Distribution network fault first-aid repair full-process visual tracking method and system

Abstract

The invention relates to a distribution network fault first-aid repair full-process visual tracking method and system, and the method comprises the following steps: building a multi-source data collection system, and obtaining the biological characteristic data, environmental parameters, vehicle positioning information and material RFID tag data of first-aid repair personnel in real time through intelligent terminal equipment; constructing a three-dimensional visual digital twinborn platform, and integrating a GIS (geographic information system), a BIM (building information modeling) power grid model and a real-time dynamic data layer; edge computing nodes are deployed for data preprocessing, and a time sequence compression algorithm is adopted to reduce the transmission bandwidth requirement; establishing a dynamic resource optimization model, and performing repair path planning and resource allocation based on a Markov decision process; the intelligent perception of the security situation is realized, the personnel behavior video stream is analyzed through the neural network model, and the illegal operation is pre-warned in real time.

Description

Technical Field

[0001] The present invention relates to a method and system for visual tracking of the entire process of distribution network fault repair, belonging to the technical field of distribution network automation. Background Art

[0002] With the continuous development of the power system, as an important part of the power system, the safe and stable operation of the distribution network is of great significance for ensuring people's livelihood and social and economic development. Distribution network fault repair is a key link in the operation and maintenance of the power system, and its efficiency and quality directly affect power supply reliability and user satisfaction. At present, distribution network fault repair mainly relies on traditional methods such as manual inspection, telephone reporting, and manual dispatching. There are many deficiencies in information acquisition, resource allocation, and safety control in this way.

[0003] In the prior art, there have been some technical solutions for distribution network fault repair. For example, CN112308250A discloses a distribution network emergency repair dispatching system and method. The system includes a distribution network operation status monitoring module, an equipment asset management module, an enterprise ledger module, a vehicle management module, a voice call module, a prediction analysis module, a human-computer interaction module, and a repair task management module, which are used to obtain distribution network historical state data and distribution network historical fault data, and generate repair tasks according to distribution network faults. Although this solution realizes the basic repair dispatching function, it lacks real-time monitoring of the biometric data of repair personnel and environmental parameters, and cannot comprehensively grasp the on-site situation.

[0004] CN117709927A proposes a distribution network repair virtual command method and system. This method uses the panoramic Internet of Things perception technology of the distribution network to access multi-source fault warning data, builds an intelligent judgment model for distribution network fault repair based on transfer learning and convolutional neural network theory, and uses text-to-speech technology and process automation technology for intelligent supervision and command. However, this solution mainly focuses on fault diagnosis and the command process, and lacks consideration for the dynamic optimization and allocation of repair resources and the perception of safety situations.

[0005] CN118313106A discloses a data-driven power grid repair risk assessment method and decision-making system. It constructs a power grid repair database through data-driven acquisition, uses digital twin technology to model the target repair power grid area, and conducts risk assessment and risk repair decision-making based on the power grid repair risk assessment network. This solution has innovation in risk assessment, but lacks the visual tracking of the entire repair process and the edge computing data preprocessing ability.

[0006] CN111582697A describes a method and system for evaluating and scheduling distribution network faults. The method constructs a digital urban distribution network topology map based on a geographic information system and distribution network construction information, uses a drone swarm to capture images along the inspection path and at the fault point, evaluates the disaster situation and traffic conditions through a convolutional neural network, and uses a simulated annealing algorithm to solve the optimal scheduling model. This solution has certain advantages in fault evaluation and path planning, but fails to fully utilize personnel biometric data and material RFID tag data for comprehensive decision-making.

[0007] In summary, the existing technologies have the following deficiencies in the emergency repair of distribution network faults: First, data collection is incomplete, lacking real-time acquisition of biometric data of emergency repair personnel, environmental parameters, vehicle positioning information, and material RFID tag data, and it is impossible to comprehensively grasp the on-site situation; second, the visualization degree is low, lacking a three-dimensional visualization digital twin platform integrating GIS geographic information system, BIM power grid model, and real-time dynamic data layer, and it is impossible to intuitively display the fault point and its influence range; third, the data processing efficiency is low, lacking edge computing nodes for data preprocessing, and it is impossible to effectively reduce the transmission bandwidth requirements; fourth, the resource allocation efficiency is not high, lacking a dynamic resource optimization model based on the Markov decision process, and it is difficult to achieve the optimal emergency repair path planning and resource allocation; fifth, the safety control is insufficient, lacking a safety situation intelligent perception system that analyzes the video stream of personnel behavior through a neural network model, and it is impossible to give real-time warnings of illegal operations. These problems have led to disadvantages such as low efficiency, high cost, and great safety hazards in the emergency repair of distribution network faults, and there is an urgent need for a method for emergency repair of distribution network faults that can achieve full-process visualization tracking. Summary of the Invention

[0008] The technical problem to be solved by the present invention is that, aiming at the problems of incomplete data collection, low visualization degree, low resource allocation efficiency, and insufficient safety control in the traditional emergency repair of distribution network faults, specifically manifested as the lack of real-time collection of biometric data of emergency repair personnel, environmental parameters, vehicle positions, and material information, and it is impossible to comprehensively grasp the on-site situation; lacking a three-dimensional visualization platform integrating GIS geography information, BIM power grid model, and real-time dynamic data, and it is impossible to intuitively display the fault point and its influence range; lacking a dynamic resource optimization model based on the Markov decision process, and it is difficult to achieve the optimal emergency repair path planning and resource allocation; lacking intelligent perception of personnel behavior and early warning of illegal operations, and it is impossible to effectively ensure the safety of emergency repair operations. These problems have led to disadvantages such as low efficiency, high cost, and great safety hazards in the emergency repair of distribution network faults.

[0009] The technical solution of the present invention is as follows:

[0010] On the one hand, the present invention provides a method for full-process visualization tracking of distribution network fault emergency repair, including the following steps:

[0011] Establish a multi-source data collection system to obtain real-time biometric data of emergency repair personnel, environmental parameters, vehicle positioning information, and material RFID tag data through intelligent terminal devices;

[0012] Build a 3D visualization digital twin platform, integrating a GIS geographic information system, a BIM power grid model, and a real-time dynamic data layer;

[0013] Deploy edge computing nodes for data preprocessing, and use time series compression algorithms to reduce the transmission bandwidth requirements;

[0014] Establish a dynamic resource optimization model, and perform emergency repair path planning and resource allocation based on the Markov decision process;

[0015] Realize intelligent perception of the security situation, analyze the video stream of personnel behavior through a neural network model, and give real-time warnings of illegal operations.

