Power plant dynamic three-dimensional model construction system based on digitization

Abstract

The invention discloses a power plant dynamic three-dimensional model construction system based on digitization, which relates to the technical field of power plant safety management and comprises a data acquisition module, a deployment optimization module, a moving track positioning module and an intelligent path planning module. According to the method, a laser radar point cloud and 4K image space alignment technology is adopted, geometric deviation is eliminated, a centimeter-level three-dimensional model is constructed, various IoT sensors such as temperature and humidity sensors and vibration sensors are integrated, a thermodynamic diagram and an oscillogram are dynamically labeled, multi-dimensional environment perception is achieved, distributed edge nodes are deployed, parallel processing of 10 paths of video streams is supported, and dynamic resource allocation is combined. According to the invention, the method is simple in operation, ensures stable visualization of 30 frames per second, controls the reasoning power consumption within 15 watts through a lightweight model and a low power consumption protocol, achieves the real-time path adjustment based on an A star algorithm and a dynamic blocking grid updating mechanism, combines fire spreading prediction and an obstacle expansion boundary, achieves the linkage of a dynamic indicating lamp and a three-level alarm system, and remarkably improves the emergency response speed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power plant safety management, and particularly relates to a digital-based power plant dynamic three-dimensional model construction system. BACKGROUND

[0002] A power plant is a factory that converts other forms of energy, such as fossil fuels, water energy, wind energy, nuclear energy, etc., into electric energy, and is the core link of the power system. Its main function is to convert primary energy into secondary energy through energy conversion equipment, and then transmit it to the user end through the power transmission network to meet the electricity demand of various types of users such as industry, commerce, and residents.

[0003] The disclosed patent: safety warning method and device for industrial plant and safety warning system (publication number: CN115410354B), comprising: acquiring video information of the industrial plant; acquiring a three-dimensional virtual model of the industrial plant; at least according to the video information, detecting the safety of the industrial plant, and displaying the virtual risk position and risk video screenshot on the display interface if a risk is detected, in the scheme, by constructing a three-dimensional virtual model of the entire industrial plant, the three-dimensional virtual scene of the entire industrial plant can be displayed on the display interface, and the video of all scenes in the entire industrial plant can be obtained to obtain the video information, so that the safety of the industrial plant can be detected at least according to the video information, and if a risk of the industrial plant is detected, the corresponding virtual risk position and risk video screenshot can be directly displayed on the display interface, so that the industrial plant can be comprehensively analyzed to ensure the safety of the industrial plant.

[0004] The above patent has significant shortcomings in data fusion, real-time response, dynamic path planning, hardware compatibility, and behavior recognition accuracy. Its technical solution is limited to single video analysis and static model display, and lacks deep adaptation to the dynamic complexity of industrial scenes and multi-dimensional safety requirements, which restricts the performance of the system in real-time, reliability, and expandability. SUMMARY

[0005] The present application relates to the technical field of power plant safety management, and particularly relates to a digital-based power plant dynamic three-dimensional model construction system.

[0006] In order to achieve the above object, the application adopts the following technical scheme: a power plant dynamic three-dimensional model construction system based on digitization, comprising a data acquisition module, a deployment optimization module, a mobile trajectory positioning module and an intelligent path planning module, the data acquisition module constructs a basic three-dimensional model through multi-source data fusion and integrates real-time environmental parameters, the deployment optimization module processes multiple camera pictures and sensor data streams in parallel through distributed edge computing nodes, the mobile trajectory positioning module is used for real-time tracking of dynamic trajectories of personnel and vehicles in the power plant and changes of obstacles, and the intelligent path planning module is used for dynamically generating an escape path and realizing emergency response.

[0007] As a further description of the above technical scheme:

[0008] The data acquisition module uses a UAV to carry a laser radar to perform three-dimensional point cloud scanning on the whole power plant, adopts an iterative closest point algorithm to realize multi-view point cloud splicing, and performs lens distortion correction and color balance on 4K camera images through an OpenCV library.

[0009] As a further description of the above technical scheme:

[0010] The data buffer of the deployment optimization module adopts a time window mechanism, and the latest 5 minutes of data are locally retained to deal with network interruption.

[0011] As a further description of the above technical scheme:

[0012] The trajectory data output by the mobile trajectory positioning module includes a timestamp, a coordinate sequence, and dynamic obstacle types, sizes and danger levels.

