stream. The second intercom module 204 simultaneously picks up ambient sounds to generate audio stream data.

[0103] This embodiment sets up an independent forwarding path for audio and video media data. At the Ethernet communication protocol level, the data packets generated by the monitoring camera 202 and the second intercom module 204 target the touch screen 101 or the pass-through module 104 as the target node. The aforementioned media data stream is input through the third switch 201 and transmitted to the second switch 103 inside the touch screen box 100 via the Ethernet cable across the enclosure. It is then forwarded by the first switch 102 to the touch screen 101 or the pass-through module 104.

[0104] Since PLC module 107 is not the target node of the media data stream packet, the switch network forwards the media data stream to touch screen 101 or transparent transmission module 104 based on the destination MAC address and port mapping relationship. Therefore, the audio and video media data stream does not enter the video encoding and decoding process and main control operation process of PLC module 107. Instead, it forms a logical offload with low-bandwidth sensor control data such as gas concentration and vital signs on the processing link, thereby reducing the main control processing load of PLC module 107.

[0105] S303 enables local fusion display, heterogeneous timestamp alignment, and remote synchronous publishing.

[0106] After completing the polling extraction and structured analysis of the underlying environmental and physiological parameters, PLC module 107, due to the different data arrival frequencies of the gas alarm 300 and the monitoring wristband 400, constructs a register based on the current sampling time. A circular data buffer is used, and zero-order hold interpolation is employed to align the timestamps of low-frequency arriving data. The specific logic is as follows: Let the current high-speed sampling requirement of the system master be... The most recent actual data arrival time of a certain low-frequency sensor was... ( ≤ If the system directly causes the sensor to... The value at time equals The value at each time point, i.e., before the arrival of the next real data, remains unchanged from the previous sampled value. This system uses zero-order hold-before interpolation instead of conventional linear interpolation because environmental and physiological data are step-type safety monitoring parameters. Linear interpolation would artificially create false slopes between two adjacent low-frequency sampling points, severely interfering with the calculation of the first-order trend cross-derivative in subsequent step S401, leading to false alarms. Zero-order hold-before interpolation reconstructs a strictly aligned comprehensive feature matrix for subsequent cross-joint early warning analysis. Subsequently, the data is sent to the second switch 103 via the internal network interface, and then reported to the touchscreen 101 via the first switch 102. The main processor of the touchscreen 101 simultaneously receives high-bandwidth video media data streams forwarded by the second switch 103 and the first switch 102.

[0107] The graphics rendering engine inside the touchscreen 101 performs hardware decoding operations on the video media data stream, restoring and outputting dynamic video frames. According to one embodiment of the present invention, in the display processing pipeline of the user interface, the application program, based on OSD screen character overlay technology, creates a transparent data overlay layer on top of the underlying video frames.

[0108] The UI rendering component reads the gas concentration feature vector uploaded by PLC module 107. Physiological feature vectors and the geographic latitude and longitude coordinates collected by GPS module 109 The real-time values ​​are obtained and converted into text or graphic controls, which are then drawn on the preset display area of ​​the transparent data overlay.

[0109] The front-end interface of the touch screen 101 simultaneously displays the geographical location of the construction site, environmental status parameters, personnel vital signs, and on-site video footage locally.

[0110] The transparent transmission module 104 receives audio and video media data streams and sensor control data uploaded by the PLC module 107, and simultaneously publishes them to the cloud platform server and remote management terminal using an encrypted communication protocol. The cloud platform server or remote management terminal synchronizes and matches the media data streams and sensor control data according to the data timestamps to achieve remote monitoring screen display and status data traceability.

[0111] See attached document Figure 6 This embodiment further explains the process by which the PLC module 107 makes a joint judgment based on environmental parameters and human physiological parameters, and triggers local prompts, cloud reports, and SMS alarms respectively.

[0112] S401, execute environmental-physiological dynamic cross-coupling early warning judgment.

[0113] First, the PLC module 107 performs a static limit judgment: the PLC module 107's internal memory has a multi-dimensional safety judgment benchmark parameter table preset, which is based on the gas concentration feature vector collected by the gas alarm 300. The ladder diagram control program of PLC module 107 executes the upper limit comparison instruction.