[0016] As a preferred implementation method, the intelligent terminal devices include intelligent safety helmets with UWB precise positioning functions, in-vehicle OBD diagnostic terminals, and material temperature and humidity sensing tags;

[0017] The biometric data of personnel includes heart rate, body temperature, blood oxygen saturation, and motion posture data; the collection of environmental parameters includes using explosion-proof multi-gas detectors to obtain combustible gas concentration, temperature and humidity, and oxygen content.

[0018] As a preferred implementation method, the dynamic resource optimization model incorporates:

[0019] A fault impact range prediction module based on a spatio-temporal graph convolutional network;

[0020] A vehicle scheduling optimization algorithm considering traffic condition evolution;

[0021] A Bayesian prediction model for material requirements;

[0022] An emergency repair task allocation strategy solved by a multi-objective genetic algorithm.

[0023] As a preferred implementation method, the security situation perception includes:

[0024] A personal protective equipment wearing recognition model improved based on YOLOv5;

[0025] An action specification analysis module using an ST-GCN spatio-temporal graph convolutional network;

[0026] A vital sign monitoring subsystem integrating millimeter-wave radar;

[0027] An intelligent emergency plan push mechanism based on a knowledge graph.

[0028] On the other hand, the present invention also provides a full-process visual tracking system for distribution network fault repair, including:

[0029] A multi-source data acquisition system construction module for establishing a multi-source data acquisition system to obtain real-time biometric data of repair personnel, environmental parameters, vehicle positioning information, and material RFID tag data through intelligent terminal devices;

[0030] A digital twin platform construction module for constructing a three-dimensional visual digital twin platform, integrating a GIS geographic information system, a BIM power grid model, and a real-time dynamic data layer;

[0031] A resource optimization module for deploying edge computing nodes for data preprocessing and using a time series compression algorithm to reduce the transmission bandwidth requirement;

[0032] A resource optimization module for establishing a dynamic resource optimization model and performing repair path planning and resource allocation based on the Markov decision process;

[0033] A security situation intelligent perception module for realizing intelligent perception of the security situation, analyzing the personnel behavior video stream through a neural network model, and giving real-time early warnings of illegal operations.

[0034] As a preferred implementation manner, the intelligent terminal devices include intelligent safety helmets with UWB precise positioning functions, vehicle-mounted OBD diagnostic terminals, and material temperature and humidity sensing tags;

[0035] The biometric data of personnel includes heart rate, body temperature, blood oxygen saturation, and motion posture data; the environmental parameter collection includes obtaining combustible gas concentration, temperature and humidity, and oxygen content using an explosion-proof multi-gas detector.

[0036] As a preferred implementation manner, the dynamic resource optimization model incorporates:

[0037] A fault impact range prediction module based on a spatio-temporal graph convolutional network;

[0038] A vehicle scheduling optimization algorithm considering traffic condition evolution;

[0039] A Bayesian prediction model for material requirements;

[0040] A repair task allocation strategy solved by a multi-objective genetic algorithm.

[0041] As a preferred implementation manner, the security situation perception includes:

[0042] A personal protective equipment wearing recognition model improved based on YOLOv5;

[0043] An action specification analysis module using an ST-GCN spatio-temporal graph convolutional network;

[0044] Vital sign monitoring subsystem integrating millimeter-wave radar;

[0045] Intelligent push mechanism for emergency plans based on knowledge graph.

[0046] In another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method described in any embodiment of the present invention is implemented.

[0047] In another aspect, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method described in any embodiment of the present invention is implemented.

[0048] The beneficial effects of the present invention are as follows: By establishing a multi-source data acquisition system, the real-time acquisition of biometric data, environmental parameters, vehicle positions, and material information of repair personnel is realized, comprehensively grasping the on-site situation, and improving the data integrity and real-time performance compared with traditional methods; By constructing a three-dimensional visualization digital twin platform, the fault points and influence ranges are intuitively displayed, making the repair decision-making more accurate and efficient; By deploying edge computing nodes and time series compression algorithms, the transmission bandwidth requirements are reduced, and the system response speed is improved; By establishing a dynamic resource optimization model based on Markov decision process, the optimal repair path planning and resource allocation are realized, significantly improving the resource utilization efficiency; Through the security situation intelligent perception system, the safety of repair operations is effectively guaranteed, and the incidence of safety accidents is reduced. The present invention solves the problems of incomplete data acquisition, low visualization level, low resource allocation efficiency, and insufficient safety control in traditional distribution network fault repair, improves the quality and efficiency of distribution network fault repair, reduces the repair cost, and reduces potential safety hazards.

[0049] The additional aspects and advantages of the present invention will be clarified in the following description, and some of them will be obvious from the description, or can be understood through the practice of the present invention. In addition, the various aspects and advantages of the present invention can be realized and obtained through the method steps and combinations specifically pointed out in the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic flowchart of the method according to Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] It should be understood that the step numbers used in the text are only for convenience of description and do not limit the order of execution of the steps.

[0053] It should be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0054] The terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0055] The term "and / or" refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0056] Embodiment 1:

[0057] See Figure 1 , this embodiment provides a method for visual tracking of the whole process of distribution network fault repair. This method realizes the visual tracking and management of the whole process of distribution network fault repair through technical means such as multi-source data acquisition, three-dimensional visualization platform construction, edge computing processing, dynamic resource optimization, and intelligent perception of security situation. The specific implementation steps are as follows:

[0058] Step S100: Establish a multi-source data acquisition system, and obtain the biometric data of repair personnel, environmental parameters, vehicle positioning information, and material RFID tag data in real time by deploying intelligent terminal devices.

[0059] The intelligent terminal devices include intelligent safety helmets with UWB precise positioning function, on-vehicle OBD diagnostic terminals, and material temperature and humidity sensing tags. Among them, the intelligent safety helmet adopts a high-precision UWB positioning module, and the positioning accuracy can reach ±10 cm, and the position information of the repair personnel is collected in real time; the on-vehicle OBD diagnostic terminal is connected to the vehicle electronic control unit through the CAN bus, and the sampling frequency is 10 Hz, and the vehicle position, speed, fuel consumption, etc. are obtained in real time; the material temperature and humidity sensing tag adopts low-power Bluetooth 5.0 technology, and the battery life can reach 6 months, and the sampling interval is 5 minutes.