[0013] As a further description of the above technical scheme:

[0014] The intelligent path planning module adopts an A-star algorithm, combines dynamic obstacle information and an obstacle inflation mechanism, and expands a safety boundary for detected static obstacles.

[0015] As a further description of the above technical scheme:

[0016] The three-dimensional model and sensor data of the system are respectively exported in FBX / GLTF formats and MQTT protocols, static structure is updated in full at a fixed time, and dynamic parameters are refreshed every second.

[0017] As a further description of the above technical scheme:

[0018] The mobile trajectory positioning module analyzes camera pictures based on a semantic segmentation algorithm, identifies road boundaries and moving vehicles, and detects temporary piles in combination with an optical flow method.

[0019] As a further description of the above technical scheme:

[0020] The system operation process is as follows:

[0021] S1 three-dimensional scanning and image acquisition

[0022] The unmanned aerial vehicle carries a laser radar to perform centimeter-level resolution three-dimensional point cloud scanning, covering power plant buildings, pipelines and equipment;

[0023] Install 4K cameras in various areas of the power plant to shoot high-definition images of the entire area, capturing vehicle, object, road and personnel information in the picture;

[0024] S2 environmental parameter acquisition

[0025] Deploy IoT sensor networks at key locations in the power plant to collect real-time temperature, humidity, smoke concentration and vibration intensity values;

[0026] S3 multi-source data fusion

[0027] Laser radar point cloud data is spliced by iterative closest point algorithm, denoised and registered, 4K images are corrected for lens distortion and color balance by OpenCV library, point cloud, image and original CAD drawings are fused, and geometric deviation is eliminated by spatial alignment and coordinate system calibration;

[0028] S4 three-dimensional model construction

[0029] Based on Unity3D engine, the basic three-dimensional model of the power plant is constructed, the triangular facets are optimized and the texture is mapped, forming a hierarchical structure containing buildings, equipment and pipelines;

[0030] S5 edge computing node deployment

[0031] Distributed edge nodes process 10 camera pictures and multiple sensor data streams in parallel, and dynamically adjust GPU computing power;

[0032] S6 lightweight model and communication optimization

[0033] YOLOv5s deep learning model is used, and TensorRT / OpenVINO is used to optimize inference speed, static data is transmitted through LoRa protocol, video stream is compressed to below 4Mbps through H.265 encoding, and low delay transmission is combined with 5G slice network;

[0034] S7 hardware management and energy efficiency optimization

[0035] Industrial-grade SHT35 temperature and humidity sensor is used, which is suitable for multiple vendors' equipment, interrupt low priority tasks, reduce CPU utilization, when hardware temperature exceeds 60℃, CPU frequency from S4GHz to 1.8GHz;

[0036] S8 personnel / vehicle positioning

[0037] The Beidou / GPS dual-mode badge realizes centimeter-level positioning, and the 4K camera combines the YOLOv8 model to perform identity recognition and behavior analysis.

[0038] S9 trajectory optimization and obstacle detection

[0039] The Kalman filtering algorithm fuses the badge positioning and the camera direction data, the semantic segmentation algorithm identifies the road boundary and the moving vehicle, the optical flow method detects the temporary stacking object, the extended obstacle boundary is expanded by 20 centimeters as a safety distance, and the obstacle map is updated every 2 seconds;

[0040] S10 alarm and data transmission

[0041] Triggering a three-level yellow / orange / red alarm, and preferentially pushing to the management terminal through 5G, and outputting trajectory data through the MQTT protocol;

[0042] S11 fire prediction and obstruction modeling

[0043] Based on the thermodynamic / fluid mechanics model, the fire spread trend within 30 seconds is predicted, a three-dimensional thermal radiation field distribution map is generated, the dynamic obstruction grid is updated every 2 seconds, and is mapped to the path planning algorithm;

[0044] S12 path planning and dynamic adjustment

[0045] The improved A algorithm combines the obstacle inflation mechanism to plan the escape path, when the main path fails, the standby path is switched within 1 second, the instructions are issued to the dynamic indicator light through 5G / LoRa, the green light identifies the safe passage, and the red light marks the dangerous area;

[0046] S13 emergency linkage and visualization

[0047] The path planning result is synchronized to the three-dimensional interface in the FBX / GLTF format, dynamic rendering and API calling are supported, the alarm signal is preferentially pushed through 5G, and the GPU computing power of the edge node is dynamically allocated.

