New energy station operation site safety monitoring and early warning method

Abstract

The invention is applicable to the technical field of safety monitoring and early warning, and provides a new energy station operation site safety monitoring and early warning method, which comprises the following steps: collecting multi-source heterogeneous data in real time through an edge computing unit deployed in an operation site; analyzing the video stream in real time by using a preset artificial intelligence visual analysis model, identifying personnel violation behaviors, equipment abnormal states and environmental risk factors, performing cross validation in combination with an electronic access control state and a work ticket permission state, and generating a primary alarm signal when it is detected that preset condition information is not matched; risk grade evaluation is carried out based on the primary alarm signal and environmental risk factors, and a dynamic early warning instruction is generated and synchronized to a station level platform; and the station control layer triggers video review of the associated area, a linkage access control system locks the dangerous area, early warning information is pushed to the target terminal, and an emergency processing plan is generated. And the real-time response speed and the active protection capability of a high-risk operation scene are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of safety monitoring and early warning technology, and in particular to a new energy station operation site safety monitoring and early warning method. Background Art

[0002] With the large-scale construction of new energy stations, safety management at their work sites faces challenges such as complex environments, high-risk operations, and the interplay of multiple trades. Traditional safety monitoring relies primarily on manual inspections and single-source sensor alerts. For example, these approaches rely on manually identifying violations through video surveillance footage or triggering threshold alarms using environmental sensors. Existing solutions employ motion detection and analysis to detect dynamic changes in footage, or infrared thermal imaging to monitor equipment temperature anomalies, providing basic risk warnings. However, video analysis, access control systems, work permit management, and environmental sensors operate independently, lacking cross-verification capabilities. For example, video can identify individuals without safety equipment but cannot correlate them with valid work permits or those entering unauthorized areas, leading to false or missed alerts. Cloud-based analysis models are constrained by network latency, making them difficult to analyze high-concurrency video streams in real time and lacking accuracy in complex scene recognition. Furthermore, existing methods lack effective support for scenarios such as trajectory prediction of drifting objects and automated verification of safety measures compliance, making them unable to meet the full-process control requirements for high-risk operations at new energy stations.

[0003] In view of this, a new energy station operation site safety monitoring and early warning method is proposed. Summary of the Invention

[0004] The present invention provides a new energy station operation site safety monitoring and early warning method for solving the problem that existing technologies are difficult to meet the full-process management and control needs of high-risk operations in new energy stations.

[0005] The present invention provides a new energy station operation site safety monitoring and early warning method, comprising: The edge computing unit deployed at the work site collects multi-source heterogeneous data in real time, including video streams, electronic access control status, work permit status, and environmental sensor data; The video stream is analyzed in real time using a preset artificial intelligence visual analysis model to identify personnel violations, equipment abnormalities, and environmental hazards. This is then cross-verified with the electronic access control status and the work permit status. A primary alarm signal is generated when a mismatch between the preset condition information is detected. Based on the primary alarm signal and environmental risk factors, a risk level assessment is performed, and a dynamic warning instruction is generated and synchronized to the station control layer platform; the station control layer platform performs the following operations: triggering a video review of the associated area, linking the access control system to lock the dangerous area, pushing warning information to the target terminal, and generating an emergency response plan; Among them, the edge computing unit and the station control layer platform are connected through a layered distributed architecture communication, and the artificial intelligence visual analysis model supports remote configuration.

[0006] Furthermore, the preset artificial intelligence visual analysis model is used to analyze the video stream in real time to identify personnel violations, equipment abnormalities and environmental risk factors, including: Performing frame processing on the video stream to extract a continuous video frame sequence; Tracking the movement trajectory of the operator in the continuous video sequence to generate a dynamic behavior vector; inputting the dynamic behavior vector into a preset violation behavior recognition sub-model to output a determination result of the operator's violation behavior; Extracting device feature regions from the continuous video frame sequence; extracting device structural contour features based on visible light images of the device feature regions; inputting infrared thermal imaging data of the device feature regions and the structural contour features into a preset abnormal state detection sub-model to output abnormal thermal distribution regions and safety device defect states; Detect fireworks features from the continuous video frame sequence and track the movement trajectory of drifting objects; input the environmental sensor data, the detected fireworks features and the tracked movement trajectory of drifting objects into a preset environmental risk analysis sub-model, and output the fireworks risk level and the risk position coordinates of the drifting objects.