[0060] Personnel biometric data includes heart rate, body temperature, blood oxygen saturation, and motion posture data. The heart rate collection range is 40 - 200 beats per minute, with an accuracy of ±2 beats per minute; the body temperature measurement range is 35 - 42 °C, with an accuracy of ±0.1 °C; the blood oxygen saturation measurement range is 70% - 100%, with an accuracy of ±2%; the motion posture data is collected by a nine-axis sensor, including acceleration, angular velocity, and geomagnetic field intensity, and the sampling frequency is 100 Hz.

[0061] Environmental parameter collection includes obtaining combustible gas concentration, temperature and humidity, and oxygen content using an explosion-proof multi-gas detector. The combustible gas detection range is 0 - 100% LEL, with an accuracy of ±3% LEL; the temperature detection range is -40 °C to 85 °C, with an accuracy of ±0.5 °C; the humidity detection range is 0 - 100% RH, with an accuracy of ±3% RH; the oxygen content detection range is 0 - 30%, with an accuracy of ±0.2%.

[0062] The data acquisition system adopts a hierarchical architecture design, including a sensing layer, a transmission layer, and an application layer. The sensing layer is responsible for collecting raw data, the transmission layer is responsible for data transmission and preprocessing, and the application layer is responsible for data storage and analysis. The data acquisition frequency is dynamically adjusted according to different data types. The acquisition frequency of key data such as personnel location information is 1 Hz, and the acquisition frequency of non-critical data such as environmental temperature and humidity is 0.033 Hz (once every 30 seconds).

[0063] The data acquisition system adopts a hierarchical architecture design, including a sensing layer, a transmission layer, and an application layer. The sensing layer is responsible for collecting raw data, the transmission layer is responsible for data transmission and preprocessing, and the application layer is responsible for data storage and analysis. The data acquisition frequency is dynamically adjusted according to different data types. The acquisition frequency of key data such as personnel location information is 1 Hz, and the acquisition frequency of non-critical data such as environmental temperature and humidity is 0.033 Hz (once every 30 seconds).

[0064] Step S200: Build a three-dimensional visualization digital twin platform. In this step, integrate the GIS geographic information system, the BIM power grid model, and the real-time dynamic data layer to build a three-dimensional visualization digital twin platform.

[0065] The GIS geographic information system adopts a hierarchical design, including a basic terrain layer, a road traffic layer, a building layer, a power distribution network layer, etc. The resolution of the basic terrain layer is 0.5 meters. The road traffic layer contains attribute information such as road grade, traffic status, and congestion index. The building layer contains information such as building outlines, heights, and types. The power distribution network layer contains the spatial positions and attribute information of equipment such as power distribution lines, substations, and distribution boxes.

[0066] The BIM power grid model adopts the IFC4 standard format and contains three-dimensional geometric information, material information, electrical parameters, etc. of distribution network equipment. The model accuracy reaches the LOD300 level, and can represent information such as the geometric dimensions, installation positions, and connection relationships of equipment. The BIM model and the GIS system achieve spatial registration through coordinate transformation, and the registration accuracy is better than 0.5 meters.

[0067] The real-time dynamic data layer uses the WebSocket protocol to achieve real-time data push, including dynamic information such as personnel positions, vehicle positions, equipment status, and environmental parameters. The data update frequency is dynamically adjusted according to the data type. The update frequency of key data such as personnel positions is 1Hz, and the update frequency of non-key data such as environmental parameters is 0.033Hz.

[0068] The three-dimensional visualization platform is implemented using WebGL technology and supports interactive operations such as multi-view switching, zooming and panning, and transparency adjustment. The platform rendering performance reaches 60FPS, and it supports the simultaneous display of 100 dynamic objects and 1000 static objects. The platform provides a variety of visualization expressions, including heat maps, streamline maps, bar charts, etc., for intuitively displaying information such as the distribution of emergency repair resources, personnel movement trajectories, and changes in environmental parameters.

[0069] Step S300: Deploy edge computing nodes for data preprocessing. In this step, a time series compression algorithm is used to reduce the transmission bandwidth requirements, and the edge computing nodes preprocess the collected raw data.

[0070] The edge computing nodes adopt ARMCortex-A72 architecture processors with a main frequency of 2.0GHz, 4GB of memory, 32GB of storage, and run a lightweight Linux operating system. The nodes are deployed at key locations such as substations and distribution rooms and are connected to the data acquisition terminals through wireless local area networks or low-power wide area networks.

[0071] Data preprocessing includes four links: data cleaning, anomaly detection, data fusion, and data compression. Data cleaning removes noise points through the median filtering algorithm, and the filtering window size is 5; anomaly detection uses a statistics-based method to mark data points that exceed the mean ± 3 times the standard deviation; data fusion uses the Kalman filtering algorithm to fuse multi-source sensor data to improve data reliability; data compression uses the piecewise linear approximation algorithm to compress time series data, and the compression ratio can reach 10:1, while ensuring that the root mean square error does not exceed 2% of the original data range.

[0072] The specific implementation steps of the time series compression algorithm are as follows:

[0073] 1. Segment the original time series data, and the length of each segment is N data points (N is dynamically adjusted according to the data type, and generally takes values from 10 to 100);

[0074] 2. Calculate the linear regression model for each data segment to obtain the slope k and the intercept b;

[0075] 3. Calculate the fitting error between the linear model and the original data. If the error is less than the preset threshold ε (usually taking a value of 1%-5% of the original data range), then use the linear model parameters k and b to replace the original data;

[0076] 4. If the fitting error is greater than the threshold ε, then bisect this data segment and repeat steps 2-3 for each sub-segment;

[0077] 5. Recursively execute the above process until the fitting errors of all data segments are less than the threshold ε, or the length of the data segment is less than the preset minimum length (usually taking a value of 2-5).

[0078] The edge computing node is also responsible for local storage and forwarding of data. Local storage uses a time series database, which supports efficient storage and query of time series data, and the storage capacity can store the original data of nearly 30 days. Data forwarding uses the MQTT protocol, which supports reliable data transmission and quality of service assurance.

