A 5G-based intelligent traffic management system and method for smart cities

Abstract

This invention proposes a 5G-based intelligent traffic management system and method for smart cities. By acquiring various data from the target city, a digital twin model of the city is constructed, and video and environmental data collection terminals are deployed on the target roads. These terminals upload the video and environmental data to edge computing nodes via 5G. The edge computing nodes use technologies such as object detection and semantic segmentation to identify traffic and environmental objects from the video and predict the risk index of each road segment in real time through deep learning. If the risk index exceeds the limit, the edge computing nodes send control commands to traffic lights via 5G to adjust traffic flow. Simultaneously, the edge computing nodes upload congestion warning results to an encrypted blockchain and send them to onboard devices. For emergency events, the edge computing nodes use blockchain and federated learning to find the optimal troubleshooting solution and continuously optimize it. This invention integrates edge computing, blockchain, and multidisciplinary technologies to improve the analytical efficiency and service security of traffic management.

Description

Technical Field

[0001] This invention relates to the field of smart city technology, specifically to a 5G-based intelligent traffic management system and method for smart cities. Background Technology

[0002] In modern urban life, cities are constantly expanding, populations are enormous, and the number of motor vehicles is rapidly increasing. Due to the limited transportation resources and the lag in transportation planning, traffic problems have become an increasingly important concern.

[0003] With the development of science and technology, various new technologies and concepts are emerging and being applied to the transportation field, such as intelligent transportation, smart cities, and smart traffic management. Smart city traffic control utilizes next-generation information technologies such as the Internet of Things, spatial sensing, cloud computing, and mobile internet to collect various operational data on urban traffic, analyze them, and make decisions to control traffic. However, current traffic management systems have shortcomings in terms of efficiency, safety, and management level. Summary of the Invention

[0004] Based on the above-mentioned problems, this invention proposes a smart city intelligent traffic management system and method based on 5G. By utilizing edge computing, blockchain and multidisciplinary technologies, it improves the analysis efficiency and service security of traffic management.

[0005] In view of this, one aspect of the present invention proposes a 5G-based intelligent traffic management system for smart cities, comprising: a cloud server, an edge computing node, a video acquisition terminal, an environmental data acquisition terminal, and a 5G base station; wherein,

[0006] The cloud server is configured as follows:

[0007] Acquire planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data of the target city;

[0008] A digital twin model of the target city is constructed based on the planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data.

[0009] According to the city digital twin model, multiple video acquisition terminals and environmental data acquisition terminals interconnected via 5G networks are configured on target roads within the target city area;

[0010] The video acquisition terminal and the environmental data acquisition terminal are respectively used to upload the first video data and the first environmental data collected by the 5G base station to the edge computing node;

[0011] The edge computing node is configured as follows:

[0012] Object detection and semantic segmentation are performed on the first video data to identify vehicles, people and environmental objects, thus obtaining the first video recognition data;

[0013] Using a deep learning model, real-time predictive analysis is performed on the first video recognition data and the first environmental data to assess the congestion risk index and personnel safety risk index of each road segment.

[0014] If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, a first control command will be sent to the 5G base station.

[0015] The 5G base station is configured to control the corresponding traffic lights according to the first control command to adjust the traffic flow.

[0016] The edge computing node is configured as follows:

[0017] The congestion warning results are uploaded to a blockchain node and then pushed to the vehicle-mounted device after being encrypted and verified.

[0018] In the event of an emergency, blockchain technology is used to find the optimal troubleshooting solution, and the solution is continuously optimized through a cross-domain federated learning model.

[0019] Another aspect of the present invention provides a 5G-based intelligent traffic management method for smart cities, comprising:

[0020] Acquire planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data of the target city;

[0021] A digital twin model of the target city is constructed based on the planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data.

[0022] Based on the urban digital twin model, multiple video acquisition terminals and environmental data acquisition terminals interconnected by a 5G network are configured on target roads within the target urban area;

[0023] The video acquisition terminal and the environmental data acquisition terminal respectively upload the acquired first video data and first environmental data to the edge computing node via the 5G base station;

[0024] The edge computing node performs object detection and semantic segmentation on the first video data to identify vehicles, people and environmental objects, and obtains the first video recognition data.

[0025] The edge computing node uses a deep learning model to perform real-time predictive analysis on the first video recognition data and the first environmental data to assess the congestion risk index and personnel safety risk index of each road segment.

[0026] If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, the edge computing node will send a first control command to the 5G base station.

[0027] The 5G base station controls the corresponding traffic lights to adjust the traffic flow according to the first control command.

[0028] The edge computing node uploads the congestion warning results to the blockchain node, and pushes them to the vehicle-mounted device after encryption and verification.

[0029] In the event of an emergency, the edge computing node uses blockchain technology to find the optimal troubleshooting solution and continuously optimizes the solution through a cross-domain federated learning model.

[0030] Optionally, the step of the edge computing node using a deep learning model to perform real-time predictive analysis on the first video recognition data and the first environmental data, and assessing the congestion risk index and personnel safety risk index of each road segment, includes:

[0031] The edge computing node acquires the traffic facility data, the historical traffic status data, and the historical road video data;

[0032] The historical road video data is aggregated using a time-series database to extract characteristics of vehicle and pedestrian traffic changes;

[0033] By using a deep learning model and combining it with feature sequences extracted from the changes in vehicle and pedestrian traffic, a traffic state prediction model is obtained.

[0034] Input the first video recognition data and the first environmental data into the traffic state prediction model to predict the traffic distribution in future time periods;

[0035] Based on the traffic flow distribution prediction results, the traffic flow difficulty coefficient of each road segment is assessed as a congestion risk index.

[0036] Real-time object detection algorithms and human pose estimation algorithms are applied to analyze the behavior of pedestrian flow and identify abnormal events.

[0037] Based on the traffic distribution prediction results, the abnormal events, and the environmental data, the personnel safety risk index of different areas is assessed.

[0038] Optionally, the step of sending a first control command to the 5G base station if the congestion risk index and / or the personnel safety risk index exceed a preset risk index threshold includes:

[0039] The edge computing node compares the congestion risk index and / or the personnel safety risk index with a preset risk index threshold.

[0040] If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, a risk event is determined to have occurred.

[0041] The edge computing node performs risk event identification and analysis on the first video recognition data and the first environmental data to determine the type and status of the first risk event that has occurred.

[0042] Generate a corresponding first control instruction based on the type and status of the first risk event;

[0043] The edge computing node sends the first control command to the corresponding 5G base station.

[0044] Optionally, the 5G-based smart city intelligent traffic management method further includes:

[0045] The on-board device uploads the first vehicle data of the first vehicle it is in to the corresponding edge computing node;

[0046] The edge computing node parses the first traffic data, the first road status data, and the first road facility data from the first video recognition data, and uploads the first traffic data, the first road status data, the first road facility data, the first environmental data, and the first vehicle data to the cloud server;

[0047] The cloud server analyzes the first traffic data, the first road status data, the first road facility data, the first environmental data, and the first vehicle data, and combines them with a preset global optimal traffic organization strategy and vehicle platooning scheme to divide all vehicles traveling on the road into multiple vehicle groups.

[0048] For each vehicle group, the first driving status data of each member vehicle in each group, the first mutual influence relationship data between each member vehicle, and the first current environmental condition data and the first current road condition data of each member vehicle are obtained.

[0049] Based on the first driving status data, the first mutual influence relationship data, the first current environment data, and the first current road condition data, the first risk existing in each member vehicle of each vehicle group is analyzed, and the control of each member vehicle in the vehicle group is carried out according to the first risk.

[0050] The vehicle members of each vehicle group are dynamically adjusted, and the vehicles are managed according to the adjusted vehicle group situation.

[0051] Optionally, the cloud server, by analyzing the first traffic data, the first road status data, the first road infrastructure data, the first environmental data, and the first vehicle data, and combining a preset globally optimal traffic organization strategy and vehicle platooning scheme, divides all vehicles traveling on the road into multiple vehicle groups, including:

[0052] A multi-source heterogeneous data integration platform is built on the cloud server;

[0053] The multi-source heterogeneous data integration platform cleans, fuses, and encodes the received first traffic data, first road status data, first road facility data, first environmental data, and first vehicle data to obtain first comprehensive data.

[0054] A traffic status digital twin model is generated based on the first comprehensive data and the city digital twin model (the traffic status digital twin model includes at least the road, speed, and destination of the first vehicle, and can reflect the traffic flow, congestion, and road facility status of each road segment in real time, and can accurately predict future traffic trends).

[0055] Based on the traffic state digital twin model, and combined with the preset global optimal traffic organization strategy and vehicle platooning scheme, all the first vehicles traveling on the road will be divided into multiple vehicle groups.

[0056] Optionally, the step of acquiring, on a per-vehicle-group basis, the first driving status data of each member vehicle in each group, the first mutual influence relationship data between each member vehicle, and the first current environmental condition data and the first current road condition data of each member vehicle, includes:

[0057] Each of the aforementioned member vehicles collects its own first driving status data through its own configured on-board equipment;

[0058] Each of the member vehicles establishes a communication connection with other vehicles in the same vehicle group through the on-board equipment, and exchanges first mutual influence relationship data.

[0059] Each of the aforementioned member vehicles collects first current environmental data and first current road condition data through the on-board equipment.

[0060] Optionally, the step of analyzing the first risk existing in each member vehicle of each vehicle group based on the first driving status data, the first mutual influence relationship data, the first current environment data, and the first current road condition data, and controlling each member vehicle of the vehicle group according to the first risk, includes:

[0061] Acquire historical vehicle driving status data, historical environmental data, historical road data, and historical accident data that have spatiotemporal correspondence;

[0062] Big data analytics is used to analyze the historical vehicle driving status data, historical environmental data, historical road data, and historical accident data. Based on the analysis results, a multi-dimensional risk assessment index system is constructed that includes vehicle-specific factors, road factors, environmental factors, and inter-vehicle interaction factors.

[0063] By using the historical vehicle driving status data, the historical environmental data, the historical road data, the historical accident data, and the multi-dimensional risk assessment index system, a neural network is trained to learn the mapping relationship between risk assessment indicators and risk levels, thereby obtaining a risk identification model.

