Intelligent traffic classification control method for strong wind area cross-sea traffic cluster

Abstract

The application discloses a wisdom passing hierarchical control method for strong wind area sea-crossing passing cluster, and relates to the field of bridge traffic safety early warning and control in strong wind area. The method adopts a spatial flow state discrimination mechanism based on physical topology to determine a flow state flag, and together with multi-source heterogeneous sensing features, reconstructs a state constraint vector; through bidirectional time-varying coupling simulation of computational fluid dynamics and multi-body dynamics, training data sets under extreme working conditions are obtained, and an extreme gradient boosting tree is pre-trained; the state constraint vector is input into the pre-trained extreme gradient boosting tree, and a joint instability probability is output; a transient limit risk probability is obtained by combining a penalty factor with the joint instability probability, a critical safety vehicle speed is determined according to the probability, and a double-path execution path combining VMS targeted shouting and V2X terminal direct connection is used to intervene in the traffic behavior of vehicles with different network connectivity degrees. The application can realize accurate intervention on different types of vehicles in the sea-crossing passing cluster, and improve the accuracy of risk early warning and passing efficiency under complex traffic environment.

Description

Technical Field

[0001] This application relates to the field of bridge traffic safety early warning and control in strong wind areas, and in particular to a smart traffic hierarchical control method for cross-sea traffic clusters in strong wind areas. Background Technology

[0002] In extreme geographical environments such as "bridge-island-tunnel" cross-sea traffic clusters, strong wind fields exhibit high levels of unsteadiness, spatial abruptness, and multi-scale transient eddy characteristics. For traffic safety management in such areas, current mainstream solutions still rely on the traditional "macro-meteorological early warning" and "static wind speed threshold speed limit" models.

[0003] Although the existing control system has made initial attempts to incorporate environmental parameters such as wind speed, wind direction, and road surface adhesion coefficient, it still reveals the following irreconcilable technical contradictions when dealing with increasingly complex mixed traffic flows and extreme crosswind environments: The lack of deep physical coupling among multi-source heterogeneous parameters: Existing technologies exhibit significant linear superposition characteristics in their logical architecture, failing to achieve deep physical coupling of all elements—wind, vehicle, and road. Specifically, while the system monitors the environmental wind field and road surface friction, it often overlooks the nonlinear amplification effect of the vehicle's dynamic attributes on aerodynamic response. This disconnect between physical and logical aspects prevents control commands from reaching the most severely affected individuals, easily leading to insufficient control precision for high-risk vehicles.

[0004] The lack of consideration for multi-vehicle aerodynamic interference: Most existing models are built on the assumption of "isolated single vehicles," severely underestimating the flow field distortion caused by multiple vehicles driving in parallel, overtaking at high speeds, or closely following each other in cross-sea traffic clusters. In strong winds, the aerodynamic shielding effect of large vehicles and the sudden change in crosswinds at the moment of departure can have a significant transient impact on adjacent vehicles. Existing solutions lack quantitative perception of traffic flow topology distances and risk compensation, making it impossible to implement proactive spatial coordinated intervention before chain-reaction accidents occur.

[0005] The granularity of traffic control decisions is large, and the feedback methods are limited: Traditional control strategies often employ a "one-size-fits-all" approach of global speed reduction or physical shutdown, with instructions relying heavily on broadcast announcements from roadside information boards. This "asymmetric and non-targeted" information delivery method fails to leverage the millisecond-level low-latency control advantages of Intelligent Connected Vehicles (ICVs), and also fails to provide precise, differentiated guidance for vehicles with different characteristics while ensuring overall traffic efficiency. Summary of the Invention

[0006] The purpose of this application is to provide a smart traffic hierarchical management method for cross-sea traffic clusters in strong wind areas, which can achieve precise intervention of vehicles and improve the accuracy of risk warning and traffic efficiency in complex traffic environments.