[0079] Step S400: Establish a dynamic resource optimization model. In this step, based on the Markov decision process, conduct emergency repair path planning and resource allocation, and establish a dynamic resource optimization model.

[0080] The dynamic resource optimization model integrates a fault impact range prediction module based on a spatio-temporal graph convolutional network, a vehicle scheduling optimization algorithm considering traffic condition evolution, a Bayesian prediction model for material requirements, and a repair task allocation strategy solved by a multi-objective genetic algorithm.

[0081] The fault impact range prediction module based on the spatio-temporal graph convolutional network adopts the ST-GCN (Spatial-Temporal Graph Convolutional Network) architecture. The input is the distribution network topology structure and historical fault data, and the output is the fault impact range prediction result. The network contains 3 spatio-temporal convolutional blocks, each convolutional block contains 2 graph convolutional layers and 1 time convolutional layer, the hidden layer dimension is 64, and the activation function uses ReLU. The model training uses the Adam optimizer, the learning rate is 0.001, the batch size is 32, and the number of training epochs is 100. The model prediction accuracy reaches more than 85%.

[0082] The vehicle scheduling optimization algorithm considering traffic condition evolution is based on a time-varying road network model. The road network is represented as a weighted directed graph G(V, E, W), where V represents the set of nodes, E represents the set of edges, and W represents the edge weights (representing travel time). The algorithm considers the spatio-temporal evolution characteristics of traffic congestion, and predicts the road traffic conditions in the future period through historical traffic data and real-time traffic information. The path planning uses an improved A* algorithm, and the heuristic function considers the road congestion condition and time-varying characteristics to calculate the optimal emergency repair path.

[0083] The Bayesian prediction model for material requirements is based on historical emergency repair case data to establish the probability relationship between fault types and material requirements. The model uses a naive Bayesian classifier, and the feature vector includes factors such as fault types, faulty equipment, fault causes, and environmental conditions. The output is the probability distribution of the requirements for various materials. The model predicts the material requirement list under the current fault situation through maximum a posteriori probability estimation, and the prediction accuracy reaches over 80%.

[0084] The emergency repair task allocation strategy solved by the multi-objective genetic algorithm models the emergency repair task allocation problem as a multi-objective optimization problem. The objective functions include minimizing the total emergency repair time, minimizing resource consumption, maximizing emergency repair efficiency, etc. The algorithm uses the NSGA-II (Non-dominated Sorting Genetic Algorithm II) framework. The chromosome coding uses integer coding to represent the allocation relationship between emergency repair tasks and emergency repair teams. The crossover operation uses single-point crossover with a crossover probability of 0.8; the mutation operation uses uniform mutation with a mutation probability of 0.1; the selection operation uses tournament selection with a tournament size of 3. The population size is 100, the number of iterations is 200, and the convergence condition is that the optimal solution has no improvement for 30 consecutive generations.

[0085] The Markov decision process (MDP) model is defined as follows:

[0086] 1. State space S: includes the distribution state of emergency repair resources, fault state, traffic state, etc.;

[0087] 2. Action space A: includes resource allocation decisions, path planning decisions, task allocation decisions, etc.;

[0088] 3. State transition function P(s'|s,a): describes the probability of transitioning from state s to state s' after taking action a in state s;

[0089] 4. Reward function R(s,a): describes the immediate reward obtained by taking action a in state s, which is related to factors such as emergency repair efficiency and resource consumption;

[0090] 5. Discount factor γ: takes a value of 0.95 and is used to balance immediate rewards and long-term benefits.

[0091] The MDP model solves the optimal policy through the value iteration algorithm, and the iteration formula is:

[0092] V(s) = max_a[R(s,a) + γ∑P(s'|s,a)V(s')]

[0093] Where V(s) represents the value function of state s, and the iteration termination condition is that the maximum value of the difference between the value functions of two consecutive iterations is less than the preset threshold ε (with a value of 0.01).

[0094] Step S500: Implement intelligent safety situation perception. In this step, the neural network model analyzes the personnel behavior video stream to give real-time warnings of illegal operations and realizes intelligent safety situation perception.

[0095] Safety situation perception includes a personal protective equipment wearing recognition model improved based on YOLOv5, an action specification analysis module using the ST-GCN spatio-temporal graph convolutional network, a vital sign monitoring subsystem integrating millimeter-wave radar, and an intelligent emergency plan push mechanism based on a knowledge graph.

[0096] The personal protective equipment wearing recognition model improved based on YOLOv5 is used to detect whether repair personnel correctly wear personal protective equipment such as safety helmets, insulating gloves, and insulating boots. The model adopts the YOLOv5s architecture, the input image size is 640×640 pixels, the backbone network adopts CSPDarknet53, the feature pyramid network adopts PANet, and the detection head adopts multi-scale prediction. The model is trained on a safety protection equipment dataset, which contains 10,000 labeled images covering various lighting conditions and scenarios. The training uses the SGD optimizer, with an initial learning rate of 0.01, a momentum of 0.937, a weight decay of 0.0005, a batch size of 16, and 300 training epochs. The detection accuracy (mAP@0.5) of the model reaches over 92%, and the inference speed can reach 25FPS on NVIDIA Jetson Xavier NX.

[0097] The action specification analysis module using the ST-GCN spatio-temporal graph convolutional network is used to identify whether the operation actions of repair personnel conform to the specifications. The model represents the human skeleton as a graph structure, where the nodes represent human key points and the edges represent bone connections. The model contains 10 ST-GCN layers, the hidden layer dimension gradually increases from 64 to 256, and finally the action category is output through global average pooling and a fully connected layer. The model is trained on a power operation action dataset, which contains 50 standard operation actions and 20 illegal operation actions, and each action contains 100 samples. The training uses the Adam optimizer, with a learning rate of 0.0005, a batch size of 32, and 100 training epochs. The recognition accuracy of the model reaches over 90%.

[0098] The vital sign monitoring subsystem integrating millimeter-wave radar senses the vital sign data of emergency repair personnel through the fusion of millimeter-wave radar and wearable sensors. The millimeter-wave radar operates at a frequency of 77 GHz and can detect human micro-movements through clothing to achieve non-contact monitoring of heart rate and respiratory rate. The system uses the Kalman filtering algorithm to fuse the millimeter-wave radar data and wearable sensor data to improve the accuracy and reliability of vital sign monitoring. The system sets multiple warning thresholds, and when the vital sign parameters exceed the normal range, different levels of warnings are triggered according to the degree of deviation.

