Power plant specific area personnel behavior identification and early warning system

Abstract

The invention discloses a power plant specific area personnel behavior identification and early warning system, and relates to the technical field of industrial safety intelligent monitoring, the power plant specific area personnel behavior identification and early warning system comprises a data acquisition module, an edge calculation module and a central processing platform, the edge computing module carries out lightweight processing and localized decision making on original data, the load of the central processing platform is reduced, and the central processing platform realizes full-process safety management and control of a high-risk area through multi-modal data fusion, dynamic authority management and trajectory analysis. According to the invention, lightweight AI models are embedded in various cameras of a work card authentication terminal and a visual perception unit, preliminary reasoning is directly completed at the work card authentication terminal and the cameras, so that the work card authentication terminal and the cameras have edge computing capability, and independent edge servers or industrial personal computers are deployed near data sources of various regions of a power plant. And serving as a regional edge computing node.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial safety intelligent monitoring, and in particular to a personnel behavior recognition and early warning system in a specific area of ​​a power plant. Background Art

[0002] Intelligent industrial safety monitoring technology is a core area for ensuring the safety of personnel and equipment in high-risk industrial environments. This field focuses on proactive safety management and control of personnel behavior, environmental risks, and equipment status through multimodal sensing, real-time data analysis, and intelligent decision-making. Power plants, as typical high-risk environments, present risks such as high-temperature and high-pressure equipment, radiation exposure, and toxic gas leaks, placing even more stringent requirements on monitoring systems.

[0003] Existing technologies generally rely on manual inspections and fixed alarm devices, forming a passive monitoring mode. When personnel violate regulations or mistakenly enter restricted areas, the system cannot proactively identify real-time risks, resulting in delayed responses. The lag makes accident prevention capabilities weak, and violations are often only discovered in retrospect. The single nature of the perception mode leads to insufficient anti-interference capabilities. Traditional solutions use isolated sensors, and the data of each system is fragmented and cannot be cross-verified. This single-mode system is difficult to adapt to the complex working conditions of high temperature, high pressure, and high dust in power plants. The authority configuration relies on manual updates to the schedule, and the change delay often exceeds 2 hours. It is impossible to track the actual activity path of personnel in the authorized area, resulting in the inability to block behaviors such as unauthorized stays and mistaken entry into radiation zones in real time. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a personnel behavior recognition and early warning system in a specific area of ​​a power plant.

[0005] In order to achieve the above-mentioned objectives, the present invention adopts the following technical solutions: a personnel behavior identification and early warning system in a specific area of ​​a power plant, including a data acquisition module, an edge computing module and a central processing platform. The data acquisition module is responsible for the collection and preliminary processing of multi-source heterogeneous data, including personnel identity, behavior, location and environmental parameters. The edge computing module performs lightweight processing and localized decision-making on the original data to reduce the load on the central processing platform. The central processing platform realizes full-process security management and control of high-risk areas through multimodal data fusion, dynamic authority management and trajectory analysis.

[0006] As a further description of the above technical solution: The data acquisition module includes: the visual perception unit captures personnel movements and environmental information through thermal imaging cameras, wide-angle high-definition cameras, and behavioral analysis cameras; the work badge authentication terminal uses RFID / NFC to read work badge information and integrates liveness detection to prevent identity fraud; the wearable device unit uses the UWB positioning module and acceleration sensor to report personnel location and abnormal status in real time; the environmental perception unit collects temperature, humidity, gas concentration, radiation dose, and marks high-risk environmental events.

[0007] As a further description of the above technical solution: The edge computing module deploys a lightweight AI model to extract behavioral features, filter invalid data, and perform data compression and structured packaging to adapt to a unified transmission protocol.

[0008] As a further description of the above technical solution: The edge computing module pre-installs a security policy library to trigger local responses, reduce cloud dependence, blur the facial and ID information of irrelevant personnel in real time, and only upload the desensitized key frames of risk events.

[0009] As a further description of the above technical solution: The central processing platform calibrates the badge punch-in time, video timestamp and UWB positioning coordinates, builds a global spatial consistency view, and uses DS evidence theory to comprehensively determine risks.

[0010] As a further description of the above technical solution: The central processing platform updates permissions based on the position-area-time matrix, generates dynamic verification logs, and issues temporary permissions through mobile approval. It limits areas, time periods, and operation types, draws electronic fences, detects trajectory deviations or stay timeouts, optimizes response strategies through linkage with environmental parameters, and pushes risk levels to the early warning module.