[0064] The first driving status data, the first mutual influence relationship data, the first current environment data, and the first current road condition data are input into the trained risk identification model to obtain the risk score and risk type of the first risk.

[0065] For high-risk vehicles whose risk scores exceed a preset score threshold, determine the first risk type corresponding to the first risk.

[0066] Construct a dynamic model to describe the propagation of risk in a vehicle group;

[0067] Combining the dynamic model and the first risk type, analyze the impact of the high-risk state of the high-risk vehicle on the risk level of other member vehicles in the vehicle group;

[0068] Based on the risk level, the risk propagation topology of the entire vehicle group is constructed in real time using the interaction data between all member vehicles in the vehicle group, and the propagation impact is assessed.

[0069] Considering the risk identification results and the impact of their spread, and in conjunction with the first optimization objective, a customized first risk control strategy is formulated for each of the aforementioned member vehicles through a preset control strategy generation model;

[0070] The first risk management strategy includes: for low-risk situations, the cloud server issues safety policy suggestions to the vehicle; for medium-risk situations, the cloud server sets hard constraints on the vehicle, such as a maximum speed limit; for high-risk situations, the edge computing node issues a warning, and the cloud server switches the vehicle to remote control mode.

[0071] The member vehicles provide feedback on the effectiveness of the control strategy and assign scores. The cloud server continuously optimizes the risk identification model and the control strategy generation model based on the scoring data to ensure that the vehicle group achieves higher driving efficiency and overall effectiveness under the premise of controllable risk.

[0072] Optionally, the step of dynamically adjusting the vehicle members of each vehicle group and controlling the vehicles according to the adjusted vehicle group situation includes:

[0073] The interaction and driving efficiency of the vehicle members within each vehicle group are monitored in real time. The overall collaborative efficiency of the vehicle group is evaluated and scored based on indicators such as energy consumption, driving safety, and matching degree with road conditions.

[0074] Continuously track changes in external conditions;

[0075] If the overall coordination efficiency evaluation score of the vehicle group is lower than the preset coordination efficiency evaluation threshold and / or the existing vehicle group cannot maintain the optimal coordination state under changing external conditions, the group reorganization demand detection is triggered, and a preset reorganization optimization strategy is used to determine whether reorganization is necessary.

[0076] If reorganization is required, the vehicle reorganization problem is modeled as a Markov decision process based on reinforcement learning algorithms. The goal is to obtain higher overall collaborative rewards through reorganization, output the optimal reorganization scheme, and determine the new group members and the formation sequence of each vehicle.

[0077] Control the trajectory of each vehicle within the original vehicle group to avoid collisions during the transition phase;

[0078] The new vehicle group gradually adjusts the spacing and speed of each vehicle to transition to the new formation.

[0079] The status of each vehicle in the new vehicle group and their mutual influence are analyzed and modeled. Combined with environmental and traffic factors, the risk points existing in the new vehicle group are identified, and targeted driving strategies are formulated for each vehicle to ensure that the new vehicle group achieves a state of coordinated efficiency and controllable risks.

[0080] The cloud server monitors the operation status of all vehicle groups in real time and controls and handles abnormal situations.

[0081] Collect real-time operating data of all vehicles, and continuously optimize the reorganization optimization strategy and vehicle management strategy model based on the real-time operating data;

[0082] The new reorganization and optimization strategy and the new vehicle management strategy model are distributed to the vehicle-mounted equipment and the edge computing node to guide the real-time management of the vehicle group.

[0083] Optionally, the globally optimal traffic organization strategy and vehicle platooning scheme are obtained by analyzing multi-source data on vehicles, traffic, and road conditions across the entire road network collected by the cloud server, and applying artificial intelligence algorithms for global decision optimization. Specifically, this includes the following steps:

[0084] By aggregating multi-dimensional traffic data uploaded by roadside facilities and vehicle-mounted equipment, and through data fusion, modeling, and intelligent analysis, combined with the traffic state digital twin model, a digital twin of the traffic situation of the entire road network is obtained.

[0085] Based on road network topology, traffic rules, vehicle dynamics knowledge, and the digital twin of the entire road network traffic situation, the motion, interaction, and state transition of vehicles in the road network are simulated.

[0086] During the simulation, traffic organization strategies are used as the behavioral strategy space of the reinforcement learning agent.

[0087] Design a global reward function to evaluate traffic organization strategies, and use the global reward function to guide reinforcement learning to obtain the globally optimal traffic organization strategy;

[0088] Based on the globally optimal traffic organization strategy obtained at the road network level, a multi-agent cooperative algorithm is designed for platoonable vehicle clusters to output a vehicle platooning scheme that includes platooning sequence, vehicle spacing, and vehicle speed.

[0089] The intelligent traffic management method for smart cities based on 5G, employing the technical solution of this invention, includes: acquiring planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data of a target city; constructing a digital twin model of the target city based on the planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data; configuring multiple video acquisition terminals and environmental data acquisition terminals interconnected via 5G networks on target roads within the target city area based on the digital twin model; uploading the first video data and the first environmental data collected by the video acquisition terminals and the environmental data acquisition terminals to an edge computing node via a 5G base station; and the edge computing node processing the collected data... The first video data is subjected to object detection and semantic segmentation to identify vehicles, personnel, and environmental objects, resulting in first video recognition data. The edge computing node uses a deep learning model to perform real-time predictive analysis on the first video recognition data and the first environmental data, assessing the congestion risk index and personnel safety risk index for each road segment. If the congestion risk index and / or the personnel safety risk index exceed a preset risk index threshold, the edge computing node sends a first control command to the 5G base station. The 5G base station controls the corresponding traffic lights according to the first control command to adjust traffic flow. The edge computing node uploads the congestion warning result to a blockchain node, which, after encryption and verification, pushes it to the vehicle-mounted equipment. In case of an emergency, the edge computing node uses blockchain technology to find the optimal troubleshooting solution and continuously optimizes the solution through a cross-domain federated learning model. This invention improves analysis efficiency and service security through edge computing and blockchain technologies, and enhances traffic management by integrating multidisciplinary technologies. Attached Figure Description

[0090] Figure 1 This is a schematic block diagram of a smart city intelligent traffic management system based on 5G, provided in one embodiment of the present invention.

[0091] Figure 2 This is a flowchart of a smart city intelligent traffic management method based on 5G, provided by an embodiment of the present invention. Detailed Implementation

[0092] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0093] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0094] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0095] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0096] The following reference Figures 1 to 2 This invention describes a 5G-based intelligent traffic management system and method for smart cities, provided by some embodiments of the present invention.

[0097] like Figure 1 As shown, one embodiment of the present invention provides a 5G-based intelligent traffic management system for smart cities, comprising: a cloud server, an edge computing node, a video acquisition terminal, an environmental data acquisition terminal, and a 5G base station; wherein,

[0098] The cloud server is configured as follows:

[0099] Acquire planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data of the target city;

[0100] A digital twin model of the target city is constructed based on the planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data.

[0101] According to the city digital twin model, multiple video acquisition terminals and environmental data acquisition terminals interconnected via 5G networks are configured on target roads within the target city area;

[0102] The video acquisition terminal and the environmental data acquisition terminal are respectively used to upload the first video data and the first environmental data collected by the 5G base station to the edge computing node;

[0103] The edge computing node is configured as follows:

[0104] Object detection and semantic segmentation are performed on the first video data to identify vehicles, people and environmental objects, thus obtaining the first video recognition data;

[0105] Using a deep learning model, real-time predictive analysis is performed on the first video recognition data and the first environmental data to assess the congestion risk index and personnel safety risk index of each road segment.

[0106] If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, a first control command will be sent to the 5G base station.

[0107] The 5G base station is configured to control the corresponding traffic lights according to the first control command to adjust the traffic flow.

[0108] The edge computing node is configured as follows:

[0109] The congestion warning results are uploaded to a blockchain node and then pushed to the vehicle-mounted device after being encrypted and verified.

[0110] In the event of an emergency, blockchain technology is used to find the optimal troubleshooting solution, and the solution is continuously optimized through a cross-domain federated learning model.

[0111] It should be known that, Figure 1 The block diagram of the 5G-based smart city intelligent traffic management system shown is for illustrative purposes only, and the number of modules shown does not limit the scope of protection of this invention.

[0112] Please see Figure 2 Another embodiment of the present invention provides a 5G-based intelligent traffic management method for smart cities, comprising:

[0113] Acquire planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data of the target city;

[0114] In this step, high-precision satellite remote sensing and aerial photography technologies are used to acquire real-world 3D data and image data of the target city; drones, vehicle-mounted cameras, and other equipment are used to collect real-world video data of the road network at different times; urban planning data, road network data, and basic traffic facility data are obtained through urban public databases and operator interfaces; real-time traffic video feature data of different road sections are collected through video transmission networks and physical sensors; historical traffic status data are reconstructed using traffic camera recordings, consumption records, geofencing data, and video feature data; all these raw data are uploaded to a cloud server, and data fusion processing is performed using technologies such as deep learning to construct a complete digital twin model of the target city. This solution enables the construction of multi-source, fine-grained data support for the target city, achieving digital management and control; it allows for the flexible extraction of required traffic parameters according to different scenario needs, improving work efficiency; and it enables the learning of historical traffic patterns through spatiotemporal data fusion, enhancing management decision-making.

[0115] A digital twin model of the target city is constructed based on the planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data.

[0116] In this step, 3D reconstruction technology can be used to fuse ground 3D data with aerial and satellite imagery to construct a 3D spatial grid structure for the target city. Based on road planning data, road network data, and traffic facility data, various traffic facilities and road segments are located within the 3D spatial grid structure, and attribute information is labeled. Real-world images and videos are classified and identified according to feature codes, and the locations of road surface targets are labeled. Real-time traffic flow and other parameters collected by sensors are imported into the corresponding locations in the model. Through deep learning and other methods, the labeled dataset is learned to reconstruct historical traffic characteristics. Planning data, real-world data (real-world 3D data, real-world image data, and real-world video data), and historical traffic status data are comprehensively integrated within the 3D spatial grid structure to construct a comprehensive spatiotemporal database. While preserving realistic details, it supports virtual scene modeling and replay, achieving digital simulation and obtaining a digital twin model of the city. This model simultaneously considers realism and computability, enhancing spatiotemporal data mining capabilities through deep learning technology. It can provide quantitative decision support for urban video management, tourism guidance, and emergency training, promoting the construction of a smart system.