[0048] The present application has the following beneficial effects:

[0049] 1、In the present application, firstly, the spatial alignment technology of laser radar point cloud and 4K image is adopted, geometric deviation is eliminated, a centimeter-level three-dimensional model is constructed, multiple IoT sensors such as temperature and humidity and vibration are integrated, a thermal map and a waveform chart are dynamically labeled, multi-dimensional environment perception is realized, distributed edge nodes are deployed, 10 video streams are supported for parallel processing, combined with dynamic resource allocation, stable visualization of 30 frames / second is ensured, and through a lightweight model and a low-power protocol, the inference power consumption is controlled to be less than 15 watts.

[0050] 2、In the application, based on A-star algorithm and dynamic obstacle grid update mechanism, combined with fire spread prediction and obstacle expansion boundary, real-time path adjustment is realized, linkage dynamic indicator light and three-level alarm system, end-to-end delay <500ms, significantly improve the emergency response speed, Beidou / GPS dual-mode badge + inertial navigation provides centimeter-level positioning, Kalman filter eliminates noise, combined with YOLOv8 behavior recognition, dynamically mark high-risk behavior, enhance personnel safety management and control, support remote firmware upgrade and customization driver, reduce deployment complexity, dynamic temperature monitoring prolongs the service life of the equipment, avoids the risk of hardware overheating. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 It is a side view of the application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0053] Referring to Figure 1 , the application provides an embodiment: a digital-based power plant dynamic three-dimensional model construction system, comprising a data acquisition module, a deployment optimization module, a mobile trajectory positioning module and an intelligent path planning module, the data acquisition module constructs a basic three-dimensional model through multi-source data fusion and integrates real-time environmental parameters, the deployment optimization module processes multiple camera pictures and sensor data streams in parallel through distributed edge computing nodes, the mobile trajectory positioning module is used for real-time tracking of dynamic trajectories of personnel and vehicles in the power plant and changes of obstacles, and the intelligent path planning module is used for dynamically generating an escape path and realizing emergency response.

[0054] The data acquisition module uses a UAV carrying a laser radar to perform three-dimensional point cloud scanning on the entire power plant, multi-view point cloud splicing is achieved using an iterative closest point algorithm, 4K camera images are corrected for lens distortion and color balance using an OpenCV library, the data cache of the deployment optimization module uses a time window mechanism, the most recent time data is retained locally to deal with network interruptions, the trajectory data output by the mobile trajectory positioning module includes a timestamp, coordinate sequence, and dynamic obstacle type, size, and danger level, the intelligent path planning module uses an A-star algorithm, combined with dynamic obstacle information and an obstacle inflation mechanism, for detected static obstacles, an expanded safety boundary, the three-dimensional model of the system and sensor data are exported in FBX / GLTF format and MQTT protocol respectively, static structures are updated in full at regular intervals, dynamic parameters are refreshed every second, the mobile trajectory positioning module analyzes camera images based on a semantic segmentation algorithm, identifies road boundaries and moving vehicles, and detects temporary piles using an optical flow method.

[0055] The system operation process is as follows:

[0056] S1 Three-dimensional scanning and image acquisition

[0057] The UAV carries a laser radar to perform centimeter-level resolution three-dimensional point cloud scanning, covering power plant buildings, pipelines, and equipment.

[0058] 4K cameras are installed in various areas of the power plant to capture high-definition images of the entire area, capturing information about vehicles, objects, roads, and personnel in the images.

[0059] S2 Environment parameter acquisition

[0060] IoT sensor networks are deployed at key locations in the power plant to collect real-time temperature and humidity values, smoke concentration values, and vibration intensity.

[0061] S3 Multi-source data fusion

[0062] Laser radar point cloud data is spliced, denoised, and registered using an iterative closest point algorithm, 4K images are corrected for lens distortion and color balance using an OpenCV library, and point cloud, image, and original CAD drawings are fused to eliminate geometric deviations through spatial alignment and coordinate system calibration.

[0063] S4 Three-dimensional model construction

[0064] A power plant basic three-dimensional model is constructed based on the Unity3D engine, triangle facets are optimized and textures are mapped to form a hierarchical structure containing buildings, equipment, and pipelines.

[0065] S5 Edge computing node deployment

[0066] Distributed edge nodes process 10 camera images and multiple sensor data streams in parallel, dynamically adjusting GPU computing power.