[0011] As a further description of the above technical solution: The system process is as follows: S1: Multi-source data acquisition Visual perception unit: thermal imaging cameras, wide-angle HD cameras, and behavior analysis cameras capture people's outlines, safety equipment wearing status, and high-risk behaviors in real time; ID card authentication terminal: mandatory verification of employee ID card information at area entrances, and integrated liveness detection to prevent fraudulent use; Wearable device unit: Smart helmets report personnel location and abnormal status in real time; Environmental sensing unit: temperature, humidity, gas concentration, and radiation sensors monitor environmental risk parameters; S2: Edge computing preprocessing Lightweight AI reasoning: The camera-side deployment model extracts behavioral features in real time and filters out invalid data; Localized rule execution: preset rules trigger local responses; Data desensitization and compression: Blurring the faces of irrelevant people in the video, retaining only the key frames of risk events, and compressing them before uploading; S3: Data Upload The pre-processed standardized data is transmitted to the central processing platform via 5G private network or Ethernet dual-channel redundancy, and is automatically cached and incrementally restored in the event of network interruption; S4: Multimodal Data Fusion Spatiotemporal alignment: calibrate badge punch-in time, camera timestamps, and positioning data to millisecond-level timing; Fusion decision-making: Use DS evidence theory to comprehensively determine risk and calculate risk level by combining environmental parameters; S5: Dynamic permission verification Based on the position-region-time permission matrix, personnel permissions are matched in real time. Temporary permissions are approved and issued through the mobile terminal and automatically reset upon expiration. S6: Graded warning trigger Level 1 warning: not wearing safety equipment, entering an unauthorized area; Response action: On-site sound and light alarm, push reminder to the wristband of the person involved; Level 2 warning: exceeding the authorized stay limit, illegal use of tools; Response actions: Lock area access, link the DCS system to limit power, and push work orders to management personnel; Level 3 warning: people are stranded in fire / leakage, and high risks are superimposed; Response actions: Initiate a plant-wide emergency shutdown, broadcast evacuation instructions, push AR escape routes, and dispatch rescue personnel; S7: Edge Responsive Execution Early warning instructions are sent to edge nodes via incremental protocols, directly triggering local actions. Deep integration with the power plant DCS system allows for automatic device linkage. S8: Full-link traceability Stores anonymized data associated with events; Supports multi-dimensional search by time, person, and region, and one-click tracing back the complete event chain; S9: Compliance report generation Automatically extract safety indicators and generate reports that comply with ISO45001 standards, suitable for regulatory review.

[0012] The present invention has the following beneficial effects: 1. In this invention, we first build a set of active safety protection system, which solves the passivity, singleness and authority fragmentation problems of traditional power plant monitoring system through technological innovation, and realizes the full process closed-loop management of personnel behavior in high-risk areas. The system forms a closed-loop verification through the four-dimensional data fusion of work badge authentication (identity) - liveness detection (face comparison) - trajectory tracking (behavior) - environmental perception (risk parameters). The work badge authentication terminal integrates RFID reading and writing and near-infrared camera, and uses ArcFace algorithm to compare the work badge photo with facial features in real time. If the cosine similarity is lower than the threshold, it determines that the identity is used fraudulently, and resists forgery attacks through random blinking / head turning detection. At the same time, UWB positioning The positioning module realizes centimeter-level real-time positioning, and combines with thermal imaging cameras to generate personnel outline coordinates, effectively penetrating smoke, dust and low-light environments. When the deviation between the work badge authorization area and the actual positioning exceeds the threshold, it triggers the judgment of unauthorized behavior based on DS evidence theory, improves the accuracy of identifying authority violations, and reduces the false alarm rate. It innovatively designs the position-area-time permission matrix, pre-defines the accessible areas and time periods of different positions, and automatically synchronizes with the scheduling system to avoid manual configuration errors. Temporary permissions are dynamically issued through mobile approval, limiting areas, time periods and operation types. After the permission expires, it automatically triggers the access control reset. The system compares the personnel's work badge permissions with the time and space coordinates in real time and generates a dynamic verification log.