[0007] To achieve the above objectives, this application provides the following solution: This application provides a smart hierarchical management method for cross-sea traffic clusters in strong wind areas, including: Acquire multi-source heterogeneous perception features of target vehicles when traveling in cross-sea traffic clusters; the multi-source heterogeneous perception features include environmental domain features, vehicle ontology domain features, and vehicle cooperation domain features. Based on the characteristics of the vehicle cooperative domain, a spatial flow state discrimination mechanism based on physical topology is used to perform quantitative flow state discrimination and determine the flow state flag. After spatiotemporal registration of the multi-source heterogeneous sensing features, they are reconstructed together with the flow state flags into a state constraint vector; By using bidirectional time-varying coupled simulation of computational fluid dynamics and multibody dynamics, a training dataset under extreme conditions is obtained, and the extreme gradient boosting tree is pre-trained using the training dataset to obtain a pre-trained extreme gradient boosting tree. The reconstructed state constraint vector is input into the pre-trained extreme gradient boosting tree, and the joint instability probability in the initial state is output. Based on the flow regime flag, the penalty factor is determined using the analytical aerodynamic mutation penalty surrogate model, and multiplicatively coupled with the joint instability probability in the initial state to obtain the transient limit risk probability of the target vehicle. If the transient limit risk probability is greater than the safety decision red line threshold, a one-way discrete step forward heuristic search algorithm based on vehicle dynamics prior is used to determine the critical safe vehicle speed. Based on the deviation gradient between the real-time vehicle speed and the critical safe vehicle speed of the target vehicle, the multi-dimensional cooperative control command is adaptively calculated. If the target vehicle is a non-connected vehicle, then according to the multi-dimensional collaborative control instructions, the target vehicle will be subjected to macroscopic visual target addressing and flexible control based on variable message signs. If the target vehicle is a connected vehicle, then according to the multi-dimensional collaborative control command, the target vehicle will be subjected to rigid takeover of the micro-level drive-by-wire chassis based on cellular vehicle networking.

[0008] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a smart traffic hierarchical management method for cross-sea traffic clusters in strong wind areas. Multi-source heterogeneous sensing features include environmental domain features, vehicle body domain features, and vehicle cooperation domain features. After spatiotemporal registration of these multi-source heterogeneous sensing features, they are reconstructed together with flow regime flags into a state constraint vector, achieving deep physical coupling of all elements of "wind-vehicle-road," thus improving the accuracy of risk warnings in complex traffic environments. A spatial flow regime discrimination mechanism based on physical topology is introduced, which quantifies wake distortion and aerodynamic shielding effects in real-time under multi-vehicle parallel, meeting, and following states, and uses a "penalty factor to correct single-vehicle wind..." This approach aims to reduce the risk of accidents and achieve a technological leap from "isolated individual protection" to "cluster spatial collaborative scheduling." It avoids, at the source, the aerodynamic shielding effect caused by large vehicles and the significant transient impact of sudden crosswind changes on adjacent vehicles caused by aerodynamic changes, thus reducing the probability of instability accidents caused by aerodynamic mutations. By reverse-analyzing the critical safe speeds of vehicles with different characteristics and combining targeted announcements from variable message signs with direct connections to cellular vehicle-to-everything (V2X) terminals, it enables precise intervention for vehicles with varying levels of connectivity. Under the premise of ensuring absolute safety, it maximizes the traffic potential of the cross-sea traffic cluster. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas provided in this application embodiment.

[0011] Figure 2 This is a schematic diagram of the bidirectional coupling experimental process provided in the embodiments of this application.

[0012] Figure 3 A schematic diagram of the functional modules of the intelligent traffic hierarchical management and control system for cross-sea traffic clusters in strong wind areas provided in this application embodiment.

[0013] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] Terminology Explanation: WIM (Weigh-In-Motion): A dynamic weighing system used to measure the weight and load status of a vehicle in real time while it is in motion.

[0017] XGBoost (Extreme Gradient Boosting): Extreme Gradient Boosting Tree, a machine learning algorithm based on decision tree ensembles.

[0018] VMS (Variable Message Sign): Variable message signs are used to deliver traffic control and guidance information to vehicles on the road.

[0019] V2N / V2X (Vehicle-to-Everything): Vehicle-to-everything communication technology that enables data interaction between vehicles and the cloud or infrastructure.

[0020] RSU (Road Side Unit): A roadside computing and communication unit installed on both sides of the road for communicating with onboard terminals.

[0021] ICV (Intelligent Connected Vehicle): Intelligent connected vehicles.

[0022] OBU (On-Board Unit): A vehicle-mounted terminal, a communication device installed inside a vehicle.

[0023] CFD (Computational Fluid Dynamics)-MBD (Multi-Body Dynamics) bidirectional coupled simulation: This application refers to the cross-domain data interaction between a fluid simulation model and a vehicle dynamics model to simulate the interaction process between wind field flow and vehicle dynamic response.

[0024] Traffic Flow State: Based on the spatial distribution characteristics of vehicles (longitudinal distance, lateral distance, etc.) extracted by sensing devices, traffic flow is divided into different modes such as single-vehicle, sparse, and dense, and is used to characterize the degree of aerodynamic interference among multiple vehicles.

[0025] Aerodynamic mutation penalty factor: a mathematical correction coefficient based on traffic flow quantification of the impact of multi-vehicle aerodynamic disturbances (such as wake and obstruction) on driving safety.