[0007] Furthermore, the tracking of the operator's motion trajectory in the continuous video sequence to generate a dynamic behavior vector includes: Determine whether the ratio of the area of ​​the moving region in the pixel displacement field of adjacent video frames in the continuous video sequence is greater than a preset ratio threshold; if it is greater, extract the motion vector of the region to construct a trajectory coordinate set; if it is less than or equal to, mark it as a background interference region and skip processing the current frame; Determine whether the acceleration change rate of the trajectory coordinate set exceeds a preset action threshold; if so, generate a dynamic behavior vector; if not, continue tracking the next frame sequence.

[0008] Furthermore, the dynamic behavior vector is input into a preset violation behavior identification sub-model to output the determination result of the violation behavior of the personnel, including: Extracting the spatiotemporal feature sequence from the dynamic behavior vector, inputting the spatiotemporal feature sequence into the traffic violation behavior identification sub-model, and outputting the determination result corresponding to the traffic violation behavior type.

[0009] Furthermore, the step of inputting the infrared thermal imaging data of the characteristic area of ​​the equipment and the structural contour features into a preset abnormal state detection sub-model to output the abnormal thermal distribution area and the safety device defect state includes: Based on the infrared thermal imaging data, the temperature gradient distribution of the characteristic area of ​​the equipment is calculated, and the area exceeding the preset temperature threshold is identified as the area of ​​abnormal thermal distribution; The structural profile features are compared with a preset standard equipment structure template to detect the missing or deformed state of key safety components and output the defect status of the safety device.

[0010] Furthermore, the inputting of the environmental sensor data, the detected fireworks characteristics, and the tracked motion trajectory of the drifting object into a preset environmental risk analysis sub-model to output the fireworks risk level and the risk position coordinates of the drifting object includes: Determining whether the detected fireworks feature exceeds a preset fireworks threshold to obtain a first risk judgment result; If the first risk judgment result is yes, then fusing the smoke concentration value and temperature value in the environmental sensor data to calculate a comprehensive smoke and fire risk index; The fire and smoke comprehensive risk index is matched with a preset risk level threshold interval to map and obtain the fire and smoke risk level.

[0011] Furthermore, the step of inputting the environmental sensor data, the detected fireworks characteristics, and the tracked motion trajectory of the drifting object into a preset environmental risk analysis sub-model to output the fireworks risk level and the risk position coordinates of the drifting object further includes: Analyze the velocity vector and height attenuation rate of the drifting object's trajectory and predict the coordinates of the drifting object's falling point; Determine whether the coordinates of the falling point are located in a preset dangerous area, obtain a second risk judgment result, and output the risk position coordinates of the drifting object based on the second risk judgment result.

[0012] Furthermore, the electronic access control status is cross-verified with the work ticket permission status, and a primary alarm signal is generated when a mismatch between the preset condition information is detected, including: Extract the authorized work area code and safety measures list from the work permit status; Compare the authorized operation area code with the electronic access control trigger position code, and verify the consistency between the safety measures list and the safety equipment status obtained from video analysis; A primary alarm signal is generated when any of the following conditions are detected: The authorized operation area code does not match the electronic access control trigger position code; the mandatory equipment items in the safety measures list are missing from the safety equipment status obtained by video analysis.

[0013] Furthermore, verifying the consistency between the safety measures list and the safety equipment status obtained by video analysis includes: Obtain the worker's safety helmet wearing status, insulating gloves wearing status, and safety belt usage status; Compare the wearing status of the safety helmet, insulating gloves and safety belt with the mandatory equipment items in the safety measures list item by item; When it is detected that the status of any mandatory equipment item is not enabled, it is marked as safety equipment missing and a primary alarm signal is generated.

[0014] Furthermore, the risk level assessment is performed based on the primary warning signal and environmental risk factors, and dynamic warning instructions are generated and synchronized to the station control layer platform, including: According to a preset risk assessment matrix, the violation type is associated and mapped with the pyrotechnic risk level and the coordinates of the risk position of a drifting object, and the safety device defect status is associated and mapped with the pyrotechnic risk level and the coordinates of the risk position of a drifting object to generate a comprehensive risk score; Based on the preset risk range of the comprehensive risk score, a dynamic warning level is determined and a corresponding warning instruction package is generated; The warning instruction package is synchronized to the station control layer platform in real time through the message queue.