[0114] Let the first The upper limit threshold for the safe concentration of the target gas is To avoid false alarms caused by sensor data glitches due to electromagnetic interference or airflow fluctuations in industrial environments, PLC module 107 incorporates a time-domain filtering anti-jitter mechanism in its alarm logic, with a preset confirmation time window set as follows: .

[0115] This confirmation time window The selection criteria are: the response time characteristics of the electrochemical gas sensor and the brief airflow disturbance caused by personnel moving in the confined space, preferably 3 to 5 seconds.

[0116] The size of the window affects the results as follows: if the window is set too large, the system will be slow to respond to harmful gas leaks, threatening personnel safety; if the window is set too small, sensor circuit noise or local short-term gas concentration fluctuations can easily trigger frequent false alarms.

[0117] System calculation environment abnormal state quantity The mathematical model is defined as follows: ; In the formula, The specific value is set according to the permissible concentration limit of the corresponding gas specified in the national occupational health standard; The value range is typically set to 3 to 5 seconds, and can be configured by engineers based on site ventilation conditions, gas diffusion rate, and sensor response characteristics. When PLC module 107 executes, the concentration sampling value of any target gas falls within the confirmation time window. The concentration remains consistently above the corresponding safe upper limit threshold. At that time, PLC module 107 was set to an abnormal environmental state. .

[0118] Physiological feature vectors collected by the monitoring wristband 400 PLC module 107 executes the range over-limit comparison instruction. Assume individual operators... The corresponding normal heart rate range is The normal range for systolic blood pressure is The normal range for diastolic blood pressure is The aforementioned physiological threshold ranges are set by the management personnel through input on the front-end interface of the touchscreen 101, based on the actual physical examination reports of the workers.

[0119] The selection of the normal physiological range is based on the resting vital signs standards of healthy adults, and is individually calibrated in conjunction with pre-employment medical examination data of workers. The above-mentioned gas safety thresholds and the set size of the normal physiological range directly constitute the baseline for life safety. If the range is too large, it will lead to the underreporting of real dangers; if it is too small, it will lead to frequent false alarms due to incompatibility with the industrial background environment.

[0120] The system calculates the physiological abnormality state quantity. The mathematical model is defined as follows: ; Secondly, PLC module 107 performs a first-order trend cross-coupling early warning judgment: in a confined space, minute changes in gas will be directly reflected in the physiological characteristics of personnel. PLC module 107 extracts the most recent time window. The data within are used to calculate the first... Vector of the rate of change of concentration of the target gas and individual workers heart rate change vector : ; ; When the rate of change of the concentration of a certain gas exceeds the preset environmental trend gradient threshold Furthermore, at the same time, the rate of change in the heart rate of a worker within the sliding window exceeds the physiological trend gradient threshold. At that time, even if the absolute concentration and absolute heart rate Even though the static alarm thresholds were not reached, PLC module 107 still determined that the current environment had an initial impact on a specific individual and output a coupled early warning decision variable. .

[0121] The selection criteria for the environmental and physiological trend gradient thresholds are: the critical value between normal work metabolic fluctuations and mild poisoning / hypoxia stress response, which is obtained by collecting baseline data from normal construction sites.

[0122] The impact of setting the threshold is as follows: if the gradient threshold is set too high, the system will degenerate into a traditional over-limit absolute value alarm, which will be unable to capture the small cumulative effects in the early stages of danger; if the gradient threshold is set too low, the acceleration of the worker's normal heavy physical labor heart rate combined with slight gas fluctuations in the environment will cause a coupled false alarm.

[0123] System global alarm decision status quantity Follow the synthetic logic expression: ; in, Represents any individual worker When the physiological abnormality state quantity is 1, the logical result of this item is 1.

[0124] When the global alarm judgment status quantity When the value jumps from 0 to 1, it indicates that environmental deterioration or abnormal decline in the physical function of personnel has occurred in the current confined space. The alarm interrupt handler of PLC module 107 is activated, triggering the subsequent alarm parallel distribution program.

[0125] S402, execute local alarm notification.