[0099] The intelligent push mechanism for emergency plans based on knowledge graph constructs a knowledge graph in the field of power emergency repair, including knowledge such as equipment entities, fault types, operation procedures, and safety measures. The knowledge graph contains 10,000 entities and 50,000 relationships and is stored in the Neo4j graph database. The system matches applicable emergency plans and operation procedures through knowledge reasoning according to the current fault situation and emergency repair progress and pushes them to relevant personnel. The push adopts role-based permission control to ensure that personnel of different roles obtain information consistent with their responsibilities.

[0100] The security situation awareness system adopts a distributed architecture. The front-end acquisition nodes are responsible for data acquisition and preliminary analysis, and the back-end server is responsible for in-depth analysis and decision support. The system response time is less than 1 second, the reliability reaches 99.9%, and it supports continuous operation for 7×24 hours.

[0101] As a preferred implementation mode of this embodiment, the intelligent safety helmet integrates a UWB positioning module, a biometric sensor, a head pose sensor, a camera, and a communication module. The UWB positioning module uses the DW1000 chip, operates at a frequency of 6.5 GHz, has a bandwidth of 500 MHz, a positioning accuracy of ±10 cm, and a refresh rate of 10 Hz. The biometric sensor includes a heart rate sensor, a body temperature sensor, and a blood oxygen sensor, adopts a low-power design, and the sampling frequency can be dynamically adjusted according to the working state. The head pose sensor uses the MPU9250 nine-axis sensor with a sampling frequency of 100 Hz. The camera uses the OV5640 image sensor with a resolution of 1280×720 pixels and a frame rate of 30 FPS. The communication module supports WiFi, Bluetooth, and 4G communications and automatically selects the optimal communication method according to the network environment. The battery capacity of the safety helmet is 5000 mAh and can work continuously for more than 8 hours.

[0102] The in-vehicle OBD diagnostic terminal is connected to the vehicle electronic control unit through the OBD-II interface to collect vehicle operation status data. The terminal uses an STM32F407 processor with a main frequency of 168 MHz, a memory of 192 KB, and a storage of 32 MB. The terminal integrates a GPS / Beidou dual-mode positioning module with a positioning accuracy better than 5 meters and a refresh rate of 1 Hz. The terminal also integrates a 4G communication module and a CAN bus interface, supporting real-time data upload and remote control. The data collected by the terminal includes parameters such as vehicle location, speed, engine speed, fuel consumption, battery voltage, and coolant temperature, with a sampling frequency of 10 Hz.

[0103] The material temperature and humidity sensing tag uses low-power Bluetooth 5.0 technology, is designed based on the nRF52832 chip, and integrates a temperature and humidity sensor and an RFID tag. The temperature sensor has a measurement range of -40°C to 85°C with an accuracy of ±0.5°C; the humidity sensor has a measurement range of 0 - 100%RH with an accuracy of ±3%RH. The tag is powered by a CR2032 button battery with a battery capacity of 225 mAh, a sampling interval of 5 minutes, and a battery life of up to 6 months. The tag supports the iBeacon protocol with a broadcast interval of 1 second and a transmission distance of up to 50 meters.

[0104] The personnel biometric data acquisition system adopts a hierarchical design, including a sensing layer, a processing layer, and a communication layer. The sensing layer is responsible for collecting raw biometric signals, the processing layer is responsible for signal filtering and feature extraction, and the communication layer is responsible for data transmission. Heart rate acquisition uses photoplethysmography (PPG) technology with a sampling frequency of 100 Hz, a measurement range of 40 - 200 beats per minute, and an accuracy of ±2 beats per minute. Body temperature measurement uses infrared thermometry technology with a sampling frequency of 0.2 Hz, a measurement range of 35 - 42°C, and an accuracy of ±0.1°C. Blood oxygen saturation measurement uses dual-wavelength PPG technology with a sampling frequency of 1 Hz, a measurement range of 70% - 100%, and an accuracy of ±2%. Motion posture data is collected through a nine-axis sensor, including a three-axis accelerometer (range ±16g, accuracy 0.5mg), a three-axis gyroscope (range ±2000° / s, accuracy 0.1° / s), and a three-axis magnetometer (range ±4900μT, accuracy 0.6μT), with a sampling frequency of 100 Hz.

[0105] For environmental parameter collection, an explosion-proof multi-gas detector is used. It adopts a modular design and can be configured with different gas sensors according to needs. The standard configuration includes a combustible gas sensor, an oxygen sensor, a carbon monoxide sensor, and a hydrogen sulfide sensor. The combustible gas detection uses a catalytic combustion sensor with a detection range of 0-100% LEL, an accuracy of ±3% LEL, and a response time of less than 15 seconds. The oxygen detection uses an electrochemical sensor with a detection range of 0-30%, an accuracy of ±0.2%, and a response time of less than 15 seconds. The temperature and humidity detection uses a digital temperature and humidity sensor. The temperature detection range is -40°C to 85°C with an accuracy of ±0.5°C; the humidity detection range is 0-100% RH with an accuracy of ±3% RH. The detector adopts an intrinsically safe design and complies with the ExiaIICT4 explosion-proof standard, and can be safely used in explosive gas environments.

[0106] In one embodiment, the specific implementation of the dynamic resource optimization model is optimized as follows:

[0107] The dynamic resource optimization model integrates a fault impact range prediction module based on a spatio-temporal graph convolutional network, a vehicle scheduling optimization algorithm considering traffic condition evolution, a Bayesian prediction model for material requirements, and a repair task allocation strategy solved by a multi-objective genetic algorithm.

[0108] The fault impact range prediction module based on the spatio-temporal graph convolutional network represents the distribution network as a graph structure G(V,E), where V represents the set of nodes (including substations, distribution boxes, users, etc.), and E represents the set of edges (representing line connection relationships). Each node and edge are associated with a set of temporal features, including electrical parameters such as voltage, current, and power. The ST-GCN model contains information extraction in two dimensions: spatial graph convolution and temporal convolution. The spatial graph convolution is implemented using ChebNet with a convolution kernel size of 3 to extract the spatial dependence relationship between nodes; the temporal convolution is implemented using 1D convolution with a convolution kernel size of 3 to extract the temporal dependence relationship of temporal features. The model input is a graph sequence with T time steps, and the output is the prediction of the fault impact range for the next K time steps. The model training uses historical fault data, including 1000 fault cases, and each case contains the power grid operation data before and after the fault and the annotation of the fault impact range. The training uses the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 100 training epochs. The F1 score of the model on the test set reaches 0.87.