[0015] It can be seen from the above technical solutions that the present invention has the following advantages: This application uses edge computing units deployed at the work site to collect multi-source heterogeneous data including video streams, electronic access control status, work ticket permission status, and environmental sensor data in real time. It uses a preset remote configurable artificial intelligence visual analysis model to analyze the video stream in real time to identify personnel violations, equipment abnormalities, and environmental risk factors. It also generates a primary alarm signal when the preset conditions do not match by cross-verifying the electronic access control status and the work ticket permission status. It then generates a dynamic early warning instruction based on the risk level assessment of the signal and the environmental risk factors and synchronizes it to the station control layer platform. The station control layer platform triggers the video review of the associated area, links the access control system to lock the dangerous area, pushes early warning information to the target terminal, and generates an emergency response plan. Finally, it communicates with the station control layer platform through the layered distributed architecture of the edge computing unit. This solves the problem of false alarms and missed alarms caused by the fragmentation of multiple systems in the existing solution, and effectively improves the real-time response speed and active protection capabilities of high-risk work scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of an embodiment of a method for safety monitoring and early warning of a new energy station operation site in the present invention; Figure 2 This is a flow chart of another embodiment of a method for safety monitoring and early warning of a new energy station operation site in the present invention; Figure 3This is a flow chart of another embodiment of a method for safety monitoring and early warning of a new energy station operation site in the present invention; Figure 4 The figure is a flow chart of another embodiment of a new energy station operation site safety monitoring and early warning method in the present invention. DETAILED DESCRIPTION

[0017] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "corresponding to" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] Example 1 The implementation method in this embodiment can be implemented in the system, can be implemented in the server, and can also be implemented in the terminal, without specific limitation. The following will introduce the safety monitoring and early warning method of the new energy station operation site in this application from the perspective of system implementation. Figures 1 to 4 The method provided in the embodiment of the present application includes the following steps: S1. Use edge computing units deployed at the work site to collect multi-source heterogeneous data in real time. This data includes video streams, electronic access control status, work permit status, and environmental sensor data. This step uses edge computing units deployed at the new energy station operation site, such as industrial-grade AI edge gateways, to simultaneously collect four types of heterogeneous data using a multi-protocol interface. The video stream is transmitted via RTSP from explosion-proof cameras located in high-risk areas. It includes dual-mode data streams of 1080P visible light video and infrared thermal imaging, covering personnel operation points, equipment operating areas, and the surrounding environment. Electronic access control status is obtained in real time via Modbus-RTU from access controller data, including the trigger location code corresponding to access card swipe records, door open / close status, and abnormal intrusion alarm signals. Work permit status is synchronized from the station work order management system via the OPC-UA interface and includes the authorized operator ID, permitted work area code, validity period, and a list of mandatory safety measures, such as the wearing of insulating gloves / hardhats. Environmental sensor data is aggregated via the LoRaWAN wireless network, including real-time readings from temperature and humidity sensors, anemometers, smoke detectors, and gas concentration sensors, providing comprehensive monitoring of the physical environment.

[0019] The above multi-source data are aligned to millisecond timestamps through the built-in time synchronization module of the edge computing unit, and uniformly encapsulated into JSON format message streams, providing an accurate data foundation of spatiotemporal correlation for subsequent artificial intelligence analysis.

[0020] S2. Utilizes a pre-set AI visual analysis model to analyze video streams in real time, identifying violations, abnormal equipment conditions, and environmental hazards. This model then cross-validates electronic access control status with work permit status, generating a primary alarm signal when a mismatch between pre-set conditions is detected. In this embodiment, a preset artificial intelligence visual analysis model is used to analyze the video stream in real time to identify personnel violations, equipment abnormalities, and environmental hazards. This is achieved through the following steps: S211 performs frame processing on the video stream to extract a continuous video frame sequence; The video stream is framed and processed using the edge computing unit's built-in hardware decoder, which parses the encoded video stream in real time to extract a continuous video frame sequence at 30fps. The mixed video stream is channel-splitting to separate visible light video frames from infrared thermal imaging frames. The frame resolution is dynamically adjusted based on preset area weights (e.g., personnel operation area > equipment area > environmental background area), maintaining full 1080P resolution in high-risk areas and downsampling to 720P in non-critical areas to reduce the computational load. High-precision timestamps and spatial coordinate metadata are injected into each frame to generate a continuous frame sequence with temporal and spatial tags.