[0013] 2. In the present invention, an embedded and regional hybrid edge computing architecture is adopted. Lightweight AI models are embedded in various cameras of the work badge authentication terminal and the visual perception unit, and preliminary reasoning is completed directly at the work badge authentication terminal and the camera end, so that the work badge authentication terminal and the camera have edge computing capabilities, which can be regarded as "minimized units of edge computing nodes". However, the computing power is limited and only supports specific tasks. Independent edge servers or industrial computers are deployed near the data sources in various areas of the power plant as regional edge computing nodes. The OpenPose algorithm extracts 18 skeleton key points and combines the optical flow method to analyze high-risk behaviors. A liveness detection model is embedded in the work badge terminal to reduce the redundancy of original data. The faces of irrelevant people in the video stream are blurred in real time, and only key frames of high-risk events are retained; sensitive data is retained locally, and the desensitized data is passed Dual-channel upload via 5G private network or Ethernet reduces bandwidth consumption. Data in high-risk areas is prioritized based on the CPU / memory usage of edge nodes. Desensitized video clips, work badge records, positioning trajectories, and environmental parameters are structured for storage, supporting one-click backtracking by time, personnel, and area. Thermal imaging complements wide-angle HD cameras, with wide-angle lenses covering blind spots. The dynamic HDR algorithm adapts to strong light / backlight scenarios. Gas sensors monitor CO and H2S concentrations in real time, and radiation sensors are linked to positioning data to delineate exposed areas. Compliance reports are automatically generated, extracting indicators such as the number of violations and response delay, reducing audit time. Through four major innovations: multimodal fusion verification, dynamic permission adaptation, edge intelligent computing, and hierarchical device linkage, the system fills the technical gaps in real-time, accuracy, and environmental adaptability in the field of industrial safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0016] Reference Figure 1, an embodiment provided by the present invention: a personnel behavior recognition and early warning system in a specific area of ​​a power plant, including a data acquisition module, an edge computing module and a central processing platform. The data acquisition module is responsible for the collection and preliminary processing of multi-source heterogeneous data, including personnel identity, behavior, location and environmental parameters. The edge computing module performs lightweight processing and localized decision-making on the original data to reduce the load of the central processing platform. The central processing platform realizes full-process safety management and control of high-risk areas through multimodal data fusion, dynamic authority management and trajectory analysis. The data acquisition module includes: a visual perception unit captures personnel movements and environmental information through thermal imaging cameras, wide-angle high-definition cameras, and behavior analysis cameras. The work badge authentication terminal uses RFID / NFC to read the work badge information and integrates liveness detection to prevent identity fraud. The wearable device unit reports personnel location and abnormal status in real time through the UWB positioning module and acceleration sensor. The environment The sensing unit collects temperature, humidity, gas concentration, and radiation dose, and marks high-risk environmental events. The edge computing module deploys a lightweight AI model to extract behavioral features, filter out invalid data, and perform data compression and structured packaging to adapt to a unified transmission protocol. The edge computing module presets a security policy library to trigger local responses, reduce cloud dependence, blur the facial and ID information of irrelevant personnel in real time, and only upload the desensitized key frames of risk events. The central processing platform calibrates the ID card punch-in time, video timestamp, and UWB positioning coordinates to build a global spatial consistency view and use DS evidence theory to comprehensively determine risks. The central processing platform updates permissions based on the position-area-time period matrix and generates dynamic verification logs. Temporary permissions are approved and issued through the mobile terminal, limiting areas, time periods, and operation types. Electronic fences are drawn to detect trajectory deviations or stay timeouts, linking environmental parameters to optimize response strategies, and pushing risk levels to the early warning module.