[0109] The vehicle scheduling optimization algorithm considering traffic condition evolution is based on a time-varying road network model. The road network is represented as a weighted directed graph G(V, E, W(t)), where W(t) represents the time-varying edge weight, reflecting the road travel time in different time periods. The algorithm uses historical traffic data and real-time traffic condition information to establish a time series prediction model of road travel time. The prediction model uses an LSTM (Long Short-Term Memory) network. The input is the road travel time series in the past 24 hours, and the output is the prediction of the road travel time in the next 2 hours. The LSTM network contains 2 hidden layers, with 128 neurons in each layer, and the dropout rate is 0.2. The model training uses the root mean square error (RMSE) as the loss function, the Adam optimizer, the learning rate is 0.001, the batch size is 64, and the number of training epochs is 50. The path planning uses an improved A* algorithm, and the heuristic function h(n) is defined as:

[0110] h(n) = α·d(n, goal) / v_max + β·c(n)

[0111] where d(n, goal) represents the Euclidean distance from node n to the target node, v_max represents the maximum driving speed, c(n) represents the congestion coefficient of node n, and α and β are weight coefficients, taking values of 0.7 and 0.3 respectively. The algorithm considers the collaborative scheduling of multiple vehicles and realizes the optimal scheduling of emergency repair vehicles through task decomposition and resource allocation.

[0112] The Bayesian prediction model for material requirements is based on historical emergency repair case data to establish the probability relationship between fault types and material requirements. The model uses a naive Bayes classifier. The feature vector X includes factors such as fault types, faulty equipment, fault causes, and environmental conditions, and the output Y is the probability distribution of the requirements for various materials. According to Bayes' theorem:

[0113] P(Y|X) = P(X|Y)·P(Y) / P(X);

[0114] where P(Y) is the prior probability, P(X|Y) is the likelihood function, and P(X) is the evidence factor. The model predicts the material requirement list under the current fault condition through maximum a posteriori probability estimation:

[0115] Y* = argmax_YP(Y|X);

[0116] The model training uses 5000 historical emergency repair cases, and each case contains fault information and actual material consumption records. The model uses Laplace smoothing to handle sparse features, and the smoothing parameter α = 1. The prediction accuracy of the model on the test set reaches 83%.

[0117] The emergency repair task allocation strategy solved by the multi-objective genetic algorithm models the emergency repair task allocation problem as a multi-objective optimization problem. The objective functions include:

[0118] 1. Minimize the total emergency repair time: f1(x) = max{t1(x), t2(x),..., t n (x)};

[0119] 2. Minimize resource consumption: f2(x) = ∑ i ∑ j r ij ·x ij ;

[0120] 3. Maximize the emergency repair efficiency: f3(x) = ∑ i ∑ j e ij ·x ij ;

[0121] where x represents the task allocation plan, t i (x) represents the completion time of the i-th task, r ij represents the resource consumption of assigning task i to team j, e ij represents the emergency repair efficiency of assigning task i to team j, x ij is a 0-1 variable indicating whether task i is assigned to team j.

[0122] The algorithm adopts the NSGA-II framework. The chromosome coding uses integer coding, and the length is equal to the number of tasks. The value of each gene represents the team number responsible for the task. The crossover operation adopts single-point crossover with a crossover probability of 0.8; the mutation operation adopts uniform mutation with a mutation probability of 0.1; the selection operation adopts tournament selection with a tournament size of 3. The population size is 100, the number of iterations is 200, and the convergence condition is that there is no improvement in the optimal solution for 30 consecutive generations. The algorithm outputs a set of non-dominated solutions, and the decision maker can select the most suitable solution according to the actual situation.

[0123] The Markov decision process model solves the optimal strategy through the value iteration algorithm. The iteration process is as follows:

[0124] 1. Initialize the value function V0(s) = 0,

[0125] 2. For k = 1, 2,..., perform the following iteration:

[0126] V k (s) = max_a[R(s, a) + γ∑P(s'|s, a)V k-1 (s')];

[0127] 3. When ||V k - Vk-1 Stop iterating when ||∞ < ε;

[0128] 4. Output the optimal policy π*(s) = argmax_a[R(s, a) + γ∑P(s'|s, a)V*(s')];

[0129] The dimension of the model state space is 10 6 Magnitude. To improve the computational efficiency, state abstraction and function approximation techniques are adopted. State abstraction maps the original state to a low-dimensional feature space through feature extraction; function approximation represents the value function using a deep neural network. The network contains 3 hidden layers, with 128 neurons in each layer, and the ReLU activation function is used. The model training adopts the experience replay technique, with the replay buffer size of 10000 and the batch size of 32.

[0130] In one embodiment, the specific implementation of the intelligent perception of the security situation is optimized as follows:

[0131] Step Five: Implement the intelligent perception of the security situation

[0132] In this step, the security situation perception includes a personal protective equipment wearing recognition model improved based on YOLOv5, an action specification analysis module using the ST-GCN spatio-temporal graph convolutional network, a vital sign monitoring subsystem integrating millimeter-wave radar, and an intelligent emergency plan push mechanism based on the knowledge graph.

[0133] The personal protective equipment wearing recognition model improved based on YOLOv5 is optimized for the power repair scenario. The model detection objects include 10 types of protective equipment such as safety helmets (5 colors), insulating gloves (3 types), insulating boots (2 types), safety belts, and goggles. The model adopts the YOLOv5s architecture and has the following improvements:

[0134] 1. The backbone network uses CSPDarknet53 and introduces an attention mechanism (CBAM module) to enhance the ability to extract key features;

[0135] 2. The feature pyramid network uses the improved PANet to increase cross-layer connections and improve the multi-scale feature fusion effect;

[0136] 3. The detection head uses multi-scale prediction and introduces the IoU-aware loss function to improve the bounding box regression accuracy;

[0137] 4. Data augmentation uses Mosaic and MixUp techniques to increase the diversity of training samples.