[0021] S212 tracking the movement trajectory of the operator in the continuous video sequence, generating a dynamic behavior vector; and the dynamic behavior vector is input into the preset violation behavior recognition sub-model, outputting the determination result of the violation behavior of the personnel; Among them, tracking the movement trajectory of the operator in the continuous video sequence and generating the dynamic behavior vector includes: 1. Determine whether the ratio of the moving area in the pixel displacement field of adjacent video frames in a continuous video sequence is greater than a preset ratio threshold; if it is, extract the motion vector of the area to construct a trajectory coordinate set; if it is less than or equal to, mark it as a background interference area and skip processing the current frame; 2. Determine whether the acceleration change rate of the trajectory coordinate set exceeds the preset action threshold; if so, generate a dynamic behavior vector; if not, continue tracking the next frame sequence.

[0022] First, the proportion of the moving area is calculated based on the pixel displacement field of adjacent video frames. For example, the motion vector field is extracted using the OpenCV optical flow method. When this proportion exceeds a preset ratio threshold of 20%, determined by statistically analyzing thousands of wind farm scene samples, it can filter out background interference such as blade swaying less than 15cm². The motion vector of this area is extracted to construct a trajectory coordinate set. Next, the acceleration change rate of the trajectory coordinate set is calculated. If it exceeds a preset motion threshold of 0.5g, which is set based on a biomechanical model of human motion and corresponds to high-risk actions such as climbing and running, a dynamic behavior vector containing position coordinates, velocity vectors, and joint angles is generated.

[0023] In this embodiment, the dynamic behavior vector is input into a preset violation behavior recognition sub-model, and the judgment result of the personnel violation behavior is output, including: extracting the spatiotemporal feature sequence in the dynamic behavior vector, inputting the spatiotemporal feature sequence into the violation behavior recognition sub-model, and outputting the judgment result of the corresponding violation behavior type.

[0024] Specifically, the temporal position data in the vector is converted into a spatiotemporal cube, and the 3D gradient histogram features are extracted. The judgment results are classified and output through a 3D-CNN model (input layer 128×128×16) trained with 100,000 violation samples. For example, when the joint angles detected in five consecutive frames meet the characteristics of "working in the air without a seat belt", a "high-altitude violation" alarm is triggered.

[0025] S213. Extracting device feature areas from a sequence of continuous video frames; extracting device structural contour features based on visible light images of the device feature areas; inputting infrared thermal imaging data and structural contour features of the device feature areas into a preset abnormal state detection sub-model, and outputting abnormal thermal distribution areas and safety device defect status; Equipment anomaly detection achieves accurate diagnosis by fusing dual-mode data of visible light and infrared thermal imaging. Its principle is as follows: based on the preset equipment detection frame of the YOLOv5 model, such as photovoltaic inverters, wind turbine gearboxes and other key components, the ROI area is cropped from the continuous video frame sequence, and the sub-pixel level structural contour features are extracted through the visible light image using the Canny edge detection algorithm; the infrared thermal imaging data (temperature matrix) and structural contour features (vector polygons) under the same spatiotemporal reference are input into the abnormal state detection sub-model, and the feature-level fusion technology is used to establish the temperature-structure mapping relationship to achieve collaborative diagnosis of the equipment's thermodynamic performance and mechanical integrity.

[0026] The specific implementation is as follows: 1. Based on infrared thermal imaging data, calculate the temperature gradient distribution of the device's characteristic areas and identify areas exceeding the preset temperature threshold as areas of thermal anomaly. 2. Compare the structural contour features with the preset standard equipment structure template, detect the missing or deformed state of key safety components, and output the defect status of the safety device.

[0027] Based on infrared thermal imaging data, the temperature gradient distribution of characteristic areas of the equipment is calculated. When the detected temperature in a local area exceeds a preset dynamic threshold, such as the electrical equipment surface temperature threshold of 65°C ± the ambient temperature compensation value, and the gradient change rate is greater than 3°C / cm, it is marked as an abnormal thermal distribution area. The extracted structural contour features are aligned with a preset standard equipment structure template (generated from a laser scanning point cloud) through ICP point cloud registration. Deviations are detected by calculating the Hausdorff distance. If the point cloud match of a key component (such as a grounding terminal or safety lock) is less than 90%, it is determined to be missing. If the curvature change rate of a deformed area exceeds a preset safety threshold (such as a bolt displacement tolerance of ±2mm), it is determined to be deformed. The output result is encoded as "component ID + defect type + deviation amount."