[0017] The system process is as follows: S1: Multi-source data acquisition Visual perception unit: thermal imaging cameras, wide-angle HD cameras, and behavior analysis cameras capture people's outlines, safety equipment wearing status, and high-risk behaviors in real time; ID card authentication terminal: mandatory verification of employee ID card information at area entrances, and integrated liveness detection to prevent fraudulent use; Wearable device unit: Smart helmets report personnel location and abnormal status in real time; Environmental sensing unit: temperature, humidity, gas concentration, and radiation sensors monitor environmental risk parameters; S2: Edge computing preprocessing Lightweight AI reasoning: The camera-side deployment model extracts behavioral features in real time and filters out invalid data; Localized rule execution: preset rules trigger local responses; Data desensitization and compression: Blurring the faces of irrelevant people in the video, retaining only the key frames of risk events, and compressing them before uploading; S3: Data Upload The pre-processed standardized data is transmitted to the central processing platform via 5G private network or Ethernet dual-channel redundancy, and is automatically cached and incrementally restored in the event of network interruption; S4: Multimodal Data Fusion Spatiotemporal alignment: calibrate badge punch-in time, camera timestamps, and positioning data to millisecond-level timing; Fusion decision-making: Use DS evidence theory to comprehensively determine risk and calculate risk level by combining environmental parameters; S5: Dynamic permission verification Based on the position-region-time permission matrix, personnel permissions are matched in real time. Temporary permissions are approved and issued through the mobile terminal and automatically reset upon expiration. S6: Graded warning trigger Level 1 warning: not wearing safety equipment, entering an unauthorized area; Response action: On-site sound and light alarm, push reminder to the wristband of the person involved; Level 2 warning: exceeding the authorized stay limit, illegal use of tools; Response actions: Lock area access, link the DCS system to limit power, and push work orders to management personnel; Level 3 warning: people are stranded in fire / leakage, and high risks are superimposed; Response actions: Initiate a plant-wide emergency shutdown, broadcast evacuation instructions, push AR escape routes, and dispatch rescue personnel; S7: Edge Responsive Execution Early warning instructions are sent to edge nodes via incremental protocols, directly triggering local actions. Deep integration with the power plant DCS system allows for automatic device linkage. S8: Full-link traceability Stores anonymized data associated with events; Supports multi-dimensional search by time, person, and region, and one-click tracing back the complete event chain; S9: Compliance report generation Automatically extract safety indicators and generate reports that comply with ISO45001 standards, suitable for regulatory review.