[0067] S6 Lightweight Model and Communication Optimization

[0068] Using YOLOv5s deep learning model, optimizing inference speed through TensorRT / OpenVINO, transmitting static data through LoRa protocol, compressing video stream to below 4Mbps through H.265 encoding, combining with 5G slice network low-latency transmission;

[0069] S7 Hardware Management and Energy Efficiency Optimization

[0070] Using industrial-grade SHT35 temperature and humidity sensor, adapting to multiple vendors' devices, interrupting low-priority tasks, reducing CPU utilization, CPU frequency from S4GHz to 1.8GHz when hardware temperature exceeds 60℃;

[0071] S8 Personnel / Vehicle Positioning

[0072] Beidou / GPS dual-mode badge realizes centimeter-level positioning, 4K camera combined with YOLOv8 model for identity recognition and behavior analysis;

[0073] S9 Trajectory Optimization and Obstacle Detection

[0074] Kalman filter algorithm fuses badge positioning and camera direction data, semantic segmentation algorithm identifies road boundaries and moving vehicles, optical flow method detects temporary piles, expands obstacle boundary by 20 centimeters as safety distance, updates obstacle map every 2 seconds;

[0075] S10 Alarm and Data Transmission

[0076] Triggering yellow / orange / red three-level alarms, pushing to management terminal through 5G, outputting trajectory data through MQTT protocol;

[0077] S11 Fire Prediction and Obstruction Modeling

[0078] Based on thermodynamics / fluid mechanics model to predict fire spread trend within 30 seconds, generate three-dimensional thermal radiation field distribution map, update dynamic obstruction grid every 2 seconds, map to path planning algorithm;

[0079] S12 Path Planning and Dynamic Adjustment

[0080] Improved A algorithm combined with obstacle inflation mechanism, planning escape path, switching to backup path within 1 second when main path fails, issuing instructions to dynamic indicator light through 5G / LoRa, green light identifying safe passage, red light marking dangerous area;

[0081] S13 Emergency Linkage and Visualization

[0082] The path planning result is synchronized to the three-dimensional interface in FBX / GLTF format, supports dynamic rendering and API calling, and the alarm signal is preferentially pushed through 5G. The GPU computing power of the edge node is dynamically allocated.

[0083] The data acquisition module uses a million-level triangular facet laser radar carried by a drone to perform three-dimensional point cloud scanning on the power plant with centimeter-level resolution, covering static structures such as buildings, pipelines, and equipment. At the same time, 4K cameras are installed in various areas of the power plant to provide full coverage of the plant area for shooting high-definition images, capturing vehicle, object, road information, and personnel action and quantity information. In addition, an IoT sensor network composed of temperature and humidity sensors, smoke sensors, and vibration sensors is deployed at key locations in the power plant to collect environmental parameters in real time, including temperature and humidity values, smoke concentration values, and vibration intensity. After data collection, the system performs spatial alignment and coordinate system calibration on the laser radar point cloud data, 4K images, and original CAD drawings through multi-source data fusion technology, eliminating geometric deviations. Based on the fused data, a Unity3D open-source engine is used to build a basic three-dimensional model of the power plant, optimize triangular facets, and map textures, forming a visual model containing building, equipment, and pipeline hierarchical structures. At the same time, IoT sensor data is real-time connected to the three-dimensional model, dynamically labeling environmental parameters at corresponding locations: temperature and humidity are displayed in the form of a heat map, smoke concentration is labeled by color gradient to show the diffusion range, and vibration intensity is displayed in the form of a waveform or numerical value to show the equipment state. In the data processing stage, laser radar point cloud data needs to be processed by denoising and registration, and the iterative closest point algorithm is used to realize multi-view point cloud stitching to ensure model integrity. 4K camera images are corrected for lens distortion and color balance through the OpenCV library to eliminate edge distortion caused by wide-angle lenses. In the data consistency verification stage, the system compares the topological structure of the three-dimensional model and the CAD drawing to ensure that the equipment layout and pipeline layout are consistent, and checks the spatio-temporal synchronization of sensor data and model labeling, with an error control within ±0.5 meters. Finally, the basic three-dimensional model is exported in FBX / GLTF format, compatible with mainstream visualization platforms such as Unreal Engine and WebGL, and provides API interfaces for external system calling. Sensor data is pushed to edge computing nodes through the MQTT protocol, static structures are updated in full every 24 hours, and dynamic parameters are refreshed every second to ensure real-time synchronization between the model and the real scene. The technical implementation of this module balances high-precision modeling and dynamic data integration, providing reliable three-dimensional visualization foundation and real-time environmental perception capabilities for subsequent modules.