[0028] S214. Detect fireworks features from a continuous video frame sequence and track the motion trajectory of drifting objects; input the environmental sensor data, the detected fireworks features, and the tracked motion trajectory of drifting objects into a preset environmental risk analysis sub-model, and output the fireworks risk level and the risk position coordinates of the drifting objects.

[0029] By integrating video dynamic features with physical sensor data, we can solve the problems of traditional solutions' high misjudgment rate of early fireworks and inaccurate prediction of the landing points of drifting objects, and achieve accurate quantification and spatial positioning of environmental risks.

[0030] 1. Determine whether the detected fireworks characteristics exceed a preset fireworks threshold, and obtain a first risk determination result; 2. If the first risk judgment result is yes, the smoke concentration and temperature values ​​in the environmental sensor data are integrated to calculate the comprehensive fire and smoke risk index; 3. Match the comprehensive fireworks risk index with the preset risk level threshold range to map the fireworks risk level.

[0031] Specifically, the YOLOv7-Fire model detects the proportion of flame pixels and flickering frequency in consecutive video frames. When the flame pixel proportion is greater than 0.3% and the flickering frequency is greater than 5Hz (the preset fireworks threshold), the first risk judgment result is triggered as positive. If the initial screening is positive, the real-time data of the environmental sensors is integrated, including the smoke concentration value, with a weighting coefficient α=0.6, and the temperature value, with a weighting coefficient β=0.4. The comprehensive fireworks risk index is dynamically calculated according to the formula = α×(smoke concentration / lower explosion limit)×100+β×(temperature-ambient temperature) / 50. Risk level mapping: The index is matched to the preset threshold range: [0,30) → low risk (Level 1); [30,70) → medium risk (Level 2); [70,100] → high risk (Level 3). Finally, the corresponding level and confidence level are output, such as "Level 3, 92%".

[0032] This step predicts the fall point using a kinematic model: 1. Analyze the velocity vector and height attenuation rate of the drifting object's trajectory and predict the coordinates of the drifting object's falling point; 2. Determine whether the coordinates of the falling point are located in a preset dangerous area, obtain a second risk judgment result, and output the risk position coordinates of the drifting object based on the second risk judgment result.

[0033] The pixel displacement of the drifting object in consecutive frames is extracted and converted into the true velocity vector, horizontal velocity Vx, Vy, and vertical velocity Vz, based on the camera's internal parameters. The altitude attenuation rate is calculated based on the aerodynamic formula: Δh / Δt=Vz-(0.5×ρ×Cd×A×Vz²) / m, where ρ is the air density, Cd is the drag coefficient, A is the cross-sectional area, and m is the mass.

[0034] Taking the current frame position as the origin, the system iteratively solves t (landing time) according to the motion equation: fall point coordinates = (Vx×t, Vy×t, h0 + Vz×t -0.5gt²), and outputs the geodetic coordinates. When the fall point falls into a preset danger zone (such as a coordinate polygon fence, which is preset as the equipment area or personnel operation area), the second risk judgment result is marked as high risk, and the risk location coordinates and estimated arrival time are output, such as "(115.62E, 38.92N), arrival in 28s."

[0035] In this embodiment, generating a primary alarm signal includes the following steps: S221. Extract the authorized work area code and safety measures list from the work permit status; This step accesses the site work order management system in real time through the OPC-UA protocol interface of the edge computing unit, and uses regular expression matching and JSON parsing technology to extract structured data from the work ticket permission status.

[0036] The authorized operation area code is the coordinates of the geofence polygon vertices in the parsed work order, such as GEO-2023: [115.62E, 38.92N; 115.63E, 38.91N...]; the safety measures list is obtained by extracting the mandatory equipment item code list, such as SAF_REQ: ["HELMET", "INSU_GLOVES", "SAFETY_HARNESS"] and the validity period timestamp (VALID_UNTIL: 2023-12-31T23:59:59Z), and at the same time associating the operator ID with the biometric database for video identity verification.