[0117] Based on the urban digital twin model, multiple video acquisition terminals and environmental data acquisition terminals interconnected by a 5G network are configured on target roads within the target urban area;

[0118] In this step, the target road and surrounding space are identified based on the city's digital twin model to determine blind spots. Aerial photography of the space is conducted using a drone to identify the planned layout location of the terminal. Specifically, this includes: the city's digital twin model containing detailed 3D information about the city, such as roads and buildings; simulation using historical traffic data to identify target roads whose traffic conditions meet preset conditions (e.g., traffic flow exceeding a preset threshold, traffic environment complexity exceeding a preset complexity, traffic accident risk exceeding a preset risk value); the target road is identified through the digital twin model, and it is calculated whether the line of sight is obstructed by buildings, creating blind spots; the drone utilizes digital... The digital twin model navigates into the air to capture high-altitude images of the target road and surrounding area; cameras collect images from the high-altitude view, and the images are combined with the 3D location information of the digital twin to construct a complete visual layer; the original blind spots are identified, and target detection algorithms are used to identify the installation location with the best field of view coverage (i.e., the planned layout location); the location of blind spots is determined by the city's digital twin, and the drone's high-altitude patching achieves panoramic monitoring, accurately identifying the optimal video acquisition installation and deployment points, improving the deployment efficiency of physical facilities, and realizing panoramic real-time video perception. This provides visual support for applications such as transportation and public safety, thus enabling the efficient deployment of visual IoT and promoting the construction of smart cities. The solution utilizes 5G cell coverage tools to assess signal strength at planned locations, prioritizing areas with strong signals as optimal sites. At these optimal sites, multi-functional video acquisition terminals and environmental data acquisition terminals are deployed using 5G industrial control terminals, supporting both video acquisition and environmental sensing. These terminals establish low-latency transparent transmission channels with edge computing nodes via the 5G network. Edge computing nodes encode and decode the acquired video streams, enabling customized allocation of visual tasks. The environmental data acquisition terminals detect environmental parameters affecting data quality. Both terminals periodically transmit operational status data back to the management platform for timely fault handling. This solution achieves: visual blind spot completion, providing panoramic video surveillance of target roads; environmental parameter acquisition, improving video quality and analysis; and low-latency video delivery, effectively supporting smart management applications.

[0119] The video acquisition terminal and the environmental data acquisition terminal respectively upload the first video data (including road facility video data, traffic status video data, etc.) and the first environmental data (such as weather, temperature, rainfall, water accumulation, etc.) to the edge computing node via the 5G base station;

[0120] In this step, the video acquisition terminal and the environmental data acquisition terminal acquire first video data and first environmental data respectively according to the prescribed format; a low-latency transparent channel is established between the terminal and the 5G base station (or 5G micro base station) through the data relay function of a nearby 5G base station (or 5G micro base station); the 5G base station (or 5G micro base station) encodes and decodes the data stream and pre-extracts key frames from the video for encapsulation; the 5G base station (or 5G micro base station) reports the processed data and location tags to the edge computing node (a small computing node deployed on the base station side); the edge computing node uses technologies such as DPDK to efficiently process traffic and maintain an end-to-end transmission latency of less than 5ms; the edge computing node stores the acquired data in a distributed storage system according to tags for timely backup; and supports the edge computing node to perform preliminary processing of key data and filter sensitive frames for transmission to the cloud. This solution enables edge acquisition and backup of visual and perceptual data, supports 5G low-latency and high-bandwidth transmission, improves system response speed, and effectively supports intelligent transportation applications.

[0121] The edge computing node performs object detection and semantic segmentation on the first video data to identify vehicles, people and environmental objects, and obtains the first video recognition data.

[0122] In this step, the edge computing node loads the first video data and uses pre-trained object detection models such as YOLO or Mask R-CNN to detect vehicles, people, and environmental objects (such as various traffic facilities and obstacles) in the video image, labeling bounding boxes and categories. It then applies semantic segmentation models such as SegNet or DeepLab to perform semantic segmentation, dividing areas such as roads and pedestrian traffic. The detection and segmentation results are vectorized to generate first video recognition data containing object location coordinates, object type, and object name. Feature encoding is used to represent the relative positional relationships between different objects. The processing results are stored with timestamp labels. This method allows edge computing nodes to process high-quality video frames locally, while the rest are forwarded to the cloud server for processing. This method can accurately identify targets in videos in real time, providing a foundation for subsequent video analysis and management; it also achieves edge object detection, effectively reducing network traffic.

[0123] The edge computing node uses a deep learning model to perform real-time predictive analysis on the first video recognition data and the first environmental data to assess the congestion risk index and personnel safety risk index of each road segment.

[0124] If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, the edge computing node will send a first control command to the 5G base station.

[0125] The 5G base station controls the corresponding traffic lights to adjust the traffic flow according to the first control command.

[0126] The edge computing node uploads the congestion warning results to the blockchain node, and pushes them to the vehicle-mounted device after encryption and verification.

[0127] In this step, edge computing nodes analyze the predicted results, including the congestion risk index for each road segment, to obtain congestion warning results. These warning results are then digitally signed to generate transaction proposals. The transaction proposals are uploaded to verification nodes in the ledger network for verification. Verification nodes check the legality of the transaction content and signature. If a unanimous vote is passed, the congestion warning results are stored as an immutable blockchain record. Onboard devices periodically synchronize the latest warning results from the blockchain network. The onboard devices decrypt and parse the warning results using their private keys to obtain congestion alerts. Drivers then choose alternative or detour routes based on the alerts. This method leverages the immutability and distributed ledger advantages of blockchain to achieve reliable transmission of warnings, improving the timeliness and reliability of congestion warnings, providing decision support for intelligent driving, and effectively preventing data disruption and unauthorized access.

[0128] In the event of an emergency, the edge computing node uses blockchain technology to find the optimal troubleshooting solution and continuously optimizes the solution through a cross-domain federated learning model.

[0129] In this step, edge computing nodes, based on the first video recognition data and the first environmental data, utilize a pre-defined event recognition model to identify emergency events and locate their positions. A consortium blockchain based on a chain-like mechanism is constructed, inviting relevant rescue departments as verification bodies. Based on deep learning, historical cases are mined to provide preliminary troubleshooting solutions as transactions. Each verification body evaluates the solutions based on its expertise and records its opinions through the blockchain, selecting the optimal solution. Federated learning is used for collaborative training, continuously iterating and optimizing while ensuring privacy. The optimized solution participates in subsequent event processing as an immutable transaction record. New data is then subjected to further federated learning to improve troubleshooting capabilities. This method fully utilizes the expertise of each entity, achieving open collaboration through blockchain and federated learning; it integrates the latest resources in real time to establish a smart emergency troubleshooting system; and it effectively meets the rescue requirements in different scenarios.

[0130] In this step, the event recognition model is constructed using the following methods: acquiring a large amount of positive and negative sample data (including video footage of various emergency scenarios and labeled data of non-emergency event footage), extracting effective features (e.g., extracting key targets such as humans and vehicles through object detection; extracting target action features through pose recognition; extracting image features through image descriptors, etc.); selecting a suitable model structure (a CNN+LSTM structure can be used to recognize time-series data features, or a transfer-based neural network model can be used); selecting a suitable loss function (e.g., cross-entropy for classification tasks); using data augmentation techniques to expand the sample (e.g., image transformation, cropping, rotation, flipping, etc. to expand data capacity); training a deep learning model (e.g., using SGD or Adam optimizers for iterative training) to obtain a classifier model; calculating classification precision and recall using a validation set to find the optimal model and hyperparameters; optimizing model deployment (achieving real-time inference requirements through quantization and other means); collecting new data for online learning and model updates to optimize recognition capabilities, thereby constructing a powerful event classification and recognition model.

[0131] This invention improves analysis efficiency and service security through edge computing and blockchain technology, and enhances traffic management by integrating multidisciplinary technologies.

[0132] In some possible embodiments of the present invention, the step of the edge computing node using a deep learning model to perform real-time predictive analysis on the first video recognition data and the first environmental data, and assessing the congestion risk index and personnel safety risk index of each road segment, includes:

[0133] The edge computing node acquires the traffic facility data, the historical traffic status data, and the historical road video data;

[0134] The historical road video data is aggregated using a time-series database to extract characteristics of vehicle and pedestrian traffic changes;

[0135] In this step, the time-series database is used to store and query sequential data (e.g., data that changes over time). The process of statistically aggregating time-series data such as vehicle or pedestrian traffic from historically collected video data is called aggregation. Specifically, this involves statistically analyzing the number of vehicles or people detected in each historical video frame to obtain the growth trend of vehicle or pedestrian traffic over different time periods (e.g., every 5 minutes). Then, time-series analysis is performed on the statistical results from different time periods to extract the periodic patterns of vehicle and pedestrian traffic over time (e.g., peak hours on weekdays, peak periods on holidays, etc.). These features will be used as input to a deep learning model to train the model to learn the historical patterns of traffic and vehicle / pedestrian traffic, thereby achieving better traffic flow prediction. In short, it involves statistically summarizing a large amount of raw video data to obtain the characteristic patterns of traffic parameters changing over time.

[0136] By using a deep learning model and combining it with the feature sequences extracted from the changes in traffic flow and pedestrian flow, a traffic state prediction model (including traffic space-time correspondence patterns) is obtained.

[0137] In this step, the traffic flow and pedestrian flow change features (traffic features related to time and region) extracted from the collected historical video data are formed into a multi-dimensional time-series feature sequence, which is then combined with a deep learning model to obtain a traffic state prediction model.

[0138] The specific implementation steps are as follows: Target detection and tracking are performed on video data, extracting target count features for each region and time period; features from adjacent regions and time intervals are combined to form a multi-dimensional input sequence; a convolutional LSTM structure is applied for sequence learning, with the convolutional kernel extracting spatial-temporal correlations; the model learns the temporal and regional dependencies in the feature sequence to construct traffic pattern models; new data is added online to continuously update and optimize the model. This solution utilizes convolutional neural networks to extract spatial feature relationships from sequences, LSTM to encode temporal dependency information, and learns the distribution patterns of traffic volume in different regions and times, providing a periodic reference for subsequent predictions and improving traffic management; it achieves traffic spatiotemporal pattern recognition based on deep learning, supporting better decision-making.