[0126] Global alarm decision state based on activation The PLC module 107 initiates a multi-level alarm scheduling process in parallel. The first-level alarm distribution action is directed to the local alarm notification response.

[0127] PLC module 107 sends a communication message containing an alarm trigger status signal to touch screen 101 via Ethernet interface. After the background process of touch screen 101 listens for the message, it calls the display rendering interface to generate an alarm prompt pop-up window on the top layer of the front-end UI interaction interface, and highlights the abnormal data items that exceed the threshold on the interface.

[0128] At the same time, the PLC module 107 sends an audio prompt control signal to the touch screen 101, which then issues an audio prompt while displaying an alarm prompt pop-up window.

[0129] S403, the second-level alarm distribution action points to the cloud for tracing and reporting response.

[0130] PLC module 107 extracts ambient gas concentration data and abnormal physiological parameters at the time of alarm triggering, and combines them with the geographical coordinate data of the construction point where the touch screen box 100 is located collected by GPS module 109 to generate a structured alarm event package with a fixed frame format.

[0131] The structured alarm event package includes the electronic construction task document number, alarm time, anomaly type, anomaly parameter value, worker identification or wristband identification, and geographic location coordinate data.

[0132] The PLC module 107 sends the structured alarm event packet to the pass-through module 104 via the second switch 103 and the first switch 102. The pass-through module 104 establishes an outward network communication link and uploads the structured alarm event packet to the cloud platform server.

[0133] Regarding the specific communication protocol selection, the IoT communication protocol with message acknowledgment and retransmission mechanisms is preferably used between the transparent transmission module 104 and the cloud platform. For example, the acknowledgment, retransmission, and arrival guarantee of alarm event packets can be achieved through the service quality level configuration of the MQTT protocol.

[0134] S404 executes network degradation fault tolerance judgment and offline SMS mandatory alarm.

[0135] The third-level alarm distribution action refers to the offline SMS forced alarm mechanism, which serves as a safety backup communication link in the event of a broadband network outage. The PLC module 107 establishes a physical connection with the SMS module 108 via an RS-232 or RS-485 serial interface.

[0136] During system operation, the background program of PLC module 107 continuously monitors the heartbeat status of the Ethernet port. When PLC module 107 continuously... No Ethernet heartbeat packet was received in a given cycle, or the received heart rate data... The mutation resulted in invalid packet loss data, and the environment was in an abnormal state at this time. At that time, PLC module 107 determines that an accident has occurred inside the confined space, resulting in network paralysis or personnel being trapped and out of contact.

[0137] At this time, PLC module 107 triggers the network degradation fault tolerance mechanism, stops relying on the Ethernet link, and sends the highest priority AT offline alarm command set to SMS module 108. PLC module 107 generates alarm SMS text based on the structured alarm event packet and encodes it according to the text mode or PDU mode supported by SMS module 108.

[0138] The SMS module 108 uses its built-in wireless radio frequency antenna to access the mobile cellular network and send text information containing the loss diagnosis code, the last known concentration of hazardous gas, the alarm time, and the last obtained geographical coordinates to the pre-stored management communication terminal.

[0139] To prevent frequent SMS messages from being sent due to data fluctuations around the alarm threshold, the PLC module 107 is equipped with an alarm cooling timer. Within a preset time period after the first successful SMS message transmission, the PLC module 107 blocks duplicate SMS message sending commands of the same type to reduce repeated alarm transmissions caused by threshold fluctuations.

[0140] Specific application examples: 1. Scene preset and parameter distribution (corresponding to S10) 2. A municipal engineering team needs to enter an underground, enclosed sewage pipe network to perform dredging work. The safety management personnel create a new form (No.: TASK-20260513-01) using touchscreen 101.

[0141] Time parameter: Permitted job start timestamp Set to 08:00:00, end timestamp Set to 12:00:00.

[0142] Personnel parameters: Authorized operator "Zhang San" (ID: OP-001), the system extracts his facial feature vector to generate a temporary whitelist.

[0143] Task and baseline parameters: The document carries the business code for this operation. For hydrogen sulfide (H2S) gas, the system has a preset standard allowable upper limit concentration. It is 10 ppm.