[0026] To address the technical bottlenecks in cross-sea traffic cluster management in strong wind areas, such as low parameter coupling, multi-vehicle intervention vacuum, and inaccurate command delivery, an exemplary embodiment is proposed, such as... Figure 1 As shown, this application provides a smart passage hierarchical management method for cross-sea passage clusters in strong wind areas. The method is executed by computer equipment, specifically by a terminal or server or other computer equipment alone, or by a terminal and a server together. In the embodiments of this application, the method includes the following steps 101 to 110.

[0027] Step 101: Obtain the multi-source heterogeneous perception features of the target vehicle when it is traveling in a cross-sea traffic cluster; the multi-source heterogeneous perception features include environmental domain features, vehicle ontology domain features and vehicle cooperation domain features.

[0028] Step 102: Based on the characteristics of the vehicle cooperative domain, a spatial flow state discrimination mechanism based on physical topology is used to perform quantitative flow state discrimination and determine the flow state flag.

[0029] Step 103: After performing spatiotemporal registration on the multi-source heterogeneous sensing features, reconstruct them together with the flow state flags into a state constraint vector.

[0030] Step 104: Obtain a training dataset under extreme conditions through bidirectional time-varying coupled simulation of computational fluid dynamics and multibody dynamics, and pre-train an extreme gradient boosting tree using the training dataset to obtain a pre-trained extreme gradient boosting tree.

[0031] Step 105: Input the reconstructed state constraint vector into the pre-trained extreme gradient boosting tree and output the joint instability probability in the initial state.

[0032] Step 106: Based on the flow state flag, use the analytical aerodynamic mutation penalty surrogate model to determine the penalty factor, and multiplicatively couple it with the joint instability probability in the initial state to obtain the transient limit risk probability of the target vehicle.

[0033] Step 107: If the transient limit risk probability is greater than the safety decision red line threshold, then a one-way discrete step forward heuristic search algorithm based on vehicle dynamics prior is used to determine the critical safe vehicle speed.

[0034] Step 108: Based on the deviation gradient between the real-time vehicle speed and the critical safe vehicle speed of the target vehicle, adaptively calculate the multi-dimensional cooperative control command.

[0035] Step 109: If the target vehicle is a non-connected vehicle, then according to the multi-dimensional collaborative control instructions, perform macroscopic visual target addressing and flexible control of the target vehicle based on variable message signs.

[0036] Step 110: If the target vehicle is a connected vehicle, then according to the multi-dimensional collaborative control command, perform micro-level drive-by-wire chassis rigid takeover of the target vehicle based on cellular vehicle networking.

[0037] By implementing steps 101 to 110 above, the following core technical objectives were achieved: 1. Construct a multi-element integrated early warning model encompassing wind, vehicles, and roads. This model changes the current situation where physical parameters operate in isolation. Through feature vectorization technology, it performs high-dimensional nonlinear mapping of transient wind fields, vehicle dynamic attributes, and road surface friction characteristics. The core of this approach lies in uncovering the coupling mechanism between environmental variables and individual characteristics, eliminating the early warning blind spots caused by the coarse logic of traditional lookup methods.

[0038] 2. Establishing a spatial flow control mechanism considering multi-vehicle aerodynamic interference: This application aims to establish a spatial flow discrimination mechanism based on physical topology, which quantifies wake distortion and aerodynamic shielding effects in real time under multi-vehicle parallel, meeting, and following conditions. By using a "penalty factor" to correct the risk probability of a single vehicle, a technological leap from "isolated individual protection" to "cluster spatial collaborative scheduling" is achieved. This avoids, at the source, the aerodynamic shielding effects caused by large vehicles and the significant transient impact of sudden crosswind changes at the moment of departure on adjacent vehicles, which can lead to chain-reaction instability accidents.

[0039] 3. Combining offline simulation and online prediction, we utilize offline CFD-MBD bidirectional time-varying coupled simulation to exhaustively enumerate and accumulate high-fidelity physical prior data under extreme conditions. Based on this, we endow the online XGBoost model with strong physical logic support, balancing the accuracy of physical and mechanical calculations with the responsiveness of online predictions.

[0040] 4. Establish a tiered and differentiated information dissemination system. In response to the current mixed traffic flow, this application aims to break the deadlock of the "one-size-fits-all" broadcast-style control. By reverse-analyzing the critical safe speeds of vehicles with different characteristics, and combining targeted announcements via VMS with direct V2X terminal connections, precise intervention can be achieved for vehicles with varying levels of connectivity. This maximizes the traffic potential of the cross-sea traffic cluster while ensuring absolute safety.

[0041] In another exemplary embodiment of this application, step 101 involves spatiotemporal registration and feature extraction of multi-source heterogeneous physical parameters. Data acquisition and spatiotemporal registration are performed using sensing devices deployed on roadside edge computing nodes of the cross-sea traffic cluster. Through a rigorous filtering and coordinate registration mechanism, a high-fidelity underlying state vector is provided for the downstream nonlinear prediction model.