[0139] Input the first video recognition data and the first environmental data into the traffic state prediction model to predict the traffic distribution in future time periods;

[0140] Based on the traffic flow distribution prediction results, the traffic flow difficulty coefficient of each road segment is assessed as a congestion risk index.

[0141] Real-time object detection algorithms (such as YOLO) and human pose estimation algorithms are used to analyze the behavior of pedestrian flow and identify abnormal events;

[0142] Based on the traffic distribution prediction results, the abnormal events, and the environmental data, the personnel safety risk index of different areas is assessed.

[0143] In this embodiment, convolutional LSTM and other models are used to predict historical pedestrian flow data, obtaining the predicted pedestrian density distribution for each area in a future time period. An environmental quality weight is calculated for each area based on environmental data such as light and temperature. The predicted pedestrian density and environmental quality weight are comprehensively evaluated for each area; areas with higher density and worse environmental conditions have higher difficulty coefficients. For example, a linear combination formula can be used to weight and sum the two factors as the safety difficulty coefficient for the area. Alternatively, a deep learning model can be used, inputting two types of features into the network for end-to-end learning and outputting a difficulty score. The safety difficulty coefficient results for each area are visualized to provide a reference for safety management. This enables regional risk assessment based on multi-source dynamic data, effectively identifying key areas with high potential hazards. It provides decision support for safety emergency response, preventing and reducing accident risks.

[0144] In this embodiment, deep learning is used to vectorize video data, extract spatial and temporal features, comprehensively assess congestion safety risks, and provide decision support for traffic optimization.

[0145] In some possible embodiments of the present invention, the step of sending a first control command to the 5G base station if the congestion risk index and / or the personnel safety risk index exceed a preset risk index threshold includes:

[0146] The edge computing node compares the congestion risk index and / or the personnel safety risk index with a preset risk index threshold.

[0147] If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, a risk event is determined to have occurred.

[0148] The edge computing node performs risk event identification and analysis on the first video recognition data and the first environmental data to determine the type and status (including location) of the first risk event that has occurred.

[0149] Generate a corresponding first control instruction based on the type and status of the first risk event;

[0150] The edge computing node sends the first control command to the corresponding 5G base station.

[0151] In this embodiment, risk thresholds for certain key areas are pre-set in the system, such as a congestion index threshold of 3 and a safety index threshold of 2. Edge computing nodes calculate the real-time congestion index of a certain area as 3.5 and the safety index as 1.8. Comparing the calculated results with the thresholds, it is found that the congestion index exceeds the threshold of 3, indicating that a congestion event may be occurring or is about to occur. Event assessment is performed, and further response measures are taken, such as pushing out early warnings. By setting quantitative thresholds, the criteria for determining event occurrence can be clearly defined; real-time monitoring of whether the threshold is exceeded allows for faster detection of risk events; effective triggering of subsequent emergency measures; and improved system response dynamics and periodicity.

[0152] In this embodiment, it may also include: the 5G base station management subsystem parsing instructions to determine that the controlled object is the video acquisition terminal; the 5G base station control layer adjusting the parameters of the video acquisition terminal, such as viewing angle scaling and reserved bandwidth; the video acquisition terminal actively reporting the video stream of the event area, and the 5G base station prioritizing sending it to the edge node; the edge node sending the high-definition video portion to the cloud as needed to build a cloud-edge collaborative processing system; and timely carrying out video analysis and annotation work to thoroughly investigate the cause of the event.

[0153] In this embodiment, the first control instruction may include the following main contents: control object type (such as video capture terminal, traffic light, etc.), control object instance (if the control object is a video capture terminal, a specific terminal number needs to be specified; if the control object is a traffic light, unique identification information such as traffic light coordinates or number needs to be specified), control action (such as specifying adjustment of viewing angle range, clarity, etc. for the video capture terminal; specifying a change of state to a specific cycle or mode for the traffic light), control time (start time and duration of the control action), auxiliary information (such as control area range, predicted traffic flow change trend, priority control object, etc.), security information (instruction sequence number, instruction signature, etc., security protection information to prevent unauthorized access or data tampering), and other information (such as control termination conditions, status feedback frequency, etc., supplementary parameters). Specifically, in a scenario, the content of the first control instruction may be: the control object is traffic intersection A, the control action is set to a 60-second yellow light state, the control start time is time T, and the area range is block X, etc.

[0154] In this embodiment, 5G sensing enhancement and video edge collaboration effectively support emergency response in complex environments, improve event handling efficiency, realize dynamic allocation of video resources, and help save edge cloud resources.

[0155] In some possible embodiments of the present invention, the step of the 5G base station controlling the corresponding traffic lights to adjust traffic flow according to the first control command includes:

[0156] The 5G base station parses the first control command and identifies the controlled object as a traffic light.

[0157] The 5G base station establishes a low-power connection with the traffic signal management system.

[0158] The 5G base station sends a control signal to the accessed traffic lights, requesting them to switch their working mode to network control;

[0159] The first control command of the 5G base station sequentially controls one or more traffic lights to work together to adjust traffic flow.

[0160] The traffic signal management system transmits real-time status data back to the 5G base station.

[0161] This implementation scheme enables 5G-based joint control of traffic signals, allowing for real-time response and optimization of traffic conditions, alleviating congestion in key areas, reducing delays, and improving road utilization efficiency.

[0162] In some possible embodiments of the present invention, the 5G-based smart city intelligent traffic management method further includes:

[0163] The on-board device uploads the first vehicle data of the first vehicle it is in to the corresponding edge computing node;

[0164] In this step, the onboard device establishes a communication connection with the roadside edge computing node through an onboard communication device (such as an OBU); the onboard device acquires first vehicle data in real time; the first vehicle data is packetized in JSON / XML format; the first vehicle data is uploaded to the topic subscribed by the roadside edge computing node via a protocol (such as MQTT); the edge computing node confirms the data receipt. This solution utilizes real-time uploading of first-hand data from vehicles, which helps in traffic status perception; edge computing can acquire vehicle dynamics faster, expanding the traditional perception range; high-precision traffic topology maps are updated in real time, supporting intelligent transportation applications; vehicle and infrastructure data interaction constructs a vehicle-environment closed loop, realizing collaborative intelligent transportation, optimizing road network planning and guidance, thereby promoting the development of intelligent transportation applications towards refinement.

[0165] In this embodiment, the first vehicle data includes, but is not limited to: location and motion data (such as precise GPS location, heading angle, vehicle speed, acceleration, lateral / longitudinal acceleration, heading angular velocity, braking signal, etc.), vehicle energy and power parameters (such as battery charge, voltage, current, temperature, etc.; engine speed; energy recovery and consumption, etc.), chassis and body status (such as tire pressure, suspension height, vehicle stability, anti-lock braking system status, etc.), driving operation data (such as steering wheel angle, accelerator / brake pedal travel, gear shifting status, headlights, turn signals, etc.), vehicle systems and fault diagnosis (various ECU system status, fault codes, maintenance information, etc.), in-vehicle environment data (such as in-vehicle temperature, humidity; wiper and heating system status, etc.), ADAS and driver assistance systems (such as active cruise control system status, automatic parking assist system status, lane departure warning and forward collision warning, etc.), as well as vehicle brand, model, function, performance parameters, technical parameters of various components, etc. The aforementioned vehicle data is collected in real time through various onboard sensors, controllers, and diagnostic systems, and interacts and is analyzed in real time with roadside facilities and the cloud via the vehicle-to-everything (V2X) network. This data forms a crucial foundation for the perception, decision-making, and control of intelligent connected vehicles. By modeling and analyzing this multi-dimensional vehicle data, the current operating status of vehicles can be assessed, future dynamic behaviors can be predicted, and various advanced functions can be supported for V2X systems.

[0166] The edge computing node parses the first traffic data, the first road status data, and the first road facility data from the first video recognition data, and uploads the first traffic data, the first road status data, the first road facility data, the first environmental data, and the first vehicle data to the cloud server;

[0167] In this step, the edge computing node extracts first traffic data (such as vehicle flow, pedestrian flow, etc.), first road status data (such as the number of lanes, road surface damage, obstacle conditions, etc.), and first road facility data (traffic lights, traffic control edge terminals, traffic signs, etc.) from the first video recognition data. The edge computing node also acquires first environmental data (weather, etc.) and first vehicle data (real-time reporting or historical data). All of these data types are then organized and packaged using a distributed file system storage format. Using 5G underlying communication technology (such as MQTT) accessed by the edge computing node, the organized data is reported to the cloud server in real time. This solution enables the fusion and utilization of video and other data, improving perception utilization; reduces the amount of data transmitted to the cloud through edge preprocessing, improving upload efficiency; allows the first raw data to enter the cloud in real time, providing a foundation for subsequent analysis; realizes a collaborative working mode between the edge and the cloud, fully leveraging the advantages of both; and supports intelligent traffic refined management and optimization decision-making based on big data, thereby constructing a perception support system for intelligent traffic management.

[0168] The cloud server analyzes the first traffic data, the first road status data, the first road facility data, the first environmental data, and the first vehicle data, and combines them with a preset global optimal traffic organization strategy and vehicle platooning scheme to divide all vehicles traveling on the road (on the same road or on different roads with intersecting conditions) into multiple vehicle groups.

[0169] For each vehicle group, the first driving status data of each member vehicle in each group, the first mutual influence relationship data between each member vehicle, and the first current environmental condition data and the first current road condition data of each member vehicle are obtained.

[0170] Based on the first driving status data, the first mutual influence relationship data, the first current environment data, and the first current road condition data, the first risk existing in each member vehicle of each vehicle group is analyzed, and the control of each member vehicle in the vehicle group is carried out according to the first risk.

[0171] The vehicle members of each vehicle group are dynamically adjusted, and the vehicles are managed according to the adjusted vehicle group situation.