[0144] Physiological threshold: Zhang San's normal resting heart rate range is set to... bpm.

[0145] 3. Entry verification calculation (corresponding to S20) At 08:15:00, Zhang San arrived at the facial recognition camera 105 at the entrance of the pipeline.

[0146] The camera extracts its current facial feature vector. and compared with the authorized feature vector in the whitelist Perform dot product and norm operations. Calculate the cosine similarity. .

[0147] System preset judgment threshold Since 0.88 ≥ 0.80, the face matching state quantity... .

[0148] The system extracts the admission judgment time. It is 08:15:02, because and Admission state quantity The output is 1. PLC module 107 allows the passage and generates a verification log, which is then reported via the MQTT protocol.

[0149] Meanwhile, PLC module 107 matches the dynamic adjustment coefficient corresponding to the high-risk environment based on the business code "M_DESILT_01". PLC module 107 reconstructs the specific dynamic H2S concentration upper limit threshold for this task. And write it into the underlying control register.

[0150] 4. Heterogeneous alignment and cross-coupling early warning judgment (corresponding to S30, S40) Sliding window during the time period from 09:30:00 to 09:30:10 Inside, the gas alarm 300 (sampling rate 1Hz) and the monitoring wristband 400 (sampling rate 0.5Hz) continuously transmit data. The PLC module 107 first uses zero-order hold interpolation to complete the spatiotemporal alignment of the two.

[0151] At this moment, the absolute concentration of H2S climbed to 7.5 ppm (since 7.5 < 8 ppm, it did not reach the upper limit of the reconstructed absolute concentration, and the environmental static state quantity was not reached). At the same time, Zhang San's heart rate rose to 95 bpm (since 95 < 100 bpm, it did not reach the absolute upper limit, and was considered a physiological resting state quantity). .

[0152] However, the first-order trend analysis performed by PLC module 107 revealed: H2S concentration change rate Greater than the environmental trend gradient threshold ; Zhang San's heart rate variability Greater than the physiological trend gradient threshold .

[0153] Both components experience a rapid increase within the same sliding window, exhibiting high coupling. PLC module 107 determines that the environment has exerted acute initial pressure on the human body and forcibly outputs a coupling warning decision variable. Global alarm decision status quantity The system intercepted potential hazards before they reached a lethal concentration.

[0154] 5. Alarm Dispatch Response Local (Level 1): The audible and visual alarm sounds, and a red OSD overlay pops up on the touchscreen 101, displaying "Warning: Both Hydrogen Sulfide and Heart Rate Change Rate Exceed Standards".

[0155] Cloud (Level 2): ​​The transparent transmission module 104 pushes data packets carrying GPS coordinates (N31.23, E121.47) and the type of anomaly to the cloud.

[0156] Offline SMS and Degradation Fault Tolerance (Level 3): At 09:30:15, due to workers' panicked evacuation, the Ethernet cable on site broke, causing PLC module 107 to continuously lose heartbeat packets, and the monitoring wristband's 400 data abruptly became invalid due to packet loss. PLC module 107's logic triggered "Trapped and Disconnected Diagnosis," immediately activating the degradation fault tolerance mechanism. It stopped relying on Ethernet and, via AT commands, drove SMS module 108 to seize the independent mobile cellular channel, sending the message "Emergency alarm coordinates N31.23, E121.47, operator OP-001's environment H2S is rapidly deteriorating and network connectivity is lost, please provide immediate rescue!" in PDU mode to the safety supervisor's mobile phone. Subsequently, a cooling timer started, locking this type of SMS sending channel for 5 minutes to prevent communication channel congestion.

[0157] Experimental verification and effect comparison: To verify the technical advantages of this system in the two core mechanisms of "heterogeneous data offloading" and "time-domain filtering anti-shake early warning", a simulation test bench was built for comparative testing.

[0158] Experiment 1: Test on the impact of heterogeneous data offloading on the system's main control computing power Test objective: To verify whether the "media stream physical penetration transmission mechanism" proposed in step S30 can effectively reduce the load on the PLC main control system and ensure the real-time performance of control commands.