[0018] As the perception front end of the system, the data acquisition module is responsible for the precise collection and preprocessing of multi-source heterogeneous data. Its core modules include the visual perception unit, the work badge authentication terminal, the wearable device unit, and the environmental perception unit. Various cameras of the work badge authentication terminal and the visual perception unit are embedded with lightweight AI models. The visual perception unit achieves precise monitoring in complex scenes through a combination of multiple types of cameras. The thermal imaging camera generates a heat map based on the difference in infrared radiation. The heat map is generated by an uncooled microbolometer, which effectively penetrates the interference of smoke, dust and low-light environments, and outputs the thermal imaging video stream of the person's X, Y, Z contour coordinates and movement direction in real time. The wide-angle high-definition camera is equipped with a 120° ultra-wide-angle lens to cover the visual blind spots in equipment-intensive areas. Combined with the embedded YOLOv5 model, real-time video stream analysis of the monitoring blind spots is performed to identify the wearing status of safety equipment in real time. The YOLOv5 model divides the image into grids, and predicts the bounding box and class probability for each grid. The formula is: , where the loss function of the bounding box often uses CIoU loss to optimize the prediction , , are the bounding box center coordinates, are the width and height of the bounding box, is the confidence level, which indicates the probability of an object existing in the bounding box, ranging from 0 to 1. is a class probability vector, which indicates the probability that the object belongs to each class, such as "helmet wearing" or "not wearing", Is the center point of the prediction box The center point of the real frame The Euclidean distance between is the diagonal length of the minimum bounding box and the real box, are the width and height of the real frame, is the aspect ratio consistency measure, is a weighting factor used to balance geometric errors and adapt to strong light reflection or backlight scenes through the dynamic HDR algorithm. The behavior analysis camera extracts 18 skeletal key points of the human body based on the OpenPose algorithm and constructs a spatiotemporal action map, such as the rate of change of joint angles and motion trajectories. OpenPose detects and associates key points. The key point detection formula is: ,in, It is The positions of the key nodes are associated with the linear distribution of PAFs as follows: , For the location At the key point The confidence map value ranges from 0 to 1. is the Diracdelta function, used to indicate the position of the key point, It is Personal The confidence of the key points, PartAffinityField vector, indicating the direction of the limb, is the interpolation point between two points on the limb, is a normalization parameter ranging from 0 to 1, From the key point arrive The unit direction vector of The correlation score of key point pairs is used to construct human posture. The spatial coordinates of skeleton key points are parsed frame by frame from the video stream. The motion vector between consecutive frames is calculated by combining the optical flow method to identify high-risk behaviors such as climbing railings, running, or illegal operation of equipment. The optical flow method is based on the constant brightness assumption, and the core equation is the Horn-Schunck optical flow constraint: , where the global optimization objective function is: , is the image brightness, is the grayscale value, is the spatial gradient of the image in the x and y directions, calculated by operators such as Sobel, is the change of the image over time, that is, the time gradient, is the optical flow vector, which indicates the speed of the pixel in the x and y directions, in pixels / frame. is the regularization parameter, balancing the data term and the smoothing term, usually set to 1.0. It is the spatial derivative of the optical flow vector and is used for smoothing constraints. The work badge authentication terminal deploys a high-frequency RFID reader at the entrance of a specific area to enforce verification of employee work badge permissions. It reads the radio frequency signal of the chip embedded in the work badge through the principle of electromagnetic induction, analyzes the employee ID, position and permission information, and forcibly triggers the authentication process at the entrance of the area. It is released only when the work badge permissions match the preset area-position matrix. If the card is not punched in or the chip is damaged, the terminal triggers a red warning light and an audible and visual alarm, and takes a photo to preserve evidence. To prevent the use of work badges, the terminal integrates a near-infrared camera to collect facial images. Through the liveness detection function, the ArcFace algorithm is used to extract facial features and perform a cosine similarity comparison with the pre-stored photo of the work badge. If the similarity is lower than the threshold, it is determined to be identity fraud, triggering an alarm and generating an evidence snapshot to prevent the use of work badges. ArcFace uses additive angular margin loss to enhance feature discriminability. The loss function is: , where the output after feature extraction is the normalized feature vector , is the eigenvector With the real category The angle in radians between the weight vectors, is the additive angular margin, usually 0.5, used to increase the distance between classes. is the characteristic scaling factor, usually 64, used to magnify the angle difference, is the number of training samples, is the eigenvector and The cosine similarity of the class weight vectors, The facial feature vector is extracted after L2 normalization. It has an anti-fraud design that randomly requires blinking or turning the head to detect liveness and resist photo / video forgery attacks. The wearable device unit includes a smart helmet with an integrated UWB ultra-wideband positioning module and a three-axis accelerometer. It also has a built-in Decawave DW1000 chip. It generates real-time point cloud data of the person's location and measures the distance difference to the positioning base station using the time-of-flight method to achieve centimeter-level real-time positioning. The distance is calculated based on the signal propagation time. , is the distance in meters, is the speed of light, The signal round trip time, in seconds, is the time difference from sending to receiving. The coordinates are reported to the central platform once per second. The built-in MEMS three-axis accelerometer continuously monitors the movement status of personnel, identifies normal movement and falls and collisions through threshold judgment, triggers the local buzzer alarm and uploads the event code. Through multi-base station networking, 4 base stations are deployed and the arrival time difference algorithm is used to achieve 3D space coverage. In the arrival time difference positioning equation, Is the label to The distance difference between the base station and the reference base station, in meters, is the speed of light, The signal reaches The delay of each base station, in seconds, is the time delay of the signal reaching the reference base station, in seconds, are the coordinates of the label personnel, It is Base station coordinates, is the coordinate of the reference base station, formula: The environmental perception unit monitors the physical status of high-risk areas in real time through temperature and humidity sensors, gas concentration sensors and radiation sensors, and assists in behavioral risk assessment. The temperature and humidity sensor DHT22 digital temperature and humidity sensor collects data every 5 seconds and transmits it to the edge node via the LoRaWAN protocol. The gas concentration sensor uses the MQ-135 electrochemical type to detect hazardous gases CO and H2S, and outputs a 0-5V analog signal that is converted by ADC to generate a PPM value. When the CO concentration is ≥50ppm, it is marked as high risk. The radiation sensor uses the LND-712 GM counter tube to detect the intensity of gamma rays, converts it into a radiation dose rate through the pulse counting rate, and links it with the personnel positioning data to delineate the exposure area. All raw data, including visual features, work badge information, positioning coordinates, and environmental parameters, are lightweight pre-processed by the edge computing module and uploaded to the central processing platform in a unified manner to ensure the efficiency of multimodal data spatiotemporal alignment and fusion decision-making.