[0042] Environmental feature extraction: A three-dimensional ultrasonic anemometer array is deployed at equal intervals along the cross-sea traffic cluster to acquire the transient wind field matrix at a preset high-frequency sampling rate. To remove the high-frequency white noise from the sensors caused by natural airflow turbulence, a time-sliding window filtering algorithm is embedded in the system to calculate and output the smoothed effective environmental wind speed. With relative composite wind direction angle This serves as the subsequent high-dimensional state constraint vector. The core physical input parameter is synchronously linked with a roadside non-contact multispectral pavement condition sensor to identify the current water film thickness or ice and snow cover phase of the pavement, and based on a built-in tribological empirical model, it is mapped in real time to a dimensionless comprehensive pavement adhesion coefficient. Among them, the relative composite wind direction angle The calculation process is as follows: To accurately represent the actual wind conditions of the vehicle, the Galilean principle of relative velocity transformation is further introduced, combined with the real-time vehicle speed of the target vehicle extracted by the roadside equipment. relative to the natural wind direction of the environment The wind speed vector is orthogonally decomposed using the formula. Real-time calculation of the relative composite wind direction angle currently experienced by the target vehicle. (i.e., aerodynamic yaw angle). Therefore, environmental domain characteristics include effective environmental wind speed. Relative composite wind direction angle and comprehensive road surface adhesion coefficient .

[0043] Ontology-domain feature extraction: When a target vehicle enters the defense zone, the road surface dynamic weighing piezoelectric sensor array and the gantry lidar are synchronized at the microsecond level. This not only obtains the initial entry speed of the target vehicle, but also... In addition to vehicle type classification, based on this vehicle type classification and point cloud contour, the system inversely predicts the current centroid height of the target vehicle. The vehicle type classification primarily utilizes dynamic weighing sensors and lidar to obtain the initial entry speed of the vehicles. With dynamic axis redistribution matrix In terms of vehicle type classification, the system integrates the number of axles, wheelbase, and chassis topology features extracted from WIM, as well as the three-dimensional geometric contours (vehicle length, width, height, and frontal shape) parsed from LiDAR point clouds. These are then input into a pre-defined physical prior database for feature matching to accurately output the vehicle type classification. Therefore, the vehicle body domain features include: the initial entry speed of the target vehicle. Dynamic axis redistribution matrix and the current center of gravity height of the target vehicle .

[0044] The accurate extraction of the aforementioned ontological features provides a fundamental physical foundation for subsequent solutions to unsteady yaw moments and determination of aerodynamic side force coefficients. More crucially, it involves extracting the dynamic axis weight distribution matrix. Meanwhile, based on the bounding box algorithm of LiDAR point clouds, the geometric contour of the vehicle's windward side is accurately extracted, and combined with the vehicle model's physical prior database, the current center of gravity height of the vehicle is predicted in reverse. The aforementioned vehicle model physical prior database is not a simple name mapping, but rather embeds the corresponding vehicle model's aerodynamic reference parameter surface (including the lateral force coefficient that varies with the aerodynamic yaw angle). Rolling moment coefficient etc.), chassis inherent parameters (suspension roll stiffness) and the highly evolved envelope of the center of mass The multidimensional data matrix provides a rigorous physical loop for the accurate inversion of downstream dynamic instability. This represents the minimum height of the center of gravity for the corresponding vehicle model. This represents the maximum value of the center of gravity height for the corresponding vehicle model. In the center of gravity prediction stage, a reverse analytical mechanism based on the conservation of mass moments is utilized: the total mass extracted from the WIM... Real-time vehicle height extracted with LiDAR For input, the unloaded mass of the linked vehicle model database. Compared with the unloaded centroid reference Through formula By dynamically solving for the combined rigid body's center of mass and combining this with suspension stiffness to implement dynamic deflection and settlement compensation, real-time, high-precision center of mass inversion without contact measurement is achieved. The accurate extraction of this parameter provides crucial underlying physical support for subsequent solutions to unsteady yaw and roll moments.

[0045] Collaborative domain feature extraction: By fusing roadside millimeter-wave radar and visual cameras, and utilizing the Doppler effect and depth perception characteristics of millimeter-wave radar, high-precision relative longitudinal distances between target vehicles and adjacent vehicles can be extracted through complex weather interference (such as fog, haze, rain, and snow). and transient relative velocity difference Simultaneously utilizing target-level semantic recognition from a visual camera (capable of identifying a vehicle in the image and distinguishing between a sedan and a truck) and a 2D bounding box tracking algorithm (continuously locking the 2D contour edges of the vehicle through visual tracking to prevent target loss), the algorithm extracts the vehicle's lateral topological offset and relative lateral spacing within the lane. Therefore, the characteristics of the vehicle cooperative domain include: the relative longitudinal distance between the target vehicle and adjacent vehicles, the relative lateral distance between the target vehicle and adjacent vehicles, and the transient relative speed difference.