[0138] The model is trained on a dataset of safety protection equipment for electric power operations. The dataset contains 10,000 annotated images, covering various lighting conditions, weather conditions, and scenarios. The training uses the SGD optimizer with an initial learning rate of 0.01, a cosine annealing scheduling strategy, a momentum of 0.937, a weight decay of 0.0005, a batch size of 16, and 300 training epochs. The detection accuracy (mAP@0.5) of the model reaches 94.5%, and the inference speed can reach 25 FPS on NVIDIA Jetson Xavier NX and 5 FPS on Raspberry Pi 4B.

[0139] The action specification analysis module using the ST-GCN spatio-temporal graph convolutional network is optimized for electric power emergency repair operations. The model represents the human skeleton as a graph structure G(V,E), which contains 17 key points (nodes) and 16 bone connections (edges). The model uses OpenPose to extract the key points of the human skeleton in real time and then inputs them into the ST-GCN network for action recognition. The ST-GCN network contains 10 ST-GCN layers, and the dimension of the hidden layer gradually increases from 64 to 256. Finally, the action category is output through global average pooling and a fully connected layer. The action categories recognized by the model include 50 standard operation actions (such as insulation operation, grounding operation, measurement operation, etc.) and 20 violation operation actions (such as unprotected operation, incorrect sequence operation, etc.).

[0140] The model is trained using a dataset of electric power operation actions, which contains 70 action categories. Each action contains 100 samples, and each sample is a 30-frame skeleton sequence. The training uses the Adam optimizer with a learning rate of 0.0005, a learning rate decay strategy, and it decays to 0.1 times the original every 20 epochs. The batch size is 32, and the number of training epochs is 100. The recognition accuracy of the model on the test set reaches 92.3%. The model is deployed on an edge computing node to analyze the actions of workers in real time. When a violation operation is detected, an alarm is immediately triggered.

[0141] The vital sign monitoring subsystem integrating millimeter-wave radar uses multi-sensor fusion technology, combining the advantages of millimeter-wave radar and wearable sensors. The millimeter-wave radar uses FMCW (Frequency Modulated Continuous Wave) technology, with a working frequency of 77 GHz and a bandwidth of 4 GHz. It can detect human micro-movements through clothing to achieve non-contact monitoring of heart rate and breathing rate. The radar signal processing process includes:

[0142] 1. Range-Doppler processing: Convert the time-domain signal to a range-Doppler spectrum through a two-dimensional FFT;

[0143] 2. Target detection and tracking: Detect the target through the CFAR algorithm and use Kalman filtering for target tracking;

[0144] 3. Micro-motion extraction: extract the micro-motion information of the target through phase demodulation;

[0145] 4. Vital signs analysis: Extract heart rate and respiratory rate through spectrum analysis.

[0146] The system uses the Kalman filter algorithm to fuse millimeter-wave radar data and wearable sensor data. The state vector contains heart rate, respiratory rate, body temperature, and blood oxygen saturation, and the observation vector contains the measurement values of each sensor. The Kalman filter parameters are set as follows: the diagonal elements of the process noise covariance matrix Q are [0.5, 0.2, 0.1, 0.3], and the diagonal elements of the observation noise covariance matrix R are [1.0, 0.5, 0.2, 0.5]. The system sets multi-level warning thresholds. When the vital sign parameters exceed the normal range, different levels of warnings are triggered according to the degree of deviation:

[0147] 1. Level 1 warning (yellow): The parameter slightly deviates from the normal range, and the system records and reminds;

[0148] 2. Level 2 warning (orange): The parameter obviously deviates from the normal range, the system alarms and recommends taking a rest;

[0149] 3. Level 3 warning (red): Parameters deviate seriously from the normal range, the system issues an emergency alarm and requires immediate cessation of operations.

[0150] The knowledge graph-based emergency plan intelligent push mechanism has constructed a knowledge graph in the field of power repair, which includes equipment entities, fault types, operating procedures, safety measures and other knowledge. The knowledge graph uses the triple <subject, relationship, object> to represent knowledge, including 10,000 entities and 50,000 relationships. Entity types include equipment (transformers, switches, lines, etc.), fault types (short circuits, broken wires, insulation breakdown, etc.), operation types (detection, isolation, grounding, etc.), safety measures (personal protection, on-site alert, etc.), etc. Relationship types include "is a kind of", "component", "cause", "need", "precondition", etc.

[0151] The knowledge graph is stored in the Neo4j graph database, and knowledge retrieval is performed through the Cypher query language. The system matches applicable emergency plans and operating procedures based on the current fault situation and repair progress through knowledge reasoning. Knowledge reasoning uses a combination of rule-based reasoning and embedding-based reasoning. Rule-based reasoning uses predefined reasoning rules to derive knowledge; embedding-based reasoning uses the TransE model to map entities and relationships to a low-dimensional vector space, and performs similarity calculations and relationship predictions through vector operations.

[0152] Based on the reasoning results, the system intelligently pushes relevant emergency response plans, operating procedures, and safety precautions in combination with the user role and the current task stage. The push adopts role-based permission control to ensure that personnel of different roles can obtain information that matches their responsibilities. The pushed content includes various forms such as text descriptions, operation steps, precautions, reference pictures, and videos, and supports voice announcements and AR augmented reality displays.

[0153] Embodiment 2:

[0154] This embodiment provides a full-process visualization tracking system for distribution network fault repair, including:

[0155] A multi-source data acquisition system construction module for establishing a multi-source data acquisition system to obtain real-time biometric data, environmental parameters, vehicle positioning information, and material RFID tag data of repair personnel through intelligent terminal devices; this module is used to implement the function of step S100 in Embodiment 1 and will not be elaborated here;

[0156] A digital twin platform construction module for constructing a three-dimensional visualization digital twin platform, integrating a GIS geographic information system, a BIM power grid model, and a real-time dynamic data layer; this module is used to implement the function of step S200 in Embodiment 1 and will not be elaborated here;

[0157] A resource optimization module for deploying edge computing nodes for data preprocessing and using a time series compression algorithm to reduce the transmission bandwidth requirements; this module is used to implement the function of step S300 in Embodiment 1 and will not be elaborated here;

[0158] A resource optimization module for establishing a dynamic resource optimization model and performing repair path planning and resource allocation based on the Markov decision process; this module is used to implement the function of step S400 in Embodiment 1 and will not be elaborated here;

[0159] A safety situation intelligent perception module for realizing safety situation intelligent perception, analyzing the video stream of personnel behavior through a neural network model, and giving real-time warnings of illegal operations; this module is used to implement the function of step S500 in Embodiment 1 and will not be elaborated here.