[0159] Test method: Set up two system architectures: Traditional architecture group: The 4Mbps bitrate 1080P video stream generated by the surveillance camera and the gas sensor data are all received and forwarded by the main control processing unit (MCU / PLC) via a unified bus.

[0160] The architecture of this invention adopts the mechanism of this patent. The video stream is directly transmitted to the display and the cloud via MAC addressing through the third switch 201 and the second switch 103. Only low-bandwidth sensing control data and aligned timestamps enter the PLC module 107.

[0161] At the 60th second of the test, two channels of high-frequency gas concentration change data were simultaneously injected into the system. Performance comparison data: Monitoring software is used to extract the CPU occupancy rate of the two main control units and the end-to-end response delay of alarm commands.

[0162] Experiments show that after the video stream is connected, the CPU utilization of the traditional architecture group rises to 82%-89%, and the instruction response delay is as high as 420ms when handling sudden alarm logic. In contrast, the PLC module 107 of the architecture group of this invention, because it does not need to handle heavy video decoding packets, maintains a CPU utilization rate in a safe low range of 15%-22%, and the alarm instruction response delay is consistently within 35ms. This invention fundamentally eliminates the risk of video streams blocking the underlying control bus by physically and logically isolating high-bandwidth data.

[0163] Experiment 2: Testing the "Lead Ahead" Advantage of the Environment-Physiology Cross-Coupling Early Warning Mechanism Test objective: To verify the first-order trend cross-coupling early warning mechanism introduced in this invention. Compared to the traditional static threshold determination mechanism, it can provide more time for escape when dealing with sudden changes in confined space conditions.

[0164] Test Method: A standard concentration of toxic substitute gas was released in a sealed test chamber to simulate a toxic gas leak. Test personnel, wearing wristbands, were placed inside the chamber, and changes in their physiological indicators were recorded. The trigger timestamps for both the "traditional static limit alarm (alarm only triggered when the absolute concentration limit is reached)" and the "cross-coupled alarm of this invention (early alarm triggered by both concentration and heart rate first-order derivative exceeding the limit)" were recorded.

[0165] Comparison of Results: Experimental results show that in five simulated sudden leaks, due to the exponential increase in toxic gas volatilization and the human body's physiological stress response, the traditional algorithm only triggered an alarm on average at 45 seconds when the concentration reached its absolute upper limit. In contrast, the cross-coupling algorithm of this invention, at 18 seconds when the absolute concentration was within a safe range, accurately detected the "double exceedance of environmental and physiological gradients," triggering the alarm earlier. This algorithm provides an average of 27 seconds of crucial evacuation time for personnel working in confined spaces, significantly enhancing the system's defensive capabilities.

[0166] See attached document Figure 7 The graph contains two subgraphs, one above the other. Figure 7 (a) is a graph showing the first derivative rate of change of the target gas concentration. Figure 7 (b) for Figure 7 (a) Synchronously aligned curves of the first derivative rate of change of human heart rate. The vertical dashed line in the figure marks the traditional "static threshold alarm point," while the shaded area to the left of the vertical dashed line visually shows the rate of change of concentration (e.g., ...) Figure 7 (a) shown) and heart rate change rate (e.g. Figure 7 (b) As shown, when the system simultaneously pierces the absolute value threshold of their respective trend gradients (horizontal dashed line) within a certain time window, the system triggers the cross-coupling judgment in advance before the absolute danger occurs. The spatiotemporal correspondence of ).

[0167] See attached document Figure 8 The diagram shows two types of bars side-by-side for each operating condition: dark bars represent the traditional centralized architecture, and light bars represent the heterogeneous off-grid transmission architecture of this invention. The envelope trend clearly shows that in the third and fourth operating conditions, the dark bars exhibit an exponential surge, approaching the system's computing power limit, while the light bars, because the media stream does not enter the main control memory, show an extremely gradual increase in height, strongly supporting the physical logic of data off-grid transmission in ensuring low latency control response.