[0046] To obtain continuous physical motion trajectories, the sensor calibration matrix is ​​invoked to perform a spatial homogeneous coordinate transformation, which maps heterogeneous data to a global three-dimensional Cartesian coordinate system. Subsequently, a Kalman filter algorithm is introduced, and through a high-frequency iterative closed loop of "prediction-update", a smooth, continuous multi-vehicle dynamic cooperative motion trajectory with a unified timestamp is fitted.

[0047] In another exemplary embodiment of this application, in the actual engineering implementation of cross-sea traffic clusters in strong wind areas, the "bridge-island-tunnel" space is often accompanied by harsh physical environments such as extremely high salt spray, torrential rain, and strong typhoons. Roadside millimeter-wave radar, ultrasonic anemometers, and road dynamic weighing systems inevitably experience brief signal failures, disconnections, or hardware-level damage. To ensure absolute driving safety throughout the entire lifecycle, in a preferred embodiment of this application, a three-layer cascaded fault-tolerant degradation network with the characteristics of "automatic hardware defect perception—conservative physical boundary replacement—control law red line cutoff" is built in. That is, this application uses a real-time data quality monitoring module deployed on roadside edge computing nodes to perform online availability assessment and health verification of the acquired multi-source heterogeneous sensing features. When data loss, outlier overflow, or time-domain refresh delay is detected in the multi-source heterogeneous sensing features, the three-layer cascaded fault-tolerant degradation control strategy is adaptively activated. The three-tiered fault-tolerant degradation control strategy includes: the first-tier ontological perception degradation strategy, the second-tier environmental perception degradation strategy, and the third-tier all-hazard red line cutoff strategy.

[0048] Level 1: When the WIM dynamic weighing piezoelectric sensor or gantry lidar at the bridge entrance is unable to output the dynamic axle load distribution matrix of the target vehicle in real time due to heavy rain, water accumulation, or obstructed view. Or reverse inversion of the centroid height At this time, the data health checker of the edge scheduling node will immediately capture the abnormal state where the input variable is NaN (Not a Number) or 0. In this case, the system will not interrupt the calculation, but will actively activate the ontology perception degradation routing: the system uses the remaining unique vehicle license plate ID or the coarsely captured two-dimensional outline from the visual camera to determine the vehicle's baseline vehicle type classification (e.g., large three-axle heavy-duty van). It then forcibly extracts the worst-case comprehensive risk parameters for this vehicle type on the mechanical boundary from the preset "vehicle type physical prior database". Specifically, the system uses the maximum legal gross vehicle weight of the vehicle type to calculate the sideslip failure risk, and simultaneously uses the maximum center of gravity height corresponding to the maximum cargo box load limit of the vehicle type. To calculate the aerodynamic roll moment. That is, if the dynamic axle redistribution matrix or centroid height data in the vehicle body domain features are missing, the vehicle model physical prior database is called to retrieve the vehicle model classification corresponding to the target vehicle, and the maximum mass envelope under full load and the maximum centroid envelope under no load corresponding to the vehicle model classification are extracted and used as alternative physical parameters to reconstruct the state constraint vector.

[0049] The second level: When ultrasonic anemometers reach their maximum wind speed during typhoons, data updates often stall for hundreds of milliseconds due to strong wind vibrations or power supply instability. This application embeds a watchdog timer into the environmental feature extraction algorithm; once a valid environmental wind speed is detected... Data refresh interval This indicates a brief disconnection in the wind field sensing link. The system immediately initiates environmental perception degradation routing, cuts off the current transient real-time data acquisition channel, and instantly activates the historical memory pool backtracking mechanism to retrieve data from before the disconnection occurred. All high-frequency wind speed samples recorded by the sensor array are included. The maximum gust wind speed within the sliding window at that 3-second interval is extracted. And the most unfavorable relative composite wind angle corresponding to this peak value. The historical peak value is used as a constant input during the current quiet period, continuously driving the downstream analytical aerodynamic mutation penalty proxy model and vehicle speed reverse addressing algorithm. This peak value will remain until the hardware link recovers and the data stream passes denoising verification for 5 consecutive frames. That is, if the refresh delay of the effective environmental wind speed data in the environmental domain features exceeds a preset time threshold, the maximum effective gust peak value within the current preset sliding window time is automatically locked as the input environmental wind speed for high-frequency risk extrapolation. High-frequency risk extrapolation refers to the complete computational link of "online risk prediction and safety status reassessment" constituted by steps 105 to 107.