[0172] In some possible embodiments of the present invention, the cloud server, by analyzing the first traffic data, the first road condition data, the first road facility data, the first environmental data, and the first vehicle data, and combining a preset globally optimal traffic organization strategy and vehicle platooning scheme, divides all vehicles traveling on the road into multiple vehicle groups, including:

[0173] A multi-source heterogeneous data integration platform is built on the cloud server;

[0174] This step involves designing unified interface specifications and data format standards on cloud servers; developing access components adapted to different data sources to connect with various edge devices and vehicle terminals; building a big data processing and storage architecture based on distributed technology; implementing collection components for various heterogeneous data, supporting real-time import of vehicle dynamic data, monitoring video data, etc.; developing a data integration engine to clean, transform, and deduplicate the collected multi-source data; establishing cross-data source relationship connections to support data aggregation queries at different levels and perspectives; implementing an elastically scalable cloud-native computing framework to support massive concurrent data processing tasks; and developing high-performance backup and recovery modules to ensure the high availability and fault tolerance of the data integration platform. Through these implementations, the following technical effects can be achieved: seamless connection of various vehicle-to-everything (V2X) devices, resolving data interaction barriers; collecting and integrating massive amounts of traffic-related data to build an integrated traffic big data database; improving data utilization efficiency and supporting efficient query and analysis applications; providing elastically scalable large-scale data processing capabilities to respond to complex application scenarios; and establishing a highly available integrated service platform to meet stable and reliable technical operation requirements.

[0175] The multi-source heterogeneous data integration platform cleans, fuses, and encodes the received first traffic data, first road status data, first road facility data, first environmental data, and first vehicle data to obtain first comprehensive data.

[0176] In this step, the data collected from various data sources undergoes format normalization and conversion to eliminate non-standard data; incomplete or erroneous data is supplemented or corrected to obtain a quality-assured dataset; based on standard models and labels, the original data at different semantic levels are mapped to a unified semantic space; database internal and external key relationships and other technologies are used to achieve the association, connection, and fusion of multi-source data; while retaining the original attributes, core feature data is mined and enriched to obtain more detailed data descriptions; the fused data is fine-grainedly divided and categorized according to dimensions such as time, space, and attributes; and the fused data is structured for storage management, enabling unified query and analysis processing. This implementation step can achieve the following technical effects: further improve the quality and completeness of multi-source data; construct a rich first-transportation knowledge graph system; provide clear and unified first-transportation data as a foundation for decision support; effectively enhance the application value of first-transportation big data, providing a foundation for in-depth analysis and mining.

[0177] A traffic status digital twin model is generated based on the first comprehensive data and the city digital twin model (the traffic status digital twin model includes at least the road, speed, and destination of the first vehicle, and can reflect the traffic flow, congestion, and road facility status of each road segment in real time, and can accurately predict future traffic trends).

[0178] In this step, a dynamic traffic subsystem is established based on a digital twin model of the urban physical scene, imported with the first comprehensive data. Mathematical models describing the road network and vehicles are generated according to vehicle and road attributes. Discrete event traffic flow is sampled to establish probabilistic models for vehicle generation and demise. A transmission model of the impact of traffic environmental variables on traffic flow is established, such as the effect of road conditions on speed. Historical data is replayed through simulation to train and verify the interactions between the sub-models. Real-time data is acquired online to correct the model state and predict outputs. This implementation step can bring the following technical effects: realistically reproducing the dynamic interaction between traffic entities and the environment; providing a panoramic view of all-weather traffic operation status; effectively assessing the impact of different decision-making schemes on traffic and issuing intelligent decisions; faster response to abnormal events and prediction of future trends to support optimization; and flexible model updates to continuously improve intelligent traffic management capabilities.

[0179] Based on the traffic state digital twin model, and combined with the preset global optimal traffic organization strategy and vehicle platooning scheme, all the first vehicles traveling on the road will be divided into multiple vehicle groups.

[0180] In this step, based on factors such as the current state, destination, road location, and distance between vehicles of the first vehicle determined from the traffic state digital twin model, and combined with the preset global optimal traffic organization strategy and vehicle platooning scheme, all the first vehicles are divided into multiple reasonable vehicle groups, and the grouping flag and the membership flag of each first vehicle in the vehicle group are sent to the corresponding first vehicles.

[0181] In this embodiment, the method further includes: using road network traffic as a reinforcement learning environment and vehicles as intelligent agents; the decision-making behavior of the intelligent agents is various traffic organization and control strategies; obtaining the globally optimal traffic organization strategy through simulation and online interactive data training; identifying platoonable vehicle clusters based on road network traffic and other data; and calculating the optimal vehicle platooning scheme (including platooning sequence, spacing, speed, etc.) for each cluster by combining objectives such as dynamically adjusting platooning composition, real-time obstacle avoidance, and minimizing energy consumption.

[0182] In this embodiment, the method also includes: distributing vehicle grouping schemes through vehicle-cloud communication channels; coordinating and controlling the driving strategy of the first vehicle in each group; and quickly grouping and platooning newly entering vehicles to achieve vehicle group collaboration and efficient and orderly travel.

[0183] In this embodiment, the cloud server can analyze and optimize real-time traffic conditions based on multi-source data perception and artificial intelligence algorithms. Then, it can rationally group all electric vehicles and intelligently schedule the entire process of each group's journey, ultimately achieving efficient collaborative driving across the entire road network. This intelligent multi-vehicle collaborative paradigm can significantly improve traffic efficiency, driving safety, and energy utilization.

[0184] In some possible embodiments of the present invention, the step of acquiring, on a per-vehicle-group basis, first driving status data of each member vehicle in each group, first mutual influence relationship data between each member vehicle, and first current environmental condition data and first current road condition data of each member vehicle includes:

[0185] Each of the aforementioned member vehicles collects its own first driving status data (including speed, acceleration, position coordinates, etc.) through its own configured on-board equipment;

[0186] Understandably, each vehicle is equipped with multiple multi-source heterogeneous onboard devices (such as LiDAR, millimeter-wave radar, high-definition vision cameras, etc.); these onboard devices collect the vehicle's own status data (such as speed, acceleration, battery level, braking status, etc.) and sense external environmental data (road conditions, obstacles, positions of other vehicles, distances between vehicles, etc.). Furthermore, the initial driving status data of each member vehicle in the vehicle group can be analyzed from the data of the first vehicle.

[0187] Each of the aforementioned member vehicles establishes a communication connection with other vehicles in the same vehicle group through the on-board equipment (e.g., via V2V communication) and exchanges first mutual influence relationship data (e.g., vehicle distance, relative speed, etc.).

[0188] In this step, each member vehicle establishes a connection with other member vehicles around it through a V2V-compliant wireless communication device installed on the vehicle, enabling information exchange. Onboard sensors (vehicle-mounted equipment) obtain data in real time (such as a separately equipped millimeter-wave radar that can sense and track the position data of surrounding vehicles), thereby acquiring first mutual influence data such as vehicle distance and relative speed. The onboard positioning system provides location tracking data (e.g., a vehicle positioning module based on a BD system or GPS), which can acquire and upload spatiotemporal coordinate data of the vehicle and other surrounding vehicles in real time. The vehicle computing module (which can be integrated into the onboard equipment) comprehensively processes various types of first raw data, integrating and processing the above-mentioned first position, speed, and other raw data to calculate derived data expressing the first mutual influence (such as possible collision risk values). Furthermore, through vehicle-to-cloud interconnection, an overall traffic flow map and first real-time video of a certain area can be obtained from the cloud to assist in analyzing the first influence relationship between vehicles. Therefore, in summary, the first mutual influence data between the two vehicles during movement is mainly obtained through onboard active perception, V2V communication, and first environmental image data obtained through interconnection.

[0189] Each of the aforementioned member vehicles collects first current environmental data (such as weather conditions, road surface conditions, etc.) and first current road condition data through the on-board equipment.

[0190] In this step, the first current environment data and / or the first current road condition data can be collected by the vehicle-mounted device itself, or the first current environment data and / or the first current road condition data can be collected from the edge computing node by the video acquisition terminal and / or the environmental data acquisition terminal.

[0191] In this embodiment, the above steps enable real-time and comprehensive identification of the intrinsic relationships and external environmental influences of each member vehicle within each vehicle group. This mainly achieves the following technical effects: improving the coordination between vehicles within the group; enhancing the ability to identify environmental influences within the group; improving the accuracy and efficiency of multi-vehicle collaborative control within the group; and reducing the risks and uncertainties in the operation of the group.

[0192] In some possible embodiments of the present invention, the step of analyzing the first risk existing in each member vehicle of each vehicle group based on the first driving state data, the first mutual influence relationship data, the first current environment data, and the first current road condition data, and controlling each member vehicle of the vehicle group according to the first risk, includes:

[0193] Acquire historical vehicle driving status data, historical environmental data, historical road data, and historical accident data that have spatiotemporal correspondence;

[0194] In this step, historical data is queried and matched from a multi-source heterogeneous data integration platform; data sources are sorted and arranged according to time labels in the data attributes; spatial overlap is achieved using techniques such as raster indexing based on coordinate labels; consistency between time and location of data points is detected, and matching relationships are labeled; a structured link network is constructed through a relational database; algorithms such as EM and HMM are applied to learn implicit relationships between datasets; complex attribute dependencies in training samples are mined; and time series data and individual relationship data are integrated and unified. This implementation step can achieve the following technical effects: reconstructing the temporal feature connectivity of real historical events, improving the quality and representativeness of learning samples, mining deep implicit interaction patterns in the data, effectively supporting the predictive modeling of complex influencing factors, and improving the deep learning capabilities of data mining and intelligent decision-making.

[0195] Big data analytics is used to analyze historical vehicle driving status data, historical environmental data, historical road data, and historical accident data. Based on the analysis results, a multi-dimensional risk assessment index system is constructed, which includes vehicle-specific factors (such as battery level, malfunction, and power range), road factors (such as the number of lanes, road surface smoothness, and slope), environmental factors (such as temperature, rainfall, and water accumulation), and inter-vehicle interaction factors (such as vehicle distance and relative speed).