[0037] S222. Compare the authorized operating area code with the electronic access control trigger position code and verify the consistency of the safety measures list with the safety equipment status obtained by video analysis; 1. Obtain the operator's safety helmet wearing status, insulating gloves wearing status and safety belt use status; 2. Compare the wearing status of safety helmets, insulating gloves and safety belts with the mandatory equipment items in the safety measures list item by item; 3. When it is detected that the status of any mandatory equipment item is not enabled, it is marked as safety equipment missing and a primary alarm signal is generated.

[0038] Based on the output of the violation identification sub-model in S22, the safety helmet wearing status, insulating glove wearing status, and safety belt usage status associated with the operator's biometric features are extracted. A three-tuple check matrix [work order equipment item, video detection status, confidence level] is constructed. For example, when the safety measures list requires "INSU_GLOVES" but the video analysis output shows the glove status as "not worn" with a confidence level greater than 90%, the item is marked as missing. If any mandatory equipment item is missing, such as the safety belt status = not enabled during high-altitude work, a primary alarm signal containing the operator ID and the missing item code is immediately generated, such as ALARM_SAF: P-0032, MISSING_HARNESS.

[0039] S223. A primary alarm signal is generated when any of the following situations is detected: the authorized operation area code does not match the electronic access control trigger position code; there is a missing between the mandatory equipment item in the safety measures list and the safety equipment status obtained by video analysis.

[0040] This step performs a logical "OR" operation based on the verification results of S221-S222 to trigger an alarm: Scenario 1: Area overreach detection: When the spatial relationship between the electronic access control trigger location code (such as GATE-7) and the work ticket authorization area code (such as GEO-2023) does not match (calculated by the GIS engine, if the access control coordinate point is >5 meters outside the authorization polygon), an alarm signal ALARM_AREA: P-0032, GATE-7 is generated; Scenario 2: Equipment missing detection: When S222 outputs the safety equipment missing flag, an alarm signal ALARM_SAF: [personnel ID] + [missing item code] is generated; Alarm signal encapsulation: The above alarm and spatiotemporal metadata (timestamp, camera coordinates) are encapsulated into a JSON message (example: {"alarm_type": "AREA", "worker_id": "P-0032", "location":[115.61E,38.93N], "timestamp":"2023-05-17T08:32:45.123Z"}), and publish to the edge message bus through the MQTT protocol.

[0041] S3. Based on the primary alarm signal and environmental hazard factors, a risk level assessment is conducted, and dynamic warning instructions are generated and synchronized to the station control layer platform. The station control layer platform performs the following operations: triggering video review of the associated area, linking the access control system to lock the dangerous area, pushing warning information to the target terminal, and generating an emergency response plan. Among them, the edge computing unit and the station control layer platform are connected through a layered distributed architecture, and the artificial intelligence visual analysis model supports remote configuration.

[0042] This step uses the edge computing unit's risk assessment engine to process primary alarm signals and environmental risk factors (fireworks risk level and the coordinates of the risky location of drifting objects from S24) in real time. It first calls upon a pre-set three-dimensional risk assessment matrix: Dimension 1: Violation type weight; Dimension 2: Fireworks risk level coefficient; and Dimension 3: The spatial relationship between the drifting object and the hazardous area. For example, if a "high-altitude violation" (weight 0.8) is detected and the drifting object falls within the equipment area (coefficient 1.2), a score is dynamically calculated using the formula: Comprehensive Risk Score = Behavior Weight × (Fireworks Level Coefficient + Drifting Object Location Coefficient). Preset risk intervals are then mapped based on the score: 0-40 points → Blue Alert (Level 4), 41-70 points → Yellow Alert (Level 3), 71-90 points → Orange Alert (Level 2), and 91-100 points → Red Alert (Level 1). An alert command packet is generated, containing the alert level, risk source coordinates, and recommended actions. This is then synchronized to the station control platform in real time via the layered distributed architecture's MQTT message queue, triggering the platform to execute four types of closed-loop operations: 1. Video review: retrieve the real-time video stream from cameras associated with the hazardous area (e.g., within a 50-meter radius) and start the manual confirmation interface; 2. Access control linkage: Send a lock command (such as LOCK_ZONE: GEO-2023) to the access control controller to physically isolate the dangerous area; 3. Information push: Match target terminals based on warning levels (red warning → app pop-up window + SMS for all personnel; yellow warning → team leader); 4. Plan generation: Call matching emergency response templates (such as "drifting object impact plan") from the knowledge base, automatically fill in risk coordinates and wind speed data to generate a disposal flow chart.