[0050] Level 3: In the event of force majeure such as a super typhoon making direct landfall or a sudden large-scale power network failure across the entire bridge, if the edge computing nodes detect that the hardware offline rate of the heterogeneous sensing devices on the roadside of the entire bridge exceeds the limit (e.g., Or, due to extremely turbulent data flow, the online XGBoost inference engine and unidirectional discrete iterative search algorithm may experience time consumption exceeding the timeliness threshold (e.g., single solution). The system will determine that the current Cyber-Physical Systems (CPS) no longer possess the technical conditions to conduct micro-level differentiated targeted intervention. The system will proactively exit the micro-level individual targeted takeover mode, automatically triggering the highest-priority red line cutoff hard downgrade, completely exiting the micro-level drive-by-wire chassis hard takeover logic for intelligent connected vehicles, and the vehicle-by-vehicle VMS targeted visual addressing logic for traditional vehicles. Control will be transferred to the bridge emergency command center, triggering a one-click activation of the variable message signs and broadcast system across the entire road to forcibly project the highest-priority static absolute speed limit red line semantics (e.g., "Strong wind extreme warning across the entire bridge, mandatory speed limit of 40km / h across the entire line, please move to the nearest safety strip"). If the wind speed completely breaches the bridge's structural design safety limits, the roadside physical barriers will be activated to physically close the bridge. That is, if the offline rate of global roadside sensing devices in the roadside edge computing node is greater than the preset number red line, or the online inference time of the extreme gradient boosting tree exceeds the timeliness threshold, the micro-individual targeted takeover mode will be actively exited, and the macro-static speed limit red line instruction for the entire road section will be issued directly with one click, or the right of passage of the bridge will be physically closed.

[0051] In another exemplary embodiment of this application, step 102 described above may be replaced by steps 201 to 203.

[0052] Step 201: If the target vehicle and all adjacent vehicles within the perception range satisfy the following conditions: or If the target vehicle is outside the aerodynamic wake and obstruction zone, the flow status flag is set to 0; where, For the target vehicle Adjacent vehicles within the perception range The relative longitudinal distance between vehicles, For vertical security threshold, For the target vehicle Adjacent vehicles within the perception range The relative lateral spacing, The threshold for lateral aerodynamic influence.

[0053] Step 202: If the target vehicle and any adjacent vehicle within the perception range satisfy... , and This indicates that the target vehicle is undergoing a high-speed parallel overtaking or passing maneuver with an adjacent vehicle, and the flow status flag is set to 1; where, For the target vehicle Adjacent vehicles within the perception range The relative speed difference The threshold for relative velocity abrupt change.

[0054] Step 203: If the target vehicle and any adjacent vehicle within the perception range satisfy... , and This indicates that the convoy follows the car, and the following car remains in the turbulent wake of the preceding car for a preset period of time. The flow status flag is set to 2.

[0055] With the target vehicle For the object, a more detailed example of quantitative flow regime determination is as follows: Target vehicle any adjacent vehicle within its perception range Real-time calculation of relative longitudinal distance between vehicles Relative lateral spacing and relative speed difference The system has a preset strict physical space interference boundary (longitudinal safety threshold). Horizontal aerodynamic influence threshold Relative velocity change threshold Perform the following quantitative flow regime determination: Mode A (Single-vehicle flow): If all surrounding vehicles All meet or The system determines that the target vehicle is outside the aerodynamic wake and obstruction zone. The system outputs the flow status flag State = 0, indicating that there is no aerodynamic abrupt interference.

[0056] Mode B (Sparse Intersection Flow): If adjacent vehicles exist Intrusion into the space influence zone ( and ), and the relative speed difference between the two sides is large ( This state indicates that a large truck and a passenger car are overtaking or passing each other at high speed, which can easily induce aerodynamic instability due to wind wall obstruction and "aerodynamic sway" at the moment of departure. The system outputs the flag State = 1 and activates the transient aerodynamic penalty term.

[0057] Mode C (Dense Car-Following Flow): If adjacent vehicles exist It intrudes into the area of ​​influence of space, but the relative velocity difference is extremely small. This state indicates that the convoy is closely following each other, with the following vehicle remaining in the turbulent wake of the preceding vehicle for an extended period. The system outputs the flag State = 2 and activates the steady-state aerodynamic penalty term.

[0058] In another exemplary embodiment of this application, the aforementioned full-element physical parameters after spatiotemporal registration and logical discrimination are finally concatenated and reconstructed into a high-dimensional feature vector. It is then pushed into a memory pool for online prediction models to access with extremely low latency.

[0059] In another exemplary embodiment of this application, step 104 above relates to the generation of digital twins and datasets based on fluid-structure interaction.