[0084] The deployment optimization module deploys distributed edge computing nodes, with single node supporting parallel processing of 10 camera pictures and multiple IoT sensor data streams. Through dynamic resource allocation mechanism, GPU computing power is adjusted in real time according to the rendering complexity of three-dimensional model, ensuring that the model visualization frame rate is stable at 30 frames per second or above. To adapt to the low power consumption requirement, a lightweight deep learning model based on YOLOv5s is used, and the inference speed is optimized through TensorRT or OpenVINO framework. The model running power consumption is controlled within 15 watts. In the data transmission link, multi-protocol hybrid communication is used to reduce bandwidth consumption: static sensor data such as temperature and humidity, vibration intensity are transmitted by LoRa protocol, with communication distance covering 2 kilometers, single node power consumption below 50 milliwatts, real-time video stream data is compressed to below 4 megabits per second by using H.265 encoding technology, combined with 5G slice network to realize low delay transmission, incremental synchronization mechanism further optimizes bandwidth utilization, only transmits differential data, reduces bandwidth demand by 70%. At the same time, data caching uses time window mechanism, local retention of the last 5 minutes of data to deal with network interruption, ensuring business continuity. The hardware selection takes low cost and high compatibility as the core, the temperature and humidity sensor uses industrial grade SHT35 device, through customized driver program to adapt to different manufacturers' equipment, reducing deployment complexity. In terms of energy efficiency optimization, dynamic task priority management mechanism is introduced to prioritize critical tasks and reduce CPU utilization by 20% after interrupting low priority tasks such as historical data archiving. The hardware temperature monitoring system detects the device running state in real time, when the temperature exceeds 60 degrees Celsius, the CPU frequency is automatically reduced from 2.4 GHz to 1.8 GHz, preventing overheating damage and prolonging the service life of the device. Balancing performance and cost, for example, by expanding the boundary 20 centimeters as a safety distance through obstacle inflation algorithm to optimize path planning accuracy, while avoiding hardware redundancy. In addition, the system supports remote firmware upgrade and configuration adjustment, further reducing the cost of later maintenance. The optimization strategy of this module covers the whole link of computing, communication and hardware, providing stable and low-cost foundation support for dynamic three-dimensional model construction and real-time emergency response.

[0085] The mobile trajectory positioning module achieves centimeter-level positioning accuracy through the Beidou / GPS dual-mode badge. The badge is equipped with an inertial navigation unit. In signal shielding scenarios, such as underground areas, the trajectory continuity is maintained through attitude compensation. The positioning data is updated once per second. The 4K camera deployed globally captures the facial features of personnel, and the improved YOLOv8 deep learning model is used for identity recognition and behavior analysis. The module records actions such as running, staying, or device operation, and binds the behavior labels with the positioning data to input a lightweight three-dimensional model. To improve trajectory accuracy, the Kalman filter algorithm is used to fuse the badge positioning data and the moving direction captured by the camera, eliminating sensor noise. The spatio-temporal synchronization error is controlled within ±0.3 meters. In terms of dynamic obstacle detection, the module analyzes the camera image based on the semantic segmentation algorithm, identifies road boundaries and moving vehicles such as forklifts and inspection vehicles, and detects temporary piles such as equipment parts and building materials using the optical flow method. The module real-time labels the road congestion area, and combines vibration sensor data such as abnormal vibration with a frequency of >5 millimeters per second squared and smoke concentration changes to identify potential danger points such as equipment failure or early fire. For detected obstacles, the module uses the dilation algorithm to expand the boundary by 20 centimeters as a safety distance, and updates the dynamic obstacle map every 2 seconds through the edge computing node, with an update delay of less than 500 milliseconds to ensure real-time data. On the data output level, the personnel movement trajectory is pushed to the intelligent path planning module through the MQTT protocol, including timestamp, coordinate sequence, dynamic obstacle coordinate list, and type, size, and danger level in the format compatible with the A-star algorithm. The single-frame data volume is compressed to below 200 kilobytes. When detecting that personnel enter a high-risk area or sudden obstacles such as fire smoke spread, the module triggers yellow, orange, and red three-level alarm signals, which are transmitted to the management terminal through the 5G network, and the three-dimensional visualization interface is highlighted for labeling.