[0196] In this step, latent semantic analysis and association rules are applied to mine the dependency relationship between vehicle attributes and accidents; decision tree algorithms are used to identify the relative importance and key features of influencing factors; a preliminary multi-factor risk model is constructed using neural networks; deep learning is used to continuously optimize the model with the support of big data; statistical learning algorithms are used to train the measurement standards of each indicator in the risk system; and the accuracy and stability of the model are verified based on sufficient samples. This implementation step can achieve the following technical effects: identify the key attribute dimensions affecting the primary risk; establish a quantitative multi-factor risk assessment system guideline; provide risk reference standards for decision support; and support more scientific risk warning and control decisions.

[0197] By using the historical vehicle driving status data, the historical environmental data, the historical road data, the historical accident data, and the multi-dimensional risk assessment index system, a neural network is trained to learn the mapping relationship between risk assessment indicators and risk levels, thereby obtaining a risk identification model.

[0198] In this step, historical data is labeled according to the indicator system to construct input and output sample datasets; a deep learning model is designed using architectures such as convolutional neural networks or recurrent neural networks; sample feature vectors and accident labels are used as the model's input and output; the training process begins, optimizing the parameters of fully connected layers and pooling layers; L1 / L2 regularization is added to avoid overfitting, and Dropout technology is used to improve generalization ability; model performance is evaluated in real time on the validation set, and EarlyStopping is used to prevent overfitting; once the loss function converges, the model is saved as a risk identification model. This implementation step can achieve the following technical effects: learning the patterns of complex data, establishing input-output mapping relationships, and improving the accuracy and robustness of risk identification; identifying the risk level of new and difficult samples, achieving quantitative prediction of risk factors and their severity, and providing threshold judgment and priority basis for subsequent decision-making.

[0199] The first driving status data, the first mutual influence relationship data, the first current environment data, and the first current road condition data are input into the trained risk identification model to obtain the risk score and risk type of the first risk.

[0200] In this step, the collected first real-time dataset is preprocessed and standardized. The standardized data is used as the model input feature vector. During model training, the influence weights of features and risks are learned. The input feature vector is then imported into the trained model, which performs forward computation and outputs a risk score. Based on the score and a preset threshold, the risk range is divided, and specific risk types are identified from typical risk factors. This implementation step achieves the following technical effects: Quantitative assessment of the first specific risk, identification of the first high-risk object and its specific causes, providing a basis for subsequent refined management and control, supporting real-time risk monitoring and early warning decisions, and ensuring the controllability of traffic safety operations. Through model identification, the first risk is assessed and key areas are selected to support efficient decision-making at each stage.

[0201] For high-risk vehicles whose risk scores exceed a preset threshold, determine the first risk type of the first risk (such as insufficient energy, too close to the vehicle in front, etc.).

[0202] In this step, a weighted standard for feature values ​​corresponding to different risk types is set. The feature values ​​of high-risk vehicles are converted into scores, and each feature score is normalized. The main risk features are identified by ranking them according to their contribution. Based on the main features, a pre-defined first risk type is matched. Simultaneously, secondary features are analyzed to eliminate possible misjudgments, and the first specific risk type result is output. This implementation step achieves the following technical effects: accurately identifying the direct cause of the first risk, developing targeted risk control strategies, providing personalized first warnings and alerts, avoiding errors caused by single-feature judgments, and laying a technical foundation for formulating precise control instructions. The above process efficiently locates the source of the first risk, supporting risk management decisions.

[0203] Construct a dynamic model to describe the propagation of risk in a vehicle group;

[0204] In this step, a network topology consisting of vehicle nodes and edge connections is defined, the state transition rules for risk diffusion between nodes are described, a state space and transition matrix for node risk levels are established, the influence of node attributes and connection attributes on transition probabilities is considered, a risk output dynamic function for pollution source nodes is introduced, the propagation equation is applied to describe the overall risk concentration change pattern, and hybrid learning is used to correct the propagation rate and diffusion interval parameters, thereby realizing the dynamic simulation of the risk propagation process and results. This model can: reflect the real-time propagation process of risk between nodes, consider the influence of network topology and node attributes on propagation, quantitatively predict the risk diffusion rate and final impact range; dynamically investigate propagation parameters to optimize the understanding of risk propagation mechanisms; provide decision-making basis for risk control, such as isolating key areas; and the construction of this model is of great significance for analyzing and preventing risk propagation.

[0205] Combining the dynamic model and the first risk type, analyze the impact of the high-risk state of the high-risk vehicle on the risk level of other member vehicles in the vehicle group;

[0206] In this step, each vehicle in the group is simulated as a network node in a dynamic model. The primary risk type of high-risk vehicles is identified as the input pollution source. Based on the characteristics of the pollution source, its risk output intensity function is determined. The propagation path and probability of other member vehicles receiving the risk are calculated, and the various levels of risk states they may be in through propagation are estimated. The final risk level is adjusted by considering the characteristics of each member vehicle, and the potential risk change patterns and impact depth of each vehicle in the group are output. This yields the following technical effects: determining the scope and depth of the primary risk diffusion; identifying potential secondary risks within the group; providing forward-looking decision-making references for risk management; clearly demonstrating the impact of collaborative relationships on risk; and enhancing the group's prevention and response capabilities.

[0207] Based on the risk level, the risk propagation topology of the entire vehicle group is constructed in real time using the interaction data between all member vehicles in the vehicle group, and the propagation impact is assessed.

[0208] In this step, each vehicle within the group is constructed as a node in the network topology graph. The connection relationships of edges are determined based on the relative motion states of each workshop. Capacity is set for the connecting edges to describe the degree of interaction between workshops. The risk level of each vehicle is mapped to the degree distribution of the nodes. A risk dynamics model is applied to update the topology structure. The connectivity between nodes and the aggregation of subgraphs are analyzed to derive the propagation path. Simulation is used to determine the depth and scope of risk diffusion. This implementation step can achieve the following technical effects: dynamically reconstructing the risk propagation network topology within the group, assessing the relative impact of each stage of risk diffusion, providing refined decision-making reference support, revealing the evolutionary laws of risk propagation within the group, and planning refined prevention and control strategy deployment.

[0209] Considering the risk identification results and propagation impact, and combining the first optimization objective (such as taking the safe driving evaluation value, driving efficiency value, energy consumption control range, etc. of the vehicle group as optimization objectives), a customized first risk control strategy is formulated for each member vehicle through a preset control strategy generation model.

[0210] In this step, the objective function and decision variable space representation problem are defined, and a set of control strategies (such as speed limits, distance adjustments, and route adjustments) is prepared. A genetic algorithm or neuroevolutionary algorithm is applied to search the optimization solution space, generating a set of customized control strategy schemes for each member vehicle. The impact of each scheme on the optimization objective is evaluated through digital twin simulation. Based on the evaluation results, the optimal control strategy for each member vehicle is selected, and the customized first-risk control strategy scheme is output. This implementation step can achieve the following technical effects: maximizing the match between control measures and optimization objectives; providing personalized optimized control schemes to support decision-making; fully utilizing the advantages of team collaboration to reduce overall risk; ensuring the safe operation of each vehicle while improving team efficiency; and guiding intelligent collaboration to be achievable.

[0211] The first risk management strategy includes: for low-risk situations, the cloud server issues safety policy suggestions to the vehicle; for medium-risk situations, the cloud server sets hard constraints on the vehicle, such as a maximum speed limit; for high-risk situations, the edge computing node issues a warning, and the cloud server switches the vehicle to remote control mode.

[0212] The member vehicles provide feedback on the effectiveness of the control strategy and assign scores. The cloud server continuously optimizes the risk identification model and the control strategy generation model based on the scoring data to ensure that the vehicle group achieves higher driving efficiency and overall effectiveness under the premise of controllable risk.

[0213] In this embodiment, the above technical approach enables real-time risk monitoring of vehicles, identification of risk sources, analysis of propagation impact, and hierarchical and refined management, thereby minimizing the occurrence of dangerous situations in the vehicle group and ensuring efficient, safe, and reliable collaborative operation within the vehicle group.

[0214] In some possible embodiments of the present invention, the step of dynamically adjusting the vehicle members of each vehicle group and controlling the vehicles according to the adjusted vehicle group situation includes:

[0215] The interaction and driving efficiency of the vehicle members within each vehicle group are monitored in real time. The overall collaborative efficiency of the vehicle group is evaluated and scored based on indicators such as energy consumption, driving safety, and matching degree with road conditions.

[0216] Continuously track changes in external conditions (such as new vehicles joining, vehicles leaving midway, changes in road conditions, etc.);

[0217] If the overall coordination efficiency evaluation score of the vehicle group is lower than the preset coordination efficiency evaluation threshold and / or the existing vehicle group cannot maintain the optimal coordination state under changing external conditions, the group reorganization demand detection is triggered, and a preset reorganization optimization strategy is used to determine whether reorganization is necessary.

[0218] If reorganization is required, the vehicle reorganization problem is modeled as a Markov decision process based on reinforcement learning algorithms. The goal is to obtain higher overall collaborative rewards (such as efficiency, safety, and low energy consumption) through reorganization, output the optimal reorganization scheme, and determine the new group members and the formation sequence of each vehicle.

[0219] Through V2V and V2I communication coordination, the driving trajectory of each vehicle in the original vehicle group is controlled to avoid collisions during the transition phase; the spacing and speed of each vehicle in the new vehicle group are gradually adjusted to transition to the new formation state.

[0220] The status of each vehicle in the new vehicle group and their mutual influence are analyzed and modeled. Combined with environmental and traffic factors, the risk points existing in the new vehicle group are identified. Targeted driving strategies (such as speed, acceleration, distance, obstacle avoidance, etc.) are formulated for each vehicle to ensure that the new vehicle group achieves a coordinated, efficient and risk-controllable state.

[0221] The cloud server monitors the operation status of all vehicle groups in real time and controls and handles abnormal situations.

[0222] Collect real-time operating data of all vehicles, and continuously optimize the reorganization optimization strategy and vehicle management strategy model based on the real-time operating data;

[0223] The new reorganization and optimization strategy and the new vehicle management strategy model are distributed to the vehicle-mounted equipment and the edge computing node to guide the real-time management of the vehicle group.