[0060] Because it is extremely difficult to obtain extreme accident data on vehicle skidding / rollover under aerodynamic interference from multiple vehicles in strong wind areas in the real physical world, this application abandons the traditional assumption of "single-vehicle coupling + rigid traffic flow" in the offline stage and introduces a "multi-rigid-body global dynamic bidirectional time-varying coupling architecture" for multi-vehicle traffic flow. This architecture strips away conventional driver path tracking control and serves purely as a high-fidelity data generator for underlying physical laws, aiming to build a training dataset with strong generalization capabilities for the downstream XGBoost prediction model. Figure 2 As shown, the specific implementation logic is as follows: 1. Physical Domain Initialization: First, a fluid computational domain encompassing the spatial topology of the cross-sea traffic cluster is established. The computational domain is a three-dimensional cuboid space. To eliminate the artificial obstruction effect of external boundaries on flow field distortion, the characteristic scale of the computational domain is strictly constrained by the characteristic dimensions of the main vehicles (let the vehicle length be L and the vehicle height be H). Specifically, the boundaries are set as follows: the inlet boundary is located at least 5L on the windward side at the front of the convoy, used to apply a fully developed atmospheric boundary layer wind speed profile. The outlet boundary is located at least 15L on the leeward side at the rear of the convoy, ensuring that the wake detachment and eddy dissipation processes of multiple vehicles are fully represented without truncation. The side and top boundaries are at least 8H away from the target vehicle group and are set as symmetrical boundaries or slip walls. The bottom boundary constructs a non-slip wall incorporating the actual road surface undulations, guardrail geometry, and windbreak netting geometry of the cross-sea bridge. To achieve independent pose shifts and overtaking maneuvers for multiple vehicles, the flow computation domain abandons the traditional single global dynamic mesh. Instead, it adopts a topology that merges a static background computation domain (including the bridge infrastructure, using a high-quality static hexahedral mesh to support global Large Eddy Simulation (LES) background flow field calculations) with multiple independent nested foreground domains (establishing independent localized computation domains for the target vehicle and each adjacent vehicle; the boundary layer meshes of each subdomain rigidly translate and rotate according to the multibody dynamics response of the corresponding vehicle; and the external boundaries of the subdomains exchange flow fluxes with the background computation domain in real time via interpolation algorithms). A transient aerodynamic analysis model is established using LES, and spatial filtering techniques are used to separate flow field vortices, solving the filtered incompressible Navier-Stokes equations.

[0061] For microscale eddies that are not directly analyzed, the Smagorinsky-Lilly subgrid closed model is used, and the subgrid eddy viscosity is solved. Quantifying its dissipative effect on the mainstream field ( For subgrid eddy viscosity, For strain rate tensor, For pre-calibration constants, (For filtering scale). After the transient aerodynamic analysis model is established, it serves as the underlying physical driving engine, operating at a set discrete time step. The following steps are taken to participate in the cross-domain joint simulation closed loop: First, the pressure distribution and wall shear stress of the surface mesh of the main vehicle and the traffic vehicle group in the transient aerodynamic analysis model are integrated to extract the six components of transient aerodynamic forces (focusing on the aerodynamic lateral force). ) and aerodynamic moments (rolling moment / yaw moment) First, the data is written to the shared storage matrix. Second, the dynamics solver reads the corresponding vehicle's... and By combining the vehicle mass distribution and the road adhesion coefficient, the pose response parameters of each entity at that time step, such as lateral velocity, yaw rate, and roll rate, are calculated. Finally, the fluid solver reads the above global pose response data, drives the boundary layer mesh of each foreground nested domain to rigidly translate and rotate with the entity vehicle, and re-iterates the aerodynamic model. This application mechanism accurately reproduces the wake and secondary distortion of the occlusion flow field caused by the relative displacement of the vehicle, providing a basis for subsequent extraction of extreme conditions, instability probability features, and generation of structured training datasets. It provides a physical foundation that conforms to the strict laws of fluid mechanics.

[0062] After establishing the transient aerodynamic analysis model, a transient wind field profile of the atmospheric boundary layer with spatially fluctuating time histories is applied to the inlet boundary of the fluid computational domain for driving. Simultaneously, in the multibody dynamics domain, nonlinear multi-degree-of-freedom vehicle models of the target vehicle and multiple adjacent vehicles are instantiated in parallel, and corresponding road adhesion coefficients, vehicle mass distribution, and initial topological coordinates are assigned to them respectively. This model focuses on establishing lateral ( ), horizontal ( ) and roll ( (Multi-degree-of-freedom nonlinear dynamic equations).

[0063] The lateral dynamics equation is: ; In the formula, For the overall vehicle weight; The lateral velocity change rate, It is an aerodynamic lateral force.

[0064] The equation for yaw dynamics is: .