[0086] The intelligent path planning module receives temperature and humidity, smoke concentration and vibration intensity data from the IoT sensor network, combines thermodynamic and fluid mechanics models, calculates the fire spread speed and predicts the diffusion trend in the next 30 seconds, and when the smoke concentration exceeds the threshold, the system automatically labels the potential diffusion path, generates a three-dimensional thermal radiation field distribution map, and maps it to the dynamic obstacle grid, which is updated every 2 seconds, synchronized to the path planning algorithm through the edge computing node, the path planning uses the A-star algorithm, combined with dynamic obstacle grid information and obstacle inflation mechanism, the algorithm expands a 20-centimeter safety boundary for detected static obstacles such as equipment and blocked roads, ensuring that the path matches the actual traffic area, when the main path is blocked by fire or obstacles, the system switches to the backup path within 1 second, and sends instructions to the dynamic indicator light through 5G or LoRa communication protocol, adjusts the arrow direction to identify the feasible path, the direction of the indicator light is always from inside to outside, the green light represents a safe passage, and the red light marks a dangerous area, such as a temperature exceeding 60℃ or a smoke-filled area, the end-to-end delay of the control signal is strictly controlled within 500 milliseconds, ensuring the real-time nature of the indication update, the module synchronously accesses personnel movement trajectory data, analyzes personnel density and movement speed, predicts path congestion probability in the next 30 seconds, and automatically generates a detour scheme to divert crowds, management personnel can manually adjust route priorities through a three-dimensional visualization interface, preferentially guiding paths away from high-temperature areas or forcing specific areas to evacuate in a specified order, path planning results are synchronized to the visualization interface in FBX / GLTF format, supporting dynamic rendering and interactive labeling, and providing an API interface for external systems to call, single-frame data is compressed to less than 200 kilobytes, suitable for low-bandwidth scenarios, in terms of emergency response, the module uses a three-level alarm mechanism, with yellow / orange / red classification to warn of extreme scenarios such as complete path blockage or uncontrolled fire, the alarm signal is preferentially pushed to the management terminal through the 5G network, at the same time, the edge computing node dynamically allocates GPU computing power according to real-time load, interrupts low-priority tasks such as historical data archiving, and reduces CPU utilization by 20%, ensuring that the path update frequency is stable at more than 30 frames / second, this module integrates multi-source data fusion and algorithm linkage to build a closed-loop emergency system from fire prediction, path optimization to device linkage, significantly improving power plant evacuation efficiency and safety.

[0087] Finally, it should be noted that: the above only for the preferred embodiments of the present application, and not for limiting the present application, although the foregoing detailed description of the application, for those skilled in the art, it still can be modified, or part of the technical features of the equivalent replacement, within the spirit and principles of the present application, any modification, equivalent replacement, improvement, etc., should be included within the scope of the present application.

Claims

1. A digital-based system for constructing a dynamic three-dimensional model of a power plant, characterized by: It includes a data acquisition module, a deployment optimization module, a mobile trajectory positioning module and an intelligent path planning module. The data acquisition module builds a basic three-dimensional model and integrates real-time environmental parameters through multi-source data fusion. The deployment optimization module processes multiple camera images and sensor data streams in parallel through distributed edge computing nodes. The mobile trajectory positioning module is used to track the dynamic trajectories and obstacle changes of people and vehicles in the power plant in real time. The intelligent path planning module is used to dynamically generate escape routes and realize emergency response.

2. The system for constructing a dynamic three-dimensional model of a power plant based on digitization according to claim 1, characterized in that: The data acquisition module uses a drone equipped with a laser radar to perform three-dimensional point cloud scanning of the entire power plant, adopts an iterative closest point algorithm to achieve multi-view point cloud stitching, and uses the OpenCV library to correct lens distortion and color balance for 4K camera images.

3. The system for constructing a dynamic three-dimensional model of a power plant based on digitization according to claim 1, characterized in that: The data cache of the deployment optimization module adopts a time window mechanism, and the latest time data is retained locally to cope with network interruptions.