[0224] In this embodiment, vehicle groups can be dynamically reorganized to respond to changes in the traffic environment, and the driving strategies of each vehicle can be adjusted in real time based on the new group status, thereby ensuring that the entire vehicle group system always operates in an optimal state of collaborative efficiency and controllable risk. Simultaneously, through continuous model optimization, the intelligence level of reorganization and control decisions is continuously improved.

[0225] In some possible embodiments of the present invention, the globally optimal traffic organization strategy and vehicle platooning scheme are obtained by analyzing multi-source data on vehicles, traffic, and road conditions across the entire road network collected by the cloud server, and applying artificial intelligence algorithms for global decision optimization, specifically including the following steps:

[0226] By aggregating multi-dimensional traffic data (including real-time vehicle location, speed, road condition information, roadside facility status, etc.) uploaded by roadside facilities and vehicle-mounted equipment, and through data fusion, modeling and intelligent analysis, combined with the traffic state digital twin model, a digital twin of the traffic situation of the entire road network is obtained.

[0227] Based on road network topology, traffic rules, vehicle dynamics knowledge, and the digital twin of the entire road network traffic situation, the motion, interaction, and state transition of vehicles in the road network are simulated.

[0228] During the simulation, traffic organization strategies are used as the behavioral strategy space of the reinforcement learning agent.

[0229] Design a global reward function (including multiple objective indicators such as road network efficiency, congestion relief, and vehicle energy consumption) to evaluate traffic organization strategies, and use the global reward function to guide reinforcement learning to obtain the globally optimal traffic organization strategy;

[0230] Based on the globally optimal traffic organization strategy obtained at the road network level, a multi-agent cooperative algorithm is designed for platoonable vehicle clusters to output a vehicle platooning scheme that includes platooning sequence, vehicle spacing, and vehicle speed.

[0231] In this embodiment, the cloud utilizes artificial intelligence technology to perform intelligent analysis and decision-making on massive heterogeneous global traffic data, which can obtain the optimal traffic organization strategy and vehicle platooning scheme at the road network level. The strategy is continuously improved through digital twin closed-loop testing, ultimately forming a highly intelligent integrated traffic scheduling and organization scheme.

[0232] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0233] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0234] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0235] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0236] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0237] If the integrated units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0238] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0239] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0240] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can easily conceive of variations or substitutions without departing from the spirit and scope of the present invention, and various modifications and alterations can be made, including combinations of the different functions and implementation steps described above, as well as software and hardware implementation methods, all of which are within the protection scope of the present invention.

Claims

1. A 5G-based intelligent traffic management system for smart cities, characterized in that, include: Cloud servers, edge computing nodes, video capture terminals, environmental data capture terminals, and 5G base stations; among them, The cloud server is configured as follows: Acquire planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data of the target city; A digital twin model of the target city is constructed based on the planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data. Based on the city digital twin model, multiple video acquisition terminals and environmental data acquisition terminals interconnected via 5G networks are configured on target roads within the target city area; The video acquisition terminal and the environmental data acquisition terminal are respectively used to upload the first video data and the first environmental data collected by the 5G base station to the edge computing node; The edge computing node is configured as follows: Object detection and semantic segmentation are performed on the first video data to identify vehicles, people and environmental objects, thus obtaining the first video recognition data; Using a deep learning model, real-time predictive analysis is performed on the first video recognition data and the first environmental data to assess the congestion risk index and personnel safety risk index of each road segment. If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, a first control command will be sent to the 5G base station. The 5G base station is configured to control the corresponding traffic lights according to the first control command to adjust the traffic flow. The edge computing node is configured as follows: The congestion warning results are uploaded to a blockchain node and then pushed to the vehicle-mounted device after being encrypted and verified. In the event of an emergency, blockchain technology is used to find the optimal troubleshooting solution, and the solution is continuously optimized through a cross-domain federated learning model. The vehicle-mounted device is configured to upload the first vehicle data of the first vehicle it is in to the corresponding edge computing node; The edge computing node is further configured to: parse first traffic data, first road status data, and first road facility data from the first video recognition data, and upload the first traffic data, first road status data, first road facility data, first environmental data, and first vehicle data to the cloud server; The cloud server is also configured as follows: By analyzing the first traffic data, the first road condition data, the first road facility data, the first environmental data, and the first vehicle data, and combining the preset global optimal traffic organization strategy and vehicle platooning scheme, all vehicles traveling on the road will be divided into multiple vehicle groups. For each vehicle group, the first driving status data of each member vehicle in each group, the first mutual influence relationship data between each member vehicle, and the first current environmental condition data and the first current road condition data of each member vehicle are obtained. Based on the first driving status data, the first mutual influence relationship data, the first current environmental condition data, and the first current road condition data, the first risk existing in each member vehicle of each vehicle group is analyzed, and the control of each member vehicle in the vehicle group is carried out according to the first risk. The vehicle members of each vehicle group are dynamically adjusted, and the vehicles are managed according to the adjusted vehicle group situation; Specifically, by analyzing the first traffic data, the first road condition data, the first road facility data, the first environmental data, and the first vehicle data, and combining this with a preset globally optimal traffic organization strategy and vehicle platooning scheme, all vehicles traveling on the road are divided into multiple vehicle groups, including: A multi-source heterogeneous data integration platform is built on the cloud server; The multi-source heterogeneous data integration platform cleans, fuses, and encodes the received first traffic data, first road status data, first road facility data, first environmental data, and first vehicle data to obtain first comprehensive data. A traffic status digital twin model is generated based on the first comprehensive data and the city digital twin model. The traffic status digital twin model includes the road, speed, and destination of the first vehicle, and can reflect the traffic flow, congestion, and road facility status of each road segment in real time. Based on the traffic state digital twin model, and combined with the preset global optimal traffic organization strategy and vehicle platooning scheme, all the first vehicles traveling on the road will be divided into multiple vehicle groups. The step of analyzing the first risk existing in each member vehicle of each vehicle group based on the first driving status data, the first mutual influence relationship data, the first current environmental condition data, and the first current road condition data, and controlling each member vehicle of the vehicle group according to the first risk, includes: Acquire historical vehicle driving status data, historical environmental data, historical road data, and historical accident data that have spatiotemporal correspondence; Big data analytics is used to analyze the historical vehicle driving status data, historical environmental data, historical road data, and historical accident data. Based on the analysis results, a multi-dimensional risk assessment index system is constructed that includes vehicle-specific factors, road factors, environmental factors, and inter-vehicle interaction factors. By using the historical vehicle driving status data, the historical environmental data, the historical road data, the historical accident data, and the multi-dimensional risk assessment index system, a neural network is trained to learn the mapping relationship between risk assessment indicators and risk levels, thereby obtaining a risk identification model. The first driving status data, the first mutual influence relationship data, the first current environmental condition data, and the first current road condition data are input into the trained risk identification model to obtain the risk score and risk type of the first risk. For high-risk vehicles whose risk scores exceed a preset score threshold, determine the first risk type corresponding to the first risk. Construct a dynamic model to describe the propagation of risk in a vehicle group; Combining the dynamic model and the first risk type, analyze the impact of the high-risk state of the high-risk vehicle on the risk level of other member vehicles in the vehicle group; Based on the risk level, the risk propagation topology of the entire vehicle group is constructed in real time using the interaction data between all member vehicles in the vehicle group, and the propagation impact is assessed. Considering the risk identification results and the impact of their spread, and in conjunction with the first optimization objective, a customized first risk control strategy is formulated for each of the aforementioned member vehicles through a preset control strategy generation model; The first risk control strategy includes: for low-risk situations, the cloud server issues safety policy suggestions to the vehicle; for medium-risk situations, the cloud server sets hard constraints on the vehicle; for high-risk situations, the edge computing node issues an alert, and the cloud server switches the vehicle to remote control mode. The member vehicles provide feedback and score the implementation effect of the control strategy. The cloud server continuously optimizes the risk identification model and the control strategy generation model based on the scoring data to ensure that the vehicle group achieves higher driving efficiency and overall effectiveness under the premise of controllable risk. The dynamic adjustment of vehicle members in each vehicle group, and the management of vehicles based on the adjusted vehicle group situation, includes: The interaction and driving efficiency of the vehicle members within each vehicle group are monitored in real time. The overall collaborative efficiency of the vehicle group is evaluated and scored based on indicators such as energy consumption, driving safety, and matching degree with road conditions. Continuously track changes in external conditions; If the overall coordination efficiency evaluation score of the vehicle group is lower than the preset coordination efficiency evaluation threshold and / or the existing vehicle group cannot maintain the optimal coordination state under changing external conditions, the group reorganization demand detection is triggered, and a preset reorganization optimization strategy is used to determine whether reorganization is necessary. If reorganization is required, the vehicle reorganization problem is modeled as a Markov decision process based on reinforcement learning algorithms. The goal is to obtain higher overall collaborative rewards through reorganization, output the optimal reorganization scheme, and determine the new group members and the formation sequence of each vehicle. Control the trajectory of each vehicle within the original vehicle group to avoid collisions during the transition phase; The new vehicle group gradually adjusts the spacing and speed of each vehicle to transition to the new formation. The status of each vehicle in the new vehicle group and their mutual influence are analyzed and modeled. Combined with environmental and traffic factors, the risk points existing in the new vehicle group are identified, and targeted driving strategies are formulated for each vehicle to ensure that the new vehicle group achieves a state of coordinated efficiency and controllable risks. The cloud server monitors the operation status of all vehicle groups in real time and controls and handles abnormal situations. Collect real-time operating data of all vehicles, and continuously optimize the reorganization optimization strategy and vehicle management strategy model based on the real-time operating data; The new reorganization and optimization strategy and the new vehicle management strategy model are distributed to the vehicle-mounted equipment and the edge computing node to guide the real-time management of the vehicle group.