[0065] In the formula, Let be the yaw moment of inertia of the vehicle about the Z-axis. This is the yaw acceleration. , The geometric distance parameter from the center of mass to the front and rear axes. This is the sum of the lateral forces acting on the front tires. This is the sum of the lateral forces acting on the rear tires. It is the aerodynamic yaw moment.

[0066] The roll dynamics equation is: ; In the formula, For longitudinal / lateral velocity; This refers to the yaw rate; The roll angle; The tire lateral force is calculated based on a nonlinear tire model (such as the magic formula). The aerodynamic load is transmitted from the CFD. It is the moment of inertia; These represent the suspension roll stiffness and damping coefficient. Let X be the vehicle's roll inertia about the X-axis. The angular velocity is the roll rate. This is the roll acceleration. The lateral acceleration at the vehicle's center of gravity. For the tilt lever arm, This is the aerodynamic tilting moment.

[0067] This nonlinear vehicle multi-degree-of-freedom model is used as both an "attitude responder" and a "failure referee." It uses the transient aerodynamic loads from the CFD as external excitations and integrates the aforementioned equations in real time to calculate the vehicle's lateral slip and roll angle at the current time step. The calculated pose is then fed back to the CFD's dynamic mesh solver. Finally, when the lateral slip exceeds the lane limit or the tire normal load transfer rate... If the vehicle is deemed unstable, the current cycle is immediately terminated, and the resulting condition is labeled as "instability" and added to the training dataset. .

[0068] Before initiating transient calculations, initial wind field boundary conditions are pre-loaded to perform steady-state flow field calculations. First, all multi-degree-of-freedom vehicle models are locked, placing them in a rigid, stationary state (no displacement feedback). A steady-state wind speed profile of the atmospheric boundary layer with bridge surface roughness is applied at the CFD computational domain inlet. The Reynolds-averaged equations (such as...) are then invoked. A steady-state calculation is performed using a turbulence model to rapidly dissipate the initial non-physical fluctuations of the flow field. When the global residual of the flow field decreases to a preset threshold (such as...), the steady-state calculation is performed. The calculation is stopped when the pressure at the monitoring points on the vehicle surface remains constant. After the flow field converges, the converged flow field is used as the initial environmental variable for transient two-way coupling to avoid non-convergence or non-physical oscillations caused by sudden changes in the flow field in the early stage of transient calculation.

[0069] 2. Two-way coupled data interaction, within a set unified discrete time step (e.g., time step). Under these conditions, the system performs a strict cross-domain handshake protocol by frequently scanning and reading / writing shared data storage files: Forward transmission of aerodynamic loads: in the first At each time step, the fluid solver simultaneously performs independent integration on the surface mesh of the "main vehicle and all other vehicles" within the computational domain, extracting their respective macroscopic transient aerodynamic components (with a focus on extracting lateral aerodynamic forces). ) and aerodynamic moments (focusing on roll moment / yaw moment) And write it to the shared storage file in parallel.

[0070] Dynamic response calculation: The dynamic solver synchronously reads the corresponding values ​​of each vehicle at that moment. and This is applied as an external excitation to the corresponding vehicle rigid body center of mass. This is combined with the current vehicle speed and physical constraints (including the tire limit friction circle constraint). With suspension mechanical travel constraints, The longitudinal force of the tire. The lateral force of the tire. (For the normal vertical load), the kinematic response parameters of the main vehicle and all other vehicles at this time step are calculated in parallel using a two-way coupling method, combining Fluent, MATLAB, Simulink, and CarSim software. These parameters include: lateral velocity (lateral speed). ), yaw rate yaw rate and pitch angular velocity Then, the aforementioned global response parameters are written in reverse to the storage file.

[0071] Dynamic Mesh Reconstruction: The fluid solver synchronously reads the attitude response data of all the aforementioned vehicles through a User-Defined Function (UDF), driving the boundary layer dynamic mesh group of multiple vehicle entities within the computational domain to simultaneously undergo dynamic displacement and deflection. This mechanism realistically reproduces the chain reaction in the physical world: that is, the lateral displacement of wind-driven traffic vehicles will instantly reconstruct its wake and occlusion blind zone distribution, and this distorted flow field will immediately affect the... The master vehicle at each time step generates a highly destructive "secondary aerodynamic change", thereby realizing a true inversion of the complex mixed traffic loss and stability mechanism.

[0072] 3. Extreme Working Condition Extraction and Dataset Construction: An automated script controls the iterative repetition of the aforementioned globally bidirectional coupled closed loop, continuously traversing and changing environmental input conditions (such as wind speed) and multi-vehicle spatial flow characteristics (vehicle spacing). The iteration terminates immediately when the dynamics solver detects that the main vehicle has reached the