[0043] S31. According to the preset risk assessment matrix, the violation type is associated with the pyrotechnic risk level and the risk position coordinates of the drifting object. At the same time, the safety device defect status is associated with the pyrotechnic risk level and the risk position coordinates of the drifting object to generate a comprehensive risk score. S32. Based on the preset risk range of the comprehensive risk score, determine the dynamic warning level and generate the corresponding warning instruction package; S33. Synchronize the warning instruction package to the station control layer platform in real time through the message queue.

[0044] Specifically, the violation type, fireworks risk level, and the coordinates of the risk position of drifting objects are input into the risk assessment matrix, and the weight coefficients are associated through the matrix mapping table. The scalar value of 0-100 points is output according to the score = behavior weight × MAX (fireworks coefficient, drifting object coefficient). When the comprehensive risk score falls into the preset range (such as 72 points → orange warning level 2), the instruction package structure is automatically generated, including the warning level, risk source type list, geographic fence vertex array and recommended disposal action code. The instruction package is transmitted to the station control layer platform via the message queue partition. After the platform parses it, it immediately executes video review (starting the specified camera PTZ preset position), access control locking (issuing BACnet protocol commands), information push (interconnecting with DingTalk / Enterprise WeChat API) and plan generation (activating the plan engine Jinja2 template rendering).

[0045] Ultimately, the false alarm rate will be reduced while the response speed to major risks will be improved. The automatic generation rate of emergency response plans will reach 100%, reducing delays in manual intervention.

[0046] It is understandable that those skilled in the art can, under the guidance of the above embodiments, combine various implementation methods in the above embodiments to obtain technical solutions of multiple implementation methods.

[0047] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A new energy station operation site safety monitoring and early warning method, characterized in that: include: The edge computing unit deployed at the work site collects multi-source heterogeneous data in real time, including video streams, electronic access control status, work permit status, and environmental sensor data; The video stream is analyzed in real time using a preset artificial intelligence visual analysis model to identify personnel violations, equipment abnormalities, and environmental hazards. This is then cross-verified with the electronic access control status and the work permit status. A primary alarm signal is generated when a mismatch between the preset condition information is detected. Based on the primary alarm signal and environmental risk factors, a risk level assessment is performed, and a dynamic warning instruction is generated and synchronized to the station control layer platform; the station control layer platform performs the following operations: triggering a video review of the associated area, linking the access control system to lock the dangerous area, pushing warning information to the target terminal, and generating an emergency response plan; Among them, the edge computing unit and the station control layer platform are connected through a layered distributed architecture communication, and the artificial intelligence visual analysis model supports remote configuration.

2. The method for safety monitoring and early warning of new energy station operation site according to claim 1 is characterized in that: The use of a preset artificial intelligence visual analysis model to analyze the video stream in real time to identify personnel violations, equipment abnormalities, and environmental risk factors includes: Performing frame processing on the video stream to extract a continuous video frame sequence; Tracking the movement trajectory of the operator in the continuous video sequence to generate a dynamic behavior vector; inputting the dynamic behavior vector into a preset violation behavior recognition sub-model to output a determination result of the operator's violation behavior; Extracting device feature regions from the continuous video frame sequence; extracting device structural contour features based on visible light images of the device feature regions; inputting infrared thermal imaging data of the device feature regions and the structural contour features into a preset abnormal state detection sub-model to output abnormal thermal distribution regions and safety device defect states; Detect fireworks features from the continuous video frame sequence and track the movement trajectory of drifting objects; input the environmental sensor data, the detected fireworks features and the tracked movement trajectory of drifting objects into a preset environmental risk analysis sub-model, and output the fireworks risk level and the risk position coordinates of the drifting objects.

3. The method for safety monitoring and early warning of new energy station operation site according to claim 2 is characterized in that: Tracking the movement trajectory of the operator in the continuous video sequence to generate a dynamic behavior vector includes: Determine whether the ratio of the area of ​​the moving region in the pixel displacement field of adjacent video frames in the continuous video sequence is greater than a preset ratio threshold; if it is greater, extract the motion vector of the region to construct a trajectory coordinate set; if it is less than or equal to, mark it as a background interference region and skip processing the current frame; Determine whether the acceleration change rate of the trajectory coordinate set exceeds a preset action threshold; if so, generate a dynamic behavior vector; if not, continue tracking the next frame sequence.