2. A smart city intelligent traffic management method based on 5G, characterized in that, include: Acquire planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data of the target city; A digital twin model of the target city is constructed based on the planning data, road network data, traffic facility data, historical traffic status data, real-scene 3D data, real-scene image data, and real-scene video data. Based on the urban digital twin model, multiple video acquisition terminals and environmental data acquisition terminals interconnected by a 5G network are configured on target roads within the target urban area; The video acquisition terminal and the environmental data acquisition terminal respectively upload the acquired first video data and first environmental data to the edge computing node via the 5G base station; The edge computing node performs object detection and semantic segmentation on the first video data to identify vehicles, people and environmental objects, and obtains the first video recognition data. The edge computing node uses a deep learning model to perform real-time predictive analysis on the first video recognition data and the first environmental data to assess the congestion risk index and personnel safety risk index of each road segment. If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, the edge computing node will send a first control command to the 5G base station. The 5G base station controls the corresponding traffic lights to adjust the traffic flow according to the first control command. The edge computing node uploads the congestion warning results to the blockchain node, and pushes them to the vehicle-mounted device after encryption and verification. In the event of an emergency, the edge computing node uses blockchain technology to find the optimal troubleshooting solution and continuously optimizes the solution through a cross-domain federated learning model. The on-board device uploads the first vehicle data of the first vehicle it is in to the corresponding edge computing node; The edge computing node parses the first traffic data, the first road status data, and the first road facility data from the first video recognition data, and uploads the first traffic data, the first road status data, the first road facility data, the first environmental data, and the first vehicle data to the cloud server; The cloud server analyzes the first traffic data, the first road status data, the first road facility data, the first environmental data, and the first vehicle data, and combines them with a preset global optimal traffic organization strategy and vehicle platooning scheme to divide all vehicles traveling on the road into multiple vehicle groups. For each vehicle group, the first driving status data of each member vehicle in each group, the first mutual influence relationship data between each member vehicle, and the first current environmental condition data and the first current road condition data of each member vehicle are obtained. Based on the first driving status data, the first mutual influence relationship data, the first current environmental condition data, and the first current road condition data, the first risk existing in each member vehicle of each vehicle group is analyzed, and the control of each member vehicle in the vehicle group is carried out according to the first risk. The vehicle members of each vehicle group are dynamically adjusted, and the vehicles are managed according to the adjusted vehicle group situation; The cloud server analyzes the first traffic data, the first road status data, the first road infrastructure data, the first environmental data, and the first vehicle data, and combines this with a preset globally optimal traffic organization strategy and vehicle platooning scheme to divide all vehicles traveling on the road into multiple vehicle groups. This process includes: A multi-source heterogeneous data integration platform is built on the cloud server; The multi-source heterogeneous data integration platform cleans, fuses, and encodes the received first traffic data, first road status data, first road facility data, first environmental data, and first vehicle data to obtain first comprehensive data. A traffic status digital twin model is generated based on the first comprehensive data and the city digital twin model. The traffic status digital twin model includes the road, speed, and destination of the first vehicle, and can reflect the traffic flow, congestion, and road facility status of each road segment in real time. Based on the traffic state digital twin model, and combined with the preset global optimal traffic organization strategy and vehicle platooning scheme, all the first vehicles traveling on the road will be divided into multiple vehicle groups. The step of analyzing the first risk existing in each member vehicle of each vehicle group based on the first driving status data, the first mutual influence relationship data, the first current environmental condition data, and the first current road condition data, and controlling each member vehicle of the vehicle group according to the first risk, includes: Acquire historical vehicle driving status data, historical environmental data, historical road data, and historical accident data that have spatiotemporal correspondence; Big data analytics is used to analyze the historical vehicle driving status data, historical environmental data, historical road data, and historical accident data. Based on the analysis results, a multi-dimensional risk assessment index system is constructed that includes vehicle-specific factors, road factors, environmental factors, and inter-vehicle interaction factors. By using the historical vehicle driving status data, the historical environmental data, the historical road data, the historical accident data, and the multi-dimensional risk assessment index system, a neural network is trained to learn the mapping relationship between risk assessment indicators and risk levels, thereby obtaining a risk identification model. The first driving status data, the first mutual influence relationship data, the first current environmental condition data, and the first current road condition data are input into the trained risk identification model to obtain the risk score and risk type of the first risk. For high-risk vehicles whose risk scores exceed a preset score threshold, determine the first risk type corresponding to the first risk. Construct a dynamic model to describe the propagation of risk in a vehicle group; Combining the dynamic model and the first risk type, analyze the impact of the high-risk state of the high-risk vehicle on the risk level of other member vehicles in the vehicle group; Based on the risk level, the risk propagation topology of the entire vehicle group is constructed in real time using the interaction data between all member vehicles in the vehicle group, and the propagation impact is assessed. Considering the risk identification results and the impact of their spread, and in conjunction with the first optimization objective, a customized first risk control strategy is formulated for each of the aforementioned member vehicles through a preset control strategy generation model; The first risk control strategy includes: for low-risk situations, the cloud server issues safety policy suggestions to the vehicle; for medium-risk situations, the cloud server sets hard constraints on the vehicle; for high-risk situations, the edge computing node issues an alert, and the cloud server switches the vehicle to remote control mode. The member vehicles provide feedback and score the implementation effect of the control strategy. The cloud server continuously optimizes the risk identification model and the control strategy generation model based on the scoring data to ensure that the vehicle group achieves higher driving efficiency and overall effectiveness under the premise of controllable risk. The step of dynamically adjusting the vehicle members of each vehicle group and managing the vehicles based on the adjusted vehicle group situation includes: The interaction and driving efficiency of the vehicle members within each vehicle group are monitored in real time. The overall collaborative efficiency of the vehicle group is evaluated and scored based on indicators such as energy consumption, driving safety, and matching degree with road conditions. Continuously track changes in external conditions; If the overall coordination efficiency evaluation score of the vehicle group is lower than the preset coordination efficiency evaluation threshold and / or the existing vehicle group cannot maintain the optimal coordination state under changing external conditions, the group reorganization demand detection is triggered, and a preset reorganization optimization strategy is used to determine whether reorganization is necessary. If reorganization is required, the vehicle reorganization problem is modeled as a Markov decision process based on reinforcement learning algorithms. The goal is to obtain higher overall collaborative rewards through reorganization, output the optimal reorganization scheme, and determine the new group members and the formation sequence of each vehicle. Control the trajectory of each vehicle within the original vehicle group to avoid collisions during the transition phase; The new vehicle group gradually adjusts the spacing and speed of each vehicle to transition to the new formation. The status of each vehicle in the new vehicle group and their mutual influence are analyzed and modeled. Combined with environmental and traffic factors, the risk points existing in the new vehicle group are identified, and targeted driving strategies are formulated for each vehicle to ensure that the new vehicle group achieves a state of coordinated efficiency and controllable risks. The cloud server monitors the operation status of all vehicle groups in real time and controls and handles abnormal situations. Collect real-time operating data of all vehicles, and continuously optimize the reorganization optimization strategy and vehicle management strategy model based on the real-time operating data; The new reorganization and optimization strategy and the new vehicle management strategy model are distributed to the vehicle-mounted equipment and the edge computing node to guide the real-time management of the vehicle group.

3. The intelligent traffic management method for smart cities based on 5G according to claim 2, characterized in that, The edge computing node uses a deep learning model to perform real-time predictive analysis on the first video recognition data and the first environmental data, and evaluates the congestion risk index and personnel safety risk index of each road segment, including the following steps: The edge computing node acquires the traffic facility data, the historical traffic status data, and the historical road video data; The historical road video data is aggregated using a time-series database to extract characteristics of vehicle and pedestrian traffic changes; By using a deep learning model and combining it with feature sequences extracted from the changes in vehicle and pedestrian traffic, a traffic state prediction model is obtained. Input the first video recognition data and the first environmental data into the traffic state prediction model to predict the traffic distribution in future time periods; Based on the traffic flow distribution prediction results, the traffic flow difficulty coefficient of each road segment is assessed as a congestion risk index. Real-time object detection algorithms and human pose estimation algorithms are applied to analyze the behavior of pedestrian flow and identify abnormal events. Based on the traffic distribution prediction results, the abnormal events, and the environmental data, the personnel safety risk index of different areas is assessed.

4. The intelligent traffic management method for smart cities based on 5G according to claim 3, characterized in that, The step of sending a first control command to the 5G base station if the congestion risk index and / or the personnel safety risk index exceed a preset risk index threshold includes: The edge computing node compares the congestion risk index and / or the personnel safety risk index with a preset risk index threshold. If the congestion risk index and / or the personnel safety risk index exceed the preset risk index threshold, a risk event is determined to have occurred. The edge computing node performs risk event identification and analysis on the first video recognition data and the first environmental data to determine the type and status of the first risk event that has occurred. Generate a corresponding first control instruction based on the type and status of the first risk event; The edge computing node sends the first control command to the corresponding 5G base station.

5. The intelligent traffic management method for smart cities based on 5G according to claim 4, characterized in that, The step of acquiring, on a per-vehicle-group basis, the first driving status data of each member vehicle in each group, the first mutual influence relationship data between each member vehicle, and the first current environmental condition data and the first current road condition data of each member vehicle includes: Each of the aforementioned member vehicles collects its own first driving status data through its own configured on-board equipment; Each of the member vehicles establishes a communication connection with other vehicles in the same vehicle group through the on-board equipment, and exchanges first mutual influence relationship data. Each of the aforementioned member vehicles collects first current environmental condition data and first current road condition data through the on-board equipment.

6. The intelligent traffic management method for smart cities based on 5G according to claim 5, characterized in that, The globally optimal traffic organization strategy and vehicle platooning scheme are obtained by analyzing multi-source data on vehicles, traffic, and road conditions across the entire road network collected by the cloud server, and applying artificial intelligence algorithms for global decision optimization. Specifically, the steps include: By aggregating multi-dimensional traffic data uploaded by roadside facilities and vehicle-mounted equipment, and through data fusion, modeling, and intelligent analysis, combined with the traffic state digital twin model, a digital twin of the traffic situation of the entire road network is obtained; Based on road network topology, traffic rules, vehicle dynamics knowledge, and the digital twin of the entire road network traffic situation, the motion, interaction, and state transition of vehicles in the road network are simulated. During the simulation, traffic organization strategies are used as the behavioral strategy space of the reinforcement learning agent. Design a global reward function to evaluate traffic organization strategies, and use the global reward function to guide reinforcement learning to obtain the globally optimal traffic organization strategy; Based on the globally optimal traffic organization strategy obtained at the road network level, a multi-agent cooperative algorithm is designed for platoonable vehicle clusters to output a vehicle platooning scheme that includes platooning sequence, vehicle spacing, and vehicle speed.

Images