[0019] Edge computing nodes adopt an embedded and regional hybrid deployment mode. Lightweight AI models are embedded in various cameras of work badge authentication terminals and visual perception units, and preliminary reasoning is completed directly on the work badge authentication terminals and cameras, giving the work badge authentication terminals and cameras edge computing capabilities. They can be regarded as "minimized units of edge computing nodes", but their computing power is limited and only supports specific tasks. Independent edge servers or industrial computers are deployed near the data sources in various areas of the power plant as regional edge computing nodes, taking on local data processing, real-time decision-making and data preprocessing tasks. Such nodes cover multiple cameras, sensors and work badge authentication terminals in the area and process the equipment data in the area. The lightweight AI model deployed on the camera side extracts behavioral features from the video stream in real time, such as the wearing status of safety equipment and skeletal key points, reducing the redundancy of raw data. Based on predefined rules, such as static background frames and low-confidence detection frames, invalid or duplicate data are eliminated, and only features related to high-risk events are retained, reducing data transmission bandwidth consumption. According to the local rule engine and built-in security policy library, such as "no entry without a safety helmet" and "electronic fence crossing judgment threshold", direct Rule matching is performed at the edge to reduce cloud-based decision-making latency. For rule-triggering events, such as unauthorized intrusion or missing equipment, local responses, such as audible and visual alarms and access control locks, are immediately triggered, eliminating the need to wait for central platform instructions before completing emergency responses. Visual features, location coordinates, and environmental parameters are losslessly compressed and structured, adapting to a unified transmission protocol for heterogeneous data. Device clocks are synchronized using the NTP protocol, with millisecond-level alignment of camera timestamps, badge punch-in times, and location data, providing a foundation for multimodal fusion on the central platform. Facial and badge information of irrelevant individuals in the video stream is blurred in real time, retaining only clear data from key frames of risk events. Data is cached by sensitivity level, such as facial features and radiation dose, with non-essential data stored locally. Only desensitized data from risk events is uploaded to the central platform. AI inference tasks are dynamically allocated based on edge node computing power, such as CPU and memory usage, prioritizing data from high-risk areas. Dual-channel redundant transmission, using wired Ethernet and wireless 5G private networks, automatically switches when the primary link is interrupted. Data is automatically cached during network outages and incrementally uploaded upon restoration to ensure data integrity.