4. The method for safety monitoring and early warning of new energy station operation site according to claim 3 is characterized in that: The dynamic behavior vector is input into a preset violation behavior identification sub-model, and the determination result of the violation behavior of the personnel is output, including: Extracting the spatiotemporal feature sequence from the dynamic behavior vector, inputting the spatiotemporal feature sequence into the traffic violation behavior identification sub-model, and outputting the determination result corresponding to the traffic violation behavior type.

5. The method for safety monitoring and early warning of new energy station operation site according to claim 2 is characterized in that: The step of inputting the infrared thermal imaging data of the characteristic area of ​​the equipment and the structural contour features into a preset abnormal state detection sub-model and outputting the abnormal thermal distribution area and the safety device defect state includes: Based on the infrared thermal imaging data, the temperature gradient distribution of the characteristic area of ​​the equipment is calculated, and the area exceeding the preset temperature threshold is identified as the area of ​​abnormal thermal distribution; The structural profile features are compared with a preset standard equipment structure template to detect the missing or deformed state of key safety components and output the defect status of the safety device.

6. The method for safety monitoring and early warning of new energy station operation site according to claim 2 is characterized in that: The step of inputting the environmental sensor data, the detected fireworks characteristics, and the tracked motion trajectory of the drifting object into a preset environmental risk analysis sub-model and outputting the fireworks risk level and the risk position coordinates of the drifting object includes: Determining whether the detected fireworks feature exceeds a preset fireworks threshold to obtain a first risk judgment result; If the first risk judgment result is yes, then fusing the smoke concentration value and temperature value in the environmental sensor data to calculate a comprehensive smoke and fire risk index; The fire and smoke comprehensive risk index is matched with a preset risk level threshold interval to map and obtain the fire and smoke risk level.

7. The method for safety monitoring and early warning of new energy station operation site according to claim 2 is characterized in that: The step of inputting the environmental sensor data, the detected fireworks characteristics, and the tracked motion trajectory of the drifting object into a preset environmental risk analysis sub-model to output the fireworks risk level and the risk position coordinates of the drifting object further includes: Analyze the velocity vector and height attenuation rate of the drifting object's trajectory and predict the coordinates of the drifting object's falling point; Determine whether the coordinates of the falling point are located in a preset dangerous area, obtain a second risk judgment result, and output the risk position coordinates of the drifting object based on the second risk judgment result.

8. The method for safety monitoring and early warning of new energy station operation site according to claim 1 is characterized in that: The cross-verification is performed in combination with the electronic access control status and the work ticket permission status, and a primary alarm signal is generated when a mismatch between the preset condition information is detected, including: Extract the authorized work area code and safety measures list from the work permit status; Compare the authorized operation area code with the electronic access control trigger position code, and verify the consistency between the safety measures list and the safety equipment status obtained from video analysis; A primary alarm signal is generated when any of the following conditions are detected: The authorized operation area code does not match the electronic access control trigger position code; the mandatory equipment items in the safety measures list are missing from the safety equipment status obtained by video analysis.

9. The method for safety monitoring and early warning of the new energy station operation site according to claim 8 is characterized in that: Verifying the consistency between the safety measures list and the safety equipment status obtained by video analysis includes: Obtain the worker's safety helmet wearing status, insulating gloves wearing status, and safety belt usage status; Compare the wearing status of the safety helmet, insulating gloves and safety belt with the mandatory equipment items in the safety measures list item by item; When it is detected that the status of any mandatory equipment item is not enabled, it is marked as safety equipment missing and a primary alarm signal is generated.

10. The method for safety monitoring and early warning of a new energy station operation site according to claim 1, characterized in that: The risk level assessment based on the primary warning signal and environmental risk factors, generating dynamic warning instructions and synchronizing them to the station control layer platform includes: According to a preset risk assessment matrix, the violation type is associated and mapped with the pyrotechnic risk level and the coordinates of the risk position of a drifting object, and the safety device defect status is associated and mapped with the pyrotechnic risk level and the coordinates of the risk position of a drifting object to generate a comprehensive risk score; Based on the preset risk range of the comprehensive risk score, a dynamic warning level is determined and a corresponding warning instruction package is generated; The warning instruction package is synchronized to the station control layer platform in real time through the message queue.

Images