[0168] See attached document Figure 9The figure overlays two sets of scatter plot markers: hollow circular scatter plots are distributed on the lagging time axis position of the traditional algorithm's triggered action; solid pentagonal scatter plots are distributed on the leading time axis position representing the triggered action of the cross-coupling algorithm of this invention. The dense distribution of the scatter plots on the left side of the time axis (leading zone) visually demonstrates the technical superiority of this invention in capturing minute signs of deterioration and shortening the system's early warning dead zone time without relying on single-point extreme data, by utilizing the joint trend of multiple parameters.

Claims

1. A method for multi-source heterogeneous data fusion and decision-making oriented to a restricted space, characterized in that, The method is executed collaboratively by multiple terminals, including a touchscreen, a face recognition camera, a switch, and a PLC module, based on a communication network connection, and includes the following steps: The touchscreen generates electronic construction task documents based on the input work restriction parameters; The facial recognition camera acquires the facial feature data of the object to be verified, and performs identity verification in conjunction with the electronic construction task document. The PLC module controls the generation of an entry release command. The PLC module dynamically reconstructs the dynamic concentration upper limit threshold for the current task cycle based on the business code in the electronic construction task document. The switch acquires audio and video media data streams, gas concentration data, and individual heart rate and blood pressure data of workers within the confined space, and performs heterogeneous data splitting to separate the audio and video media data streams from low-bandwidth sensor control data, so that the audio and video media data streams do not enter the main control operation process of the PLC module. The PLC module, based on the gas concentration data and heart rate data in the low-bandwidth sensor control data, and combined with the dynamic concentration upper limit threshold, comprehensively performs limit static judgment and first-order trend cross-coupled early warning judgment. When the judgment result meets the abnormal conditions, the PLC module generates a global alarm judgment status quantity and triggers a multi-level alarm scheduling process.

2. The multi-source heterogeneous data fusion and decision-making method according to claim 1, characterized in that, The electronic construction task document includes the permission time parameter and the business code; The facial recognition camera acquires the facial feature data, and performs identity verification in conjunction with the electronic construction task document. The PLC module controls the generation of the entry release command, specifically including: The face recognition camera extracts the facial feature vector of the current object to be verified from the acquired facial feature data; Obtain the temporary authorization feature comparison whitelist library associated with the electronic construction task document, and extract the authorization feature vector in the temporary authorization feature comparison whitelist library; Calculate the cosine similarity between the facial feature vector of the current object to be verified and the authorized feature vector; When the cosine similarity is not less than the preset facial feature matching similarity judgment threshold, the facial feature matching is determined to be successful and a face matching state quantity with a value of 1 is generated. The PLC module obtains the admission judgment time and determines whether the admission judgment time is between the start timestamp and end timestamp of the licensed operation converted by the license time parameter. When the face matching status value is 1 and the access judgment time is between the start time stamp and the end time stamp of the permission operation, the PLC module generates the entry release command and sets the access status value to 1. When the face matching status value is not 1, or when the admission judgment time is not between the start time stamp and the end time stamp of the permission operation, the PLC module sets the admission status value to 0.

3. The multi-source heterogeneous data fusion and decision method according to claim 2, characterized in that, The PLC module dynamically reconstructs the dynamic concentration upper limit threshold for the current task cycle based on the business code in the electronic construction task document, specifically including: Extract the business code from the electronic construction task document, call the threshold adjustment coefficient function corresponding to the business code mapping inside the PLC module, and obtain the output adjustment coefficient; By multiplying the system's preset standard allowable concentration limit by the adjustment coefficient, the dynamic concentration upper limit threshold for the corresponding target gas in the current mission cycle is calculated.

4. The multi-source heterogeneous data fusion and decision method according to claim 1, characterized in that, The process of performing heterogeneous data splitting, separating the audio and video media data stream from the low-bandwidth sensor control data, specifically includes: The switch reads the MAC address and port mapping relationship of Ethernet packets; The audio and video media data stream with the touch screen or transparent transmission module as the target node is directly forwarded to the target receiving port; The gas concentration data, heart rate data, and blood pressure data are sent to the PLC module as low-bandwidth sensing and control data for unified reception and register address allocation, forming a logical split from the audio and video media data stream on the processing link.