[0020] The central processing platform sends dynamic rules to edge nodes through the incremental update protocol to avoid bandwidth pressure caused by the transmission of the full rule base. The edge nodes feedback the update status through the ACK confirmation mechanism. When communication with the edge node is interrupted, the platform activates the locally cached historical rule base and the authority matrix of the last 24 hours, and continuously performs basic risk assessment until the link is restored and the data is resynchronized to ensure decision continuity. The multimodal data fusion unit uses the device clock calibrated based on the NTP protocol to perform millisecond-level timing alignment on the work badge punch-in time, camera video stream timestamp, and UWB positioning coordinates to eliminate the time deviation of multi-source data. Through spatial coordinate system mapping, the location information collected by different devices, such as the contour coordinates of the thermal imaging camera and UWB positioning data, is unified to build a global spatial consistency view, and the DS evidence theory is used to make a comprehensive judgment on the multimodal data, among which the conflict coefficient for: , is the basic probability distribution function, which indicates the degree of support for the hypothesis by different evidences. is an assumption, for example, "authorized area A" or "ultraviolate authority", is the intersection of the hypotheses, is the conflict coefficient, ranging from 0 to 1, indicating the degree of conflict between the evidence. Indicates no conflict. is the combined BPA, which indicates the integration of evidence to the hypothesis The support probability of Is an empty set, indicating the no-intersection hypothesis, and the formula is: , for identity and trajectory conflict detection, when the work badge authorization area is area A, but UWB positioning shows that the person is in area B, it is determined to be an unauthorized behavior, and the behavior and environment are correlated. The risk level is calculated by combining environmental parameters and personnel actions to output fusion decision results, such as "high-risk crossing the boundary" and "low-risk equipment missing", which are used for risk warning and authority management. The dynamic authority management unit realizes precise control based on the position-area-time period authority matrix, and predefines the binding relationship between positions and areas and time periods, such as "boiler inspectors can enter the boiler core area from 8:00 to 17:00", supports the scheduling system to automatically synchronize and update the authority configuration, and compares in real time The personnel badge authority and current time and space coordinates are combined to generate a dynamic authority verification log, such as "Maintenance personnel entered the radiation zone at 18:05 - authority expired." For the issuance and revocation of temporary authority, a temporary pass code is issued through the mobile terminal approval process. For example, for emergency repair tasks, the area, time period, and operation type are limited. When the authority expires or the task is completed, the area access control is automatically reset to prohibit unauthorized personnel from staying. The trajectory analysis unit dynamically generates an electronic fence, draws a real-time personnel activity heat map based on UWB positioning data, and automatically generates the electronic fence boundary. Combined with the dynamic authority matrix, it determines cross-border behavior and triggers a trajectory deviation alarm, such as "The personnel deviates from the authorized path 2.5 meters", monitor the length of time people stay in high-risk areas. If the safety threshold is exceeded, the hierarchical evacuation protocol will be activated, such as broadcast warnings to automatic power restrictions, and finally emergency shutdowns. Environmental parameters such as sudden temperature rise will be linked to optimize evacuation priorities and dynamically adjust response strategies. The early warning module has a three-level early warning mechanism. The triggering conditions for the first-level early warning are failure to wear safety equipment and mistakenly entering unauthorized areas, such as conflicts between work badge permissions and positioning coordinates. The response action is to trigger the red warning light and buzzer on site for sound and light alarms, and retain a snapshot of the violation event. The triggering condition for the second-level early warning is unauthorized stay timeout, such as personnel staying in a high-risk area for more than the preset safety threshold. , as well as illegal use of tools such as unauthorized operation of equipment, the response action is to lock the area access control, prohibit personnel from continuing to enter or leave, and link the power plant DCS system to perform local power restrictions, such as cutting off non-core circuits, and sending emergency work orders and positioning information to management personnel. The triggering conditions for the third-level warning are major safety hazards, such as fires and leakage accidents where personnel are stranded, and high-risk behaviors superimposed on environmental abnormalities, such as personnel in high-temperature areas not wearing protective clothing. The response action is to trigger the plant-wide emergency shutdown protocol and start the evacuation broadcast, and simultaneously push the optimal escape route to the AR navigation equipment of relevant personnel, and at the same time dispatch the nearest rescue personnel to link with the emergency command center. After the early warning is triggered, the central processing platform sends a response instruction to the edge node based on the multimodal data fusion results, such as work badge permissions, behavior analysis, and environmental parameters. The edge node directly executes the local response, such as sound and light alarms, access control locks, without waiting for cloud confirmation, ensuring millisecond-level delays. It is deeply integrated with the power plant DCS system and supports automatic triggering of equipment protection actions according to the early warning level, such as level 2 early warning power restrictions and level 3 early warning shutdowns. Redundant transmission of instructions is achieved through the 5G private network. When the main link is interrupted, it automatically switches to the backup channel. According to environmental parameters such as radiation dose and gas concentration, the response strategy is dynamically optimized. For example, in high temperature areas Prioritize personnel evacuation over power rationing. The data traceability and audit module stores full-link data and structured storage of raw event-related data, including desensitized video footage, badge verification records, location tracking, and environmental parameters. It supports multi-dimensional retrieval by time, person, and region, retains data version snapshots, and supports one-click backtracking to ensure the integrity and immutability of audit traceability. Compliance reports are generated, automatically extracting security event statistical indicators such as the number of violations and response delays, and generating safety management reports that comply with ISO45001 standards. Customizable export formats are supported to meet the review requirements of regulatory authorities, and all warning events are linked to the original data.

[0021] The data acquisition module collects raw data and transmits it to the edge node. The edge node performs AI reasoning, data compression and desensitization, retaining only the characteristics of high-risk events. The pre-processed standardized data is uploaded to the central processing platform via the 5G private network or Ethernet. After receiving the data, the central processing platform performs multimodal fusion and global decision-making, issues early warning instructions based on the fusion results, and links the DCS system control equipment. Dynamic rules are sent to the edge node through the incremental protocol to ensure local rule synchronization. The early warning response results are fed back to the central processing platform, forming an "identification-decision-response-audit" closed loop, storing full-link data and generating ISO45001 compliance reports, realizing low-latency, high-precision closed-loop safety management, and meeting the active protection needs of high-risk areas in power plants.

[0022] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The personnel behavior recognition and early warning system in specific areas of the power plant is characterized by: It includes a data acquisition module, an edge computing module and a central processing platform. The data acquisition module is responsible for the collection and preliminary processing of multi-source heterogeneous data, including personnel identity, behavior, location and environmental parameters. The edge computing module performs lightweight processing and localized decision-making on the original data to reduce the load on the central processing platform. The central processing platform achieves full-process security control in high-risk areas through multimodal data fusion, dynamic authority management and trajectory analysis.