4. The system for constructing a dynamic three-dimensional model of a power plant based on digitization according to claim 1, characterized in that: The trajectory data output by the mobile trajectory positioning module includes a timestamp, a coordinate sequence, and the type, size, and hazard level of dynamic obstacles.

5. The system for constructing a dynamic three-dimensional model of a power plant based on digitization according to claim 1, characterized in that: The intelligent path planning module adopts the A-star algorithm, combined with dynamic obstacle grid information and obstacle expansion mechanism, to expand the safety boundary for detected static obstacles.

6. The system for constructing a dynamic three-dimensional model of a power plant based on digitization according to claim 1, characterized in that: The three-dimensional model and sensor data of the system are exported in FBX / GLTF format and MQTT protocol respectively. The static structure is updated in full at regular intervals, and the dynamic parameters are refreshed every second.

7. The system for constructing a dynamic three-dimensional model of a power plant based on digitization according to claim 1, characterized in that: The mobile trajectory positioning module analyzes the camera image based on the semantic segmentation algorithm, identifies road boundaries and moving vehicles, and combines the optical flow method to detect temporary piles.

8. The system for constructing a dynamic three-dimensional model of a power plant based on digitization according to claim 1, characterized in that: The system operation process is as follows: S1 3D scanning and image acquisition Drones equipped with lidar perform centimeter-level resolution 3D point cloud scanning, covering power plant buildings, pipelines, and equipment; 4K cameras are installed in various areas of the power plant to capture full-area HD images, capturing information about vehicles, objects, roads, and people appearing in the images; S2 environmental parameter collection Deploy IoT sensor networks at key locations in power plants to collect real-time data on temperature, humidity, smoke concentration, and vibration intensity. S3 multi-source data fusion The LiDAR point cloud data is stitched, denoised, and registered using an iterative closest point algorithm. The 4K images are corrected for lens distortion and color balanced using the OpenCV library. The point cloud, image, and original CAD drawings are then fused, and geometric deviations are eliminated through spatial alignment and coordinate system calibration. S4 3D model construction Build a basic 3D model of the power plant using the Unity3D engine, optimize triangles and map textures to form a hierarchical structure including buildings, equipment, and pipelines; S5 edge computing node deployment Distributed edge nodes process 10 camera images and multiple sensor data streams in parallel, dynamically adjusting GPU computing power; S6 lightweight model and communication optimization It uses the YOLOv5s deep learning model and optimizes inference speed through TensorRT / OpenVINO. Static data is transmitted via the LoRa protocol, and video streams are compressed to below 4Mbps through H.265 encoding, combined with low-latency transmission on the 5G slicing network. S7 hardware management and energy efficiency optimization It uses an industrial-grade SHT35 temperature and humidity sensor, compatible with devices from multiple manufacturers, interrupts low-priority tasks, reduces CPU utilization, and reduces the CPU frequency from 4GHz to 1.8GHz when the hardware temperature exceeds 60°C. S8 Personnel / Vehicle Positioning Beidou / GPS dual-mode badges achieve centimeter-level positioning, and 4K cameras combined with the YOLOv8 model perform identity recognition and behavior analysis. S9 trajectory optimization and obstacle detection The Kalman filter algorithm integrates badge location and camera orientation data, the semantic segmentation algorithm identifies road boundaries and moving vehicles, and the optical flow method detects temporary piles of objects. The obstacle boundary is extended by 20 cm as a safety distance, and the obstacle map is updated every 2 seconds. S10 alarm and data transmission When a yellow / orange / red level alarm is triggered, it is pushed to the management terminal via 5G first, and the trajectory data is output via the MQTT protocol; S11 Fire Prediction and Obstruction Modeling Based on thermodynamics / fluid mechanics models, the fire spread trend within 30 seconds is predicted, and a three-dimensional heat radiation field distribution map is generated. The dynamic barrier grid is updated every 2 seconds and mapped to the path planning algorithm. S12 path planning and dynamic adjustment Improved A algorithm combined with obstacle expansion mechanism to plan escape routes. If the primary route fails, the backup route is switched within 1 second. Instructions are sent to dynamic indicator lights via 5G / LoRa. Green lights indicate safe passages, and red lights mark dangerous areas. S13 Emergency Linkage and Visualization The path planning results are synchronized to the 3D interface in FBX / GLTF format, supporting dynamic rendering and API calls. Alarm signals are pushed preferentially via 5G, and edge nodes dynamically allocate GPU computing power.

Images