5. The multi-source heterogeneous data fusion and decision method according to claim 4, characterized in that, After receiving the low-bandwidth sensor control data, the PLC module also performs heterogeneous timestamp alignment processing, specifically including: Since the gas concentration data, heart rate data, and blood pressure data arrive at different frequencies, the PLC module constructs a circular data buffer based on the current sampling time in the register and caches the gas concentration data, heart rate data, and blood pressure data into the circular data buffer. The PLC module uses zero-order hold interpolation to timestamp the data in the annular data buffer and reconstructs a comprehensive feature matrix for subsequent execution of the first-order trend cross-coupling early warning judgment.

6. The multi-source heterogeneous data fusion and decision-making method according to claim 4, characterized in that, In the integrated execution of the static limit judgment and the first-order trend cross-coupling early warning judgment, the static limit judgment introduces a time-domain filtering anti-shake mechanism, and the judgment logic includes: Set a preset confirmation time window; When the gas concentration data sampling value of any target gas is continuously higher than the corresponding dynamic concentration upper limit threshold within the preset confirmation time window, the PLC module sets an environmental abnormality status.

7. The multi-source heterogeneous data fusion and decision-making method according to claim 6, characterized in that, The specific logic for the first-order trend cross-coupling early warning judgment includes: Determine whether the heart rate data or blood pressure data of any individual worker exceeds a preset physiological safety threshold. If it does, generate a physiological abnormality state quantity with a value of 1. Extract the gas concentration data and heart rate data within the preset time window, and calculate the rate of change of the target gas concentration and the rate of change of the individual worker's heart rate within the most recent time window, respectively. When it is determined that the concentration change rate exceeds the preset environmental trend gradient threshold and the heart rate change rate exceeds the preset physiological trend gradient threshold, it is determined that the current environment has had an initial impact on a specific individual, and a coupled early warning decision variable with a value of 1 is output. The trigger condition for the global alarm decision status quantity is that any of the following conditions are met: The abnormal physiological state quantity of any individual worker is 1, the abnormal environmental state quantity is 1, or the coupled early warning decision variable is 1.

8. The multi-source heterogeneous data fusion and decision-making method according to claim 1, characterized in that, The multi-level alarm scheduling process includes performing network degradation fault tolerance judgment and offline SMS forced alarm, specifically including: The background program of the PLC module continuously monitors the heartbeat status of the Ethernet port. When the PLC module fails to receive an Ethernet heartbeat packet for a preset number of consecutive cycles or the received data suddenly becomes invalid packet loss data, and there is an abnormal environmental state, it is determined that an accident has occurred inside the confined space, causing network paralysis or personnel to be trapped and disconnected, and the network degradation fault tolerance mechanism is triggered. The PLC module stops relying on the Ethernet link and sends a set of offline alarm instructions to the SMS module with a serial communication physical connection. This drives the SMS module to use the mobile cellular network to send an alarm SMS containing a disconnection diagnosis code and the last known geographical coordinates to a preset terminal.

9. The multi-source heterogeneous data fusion and decision method according to claim 1, characterized in that, The multi-level alarm scheduling process also includes performing local fusion display processing, specifically including: The graphics rendering engine inside the touch screen performs hardware decoding operations on the audio and video media data stream to restore and output dynamic video frames. Based on OSD screen character overlay technology, a transparent data overlay layer is created on top of the underlying video frame; The touchscreen reads the real-time values ​​of the gas concentration data, heart rate data, and blood pressure data uploaded by the PLC module, converts them into text or graphic controls, and draws a preset display area superimposed on the transparent data overlay layer.

10. A multi-source heterogeneous data fusion and decision system for a confined space, characterized in that, The system, applied to the multi-source heterogeneous data fusion and decision-making method according to any one of claims 1-9, comprises: The touch screen box, located outside the confined space, contains a touch screen, a first switch, a second switch, a transparent transmission module, a face recognition camera, a first intercom module, a PLC module, an SMS module, and a GPS module. A camera box and a gas alarm are arranged inside a confined space. The camera box contains a third switch, a monitoring camera, a second intercom module, and a wristband radar for receiving wireless signals. And a monitoring wristband that collects heart rate and blood pressure data, and transmits the data wirelessly to the wristband radar; The modules and devices within the system are connected via Ethernet, industrial bus interface, or serial interface.

Images