2. The power plant specific area personnel behavior recognition and early warning system according to claim 1 is characterized by: The data acquisition module includes: the visual perception unit captures personnel movements and environmental information through thermal imaging cameras, wide-angle high-definition cameras, and behavioral analysis cameras; the work badge authentication terminal uses RFID / NFC to read work badge information and integrates liveness detection to prevent identity fraud; the wearable device unit uses the UWB positioning module and acceleration sensor to report personnel location and abnormal status in real time; the environmental perception unit collects temperature, humidity, gas concentration, radiation dose, and marks high-risk environmental events.

3. The personnel behavior recognition and early warning system for a specific area of ​​a power plant according to claim 1 is characterized by: The edge computing module deploys a lightweight AI model to extract behavioral features, filter invalid data, and perform data compression and structured packaging to adapt to a unified transmission protocol.

4. The power plant specific area personnel behavior recognition and early warning system according to claim 1 is characterized by: The edge computing module pre-installs a security policy library to trigger local responses, reduce cloud dependence, blur the facial and ID information of irrelevant personnel in real time, and only upload the desensitized key frames of risk events.

5. The power plant specific area personnel behavior recognition and early warning system according to claim 1 is characterized by: The central processing platform calibrates the badge punch-in time, video timestamp and UWB positioning coordinates, builds a global spatial consistency view, and uses DS evidence theory to comprehensively determine risks.

6. The power plant specific area personnel behavior recognition and early warning system according to claim 1 is characterized by: The central processing platform updates permissions based on the position-area-time matrix, generates dynamic verification logs, and issues temporary permissions through mobile approval. It limits areas, time periods, and operation types, draws electronic fences, detects trajectory deviations or stay timeouts, optimizes response strategies through linkage with environmental parameters, and pushes risk levels to the early warning module.

7. The personnel behavior recognition and early warning system for a specific area of ​​a power plant according to any one of claims 1 to 6, characterized in that: The system process is as follows: S1: Multi-source data acquisition Visual perception unit: thermal imaging cameras, wide-angle HD cameras, and behavior analysis cameras capture people's outlines, safety equipment wearing status, and high-risk behaviors in real time; ID card authentication terminal: mandatory verification of employee ID card information at area entrances, and integrated liveness detection to prevent fraudulent use; Wearable device unit: Smart helmets report personnel location and abnormal status in real time; Environmental sensing unit: temperature, humidity, gas concentration, and radiation sensors monitor environmental risk parameters; S2: Edge computing preprocessing Lightweight AI reasoning: The camera-side deployment model extracts behavioral features in real time and filters out invalid data; Localized rule execution: preset rules trigger local responses; Data desensitization and compression: Blurring the faces of irrelevant people in the video, retaining only the key frames of risk events, and compressing them before uploading; S3: Data Upload The pre-processed standardized data is transmitted to the central processing platform via 5G private network or Ethernet dual-channel redundancy, and is automatically cached and incrementally restored in the event of network interruption; S4: Multimodal Data Fusion Spatiotemporal alignment: calibrate badge punch-in time, camera timestamps, and positioning data to millisecond-level timing; Fusion decision-making: Use DS evidence theory to comprehensively determine risk and calculate risk level by combining environmental parameters; S5: Dynamic permission verification Based on the position-region-time permission matrix, personnel permissions are matched in real time. Temporary permissions are approved and issued through the mobile terminal and automatically reset upon expiration. S6: Graded warning trigger Level 1 warning: not wearing safety equipment, entering an unauthorized area; Response action: On-site sound and light alarm, push reminder to the wristband of the person involved; Level 2 warning: exceeding the authorized stay limit, illegal use of tools; Response actions: Lock area access, link DCS system to limit power, and push work orders to management personnel; Level 3 warning: people are stranded in fire / leakage, and high risks are superimposed; Response actions: Initiate a plant-wide emergency shutdown, broadcast evacuation instructions, push AR escape routes, and dispatch rescue personnel; S7: Edge Responsive Execution Early warning instructions are sent to edge nodes via incremental protocols, directly triggering local actions. Deep integration with the power plant DCS system allows for automatic device linkage. S8: Full-link traceability Stores anonymized data associated with events; Supports multi-dimensional search by time, person, and region, and one-click tracing back the complete event chain; S9: Compliance report generation Automatically extract safety indicators and generate reports that comply with ISO45001 standards, suitable for regulatory review.

Images