[0160] As a preferred implementation method, the intelligent terminal device includes an intelligent safety helmet with UWB precise positioning function, an in-vehicle OBD diagnostic terminal, and a material temperature and humidity sensing tag;

[0161] The personnel biometric data includes heart rate, body temperature, blood oxygen saturation, and motion posture data; the environmental parameter collection includes obtaining combustible gas concentration, temperature and humidity, and oxygen content using an explosion-proof multi-gas detector.

[0162] As a preferred implementation method, the dynamic resource optimization model incorporates:

[0163] Fault impact range prediction module based on spatio-temporal graph convolutional network;

[0164] Vehicle scheduling optimization algorithm considering traffic condition evolution;

[0165] Bayesian prediction model for material requirements;

[0166] Repair task allocation strategy solved by multi-objective genetic algorithm.

[0167] As a preferred implementation, the security situation awareness includes:

[0168] Personal protective equipment wearing recognition model improved based on YOLOv5;

[0169] Action specification analysis module using ST-GCN spatio-temporal graph convolutional network;

[0170] Vital sign monitoring subsystem integrated with millimeter-wave radar;

[0171] Intelligent emergency plan push mechanism based on knowledge graph.

[0172] Embodiment III:

[0173] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the method described in any embodiment of the present invention.

[0174] Embodiment IV:

[0175] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method described in any embodiment of the present invention.

[0176] In the embodiments of the present application, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent the situation of A existing alone, A and B existing simultaneously, and B existing alone. Where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one of the following" and its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0177] Those of ordinary skill in the art can realize that the units and algorithm steps described in the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0178] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0179] In several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (hereinafter referred to as ROMs), random access memories (hereinafter referred to as RAMs), magnetic disks, or optical discs that can store program codes.

[0180] The above are only the embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A full-process visual tracking method for distribution network fault repair, characterized in that The following steps are involved: Establish a multi-source data collection system to obtain repair personnel's biometric data, environmental parameters, vehicle location information, and material RFID tag data in real time through intelligent terminal devices; Build a 3D visualization digital twin platform that integrates GIS, BIM power grid model and real-time dynamic data layer; Deploy edge computing nodes for data preprocessing and use time series compression algorithms to reduce transmission bandwidth requirements; Establish a dynamic resource optimization model to carry out repair path planning and resource allocation based on the Markov decision process; Realize intelligent perception of security situation, analyze video streams of personnel behavior through neural network models, and issue real-time warnings for illegal operations.

2. A method for visually tracking the entire process of distribution network fault repair according to claim 1, characterized in that: The intelligent terminal equipment includes an intelligent helmet with UWB precise positioning function, an on-board OBD diagnostic terminal, and a material temperature and humidity sensor tag; Personnel biometric data includes heart rate, body temperature, blood oxygen saturation and movement posture data; environmental parameter collection includes the use of explosion-proof multi-gas detectors to obtain combustible gas concentration, temperature, humidity and oxygen content.

3. A full-process visualization tracking method for distribution network fault repair according to claim 1, characterized in that The dynamic resource optimization model incorporates: Fault impact range prediction module based on spatiotemporal graph convolutional network; Vehicle dispatch optimization algorithm considering traffic condition evolution; Bayesian forecasting model for material demand; Repair task allocation strategy solved by multi-objective genetic algorithm.

4. A full-process visualization tracking method for distribution network fault repair according to claim 1, characterized in that The security situation awareness includes: Improved personal protective equipment wearing recognition model based on YOLOv5; The action specification analysis module adopts ST-GCN spatiotemporal graph convolutional network; A vital signs monitoring subsystem that integrates millimeter-wave radar; Intelligent push mechanism for emergency plans based on knowledge graph.

5. A full-process visual tracking system for distribution network fault repair, characterized in that, include: The multi-source data collection system construction module is used to establish a multi-source data collection system to obtain the repair personnel's biometric data, environmental parameters, vehicle positioning information and material RFID tag data in real time through intelligent terminal devices; Digital twin platform building module, used to build a three-dimensional visual digital twin platform, integrating GIS geographic information system, BIM power grid model and real-time dynamic data layer; Resource optimization module, which is used to deploy edge computing nodes for data preprocessing and use time series compression algorithms to reduce transmission bandwidth requirements; Resource optimization module, which is used to establish a dynamic resource optimization model and perform repair path planning and resource allocation based on the Markov decision process; The security situation intelligent perception module is used to realize intelligent perception of security situation, analyze the video stream of personnel behavior through the neural network model, and warn of illegal operations in real time.

6. A visual tracking system for the entire process of distribution network fault repair according to claim 5, characterized in that: The intelligent terminal equipment includes an intelligent helmet with UWB precise positioning function, an on-board OBD diagnostic terminal, and a material temperature and humidity sensor tag; Personnel biometric data includes heart rate, body temperature, blood oxygen saturation and movement posture data; environmental parameter collection includes the use of explosion-proof multi-gas detectors to obtain combustible gas concentration, temperature, humidity and oxygen content.

7. A visualization tracking system for the whole process of distribution network fault repair according to claim 5, characterized in that, The dynamic resource optimization model incorporates: Fault impact range prediction module based on spatiotemporal graph convolutional network; Vehicle scheduling optimization algorithm considering traffic condition evolution; Material demand Bayesian prediction model; Repair task allocation strategy solved by multi-objective genetic algorithm.

8. A visualization tracking system for the whole process of distribution network fault repair according to claim 5, characterized in that, The security situation awareness includes: Personal protective equipment wearing recognition model improved based on YOLOv5; Action specification analysis module using ST-GCN spatio-temporal graph convolutional network; Vital sign monitoring subsystem integrated with millimeter wave radar; Intelligent emergency plan push mechanism based on knowledge graph.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that When the processor executes the program, it realizes the whole-process visual tracking method for distribution network fault repair as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it realizes the whole-process visual tracking method for distribution network fault repair as described in any one of claims 1 to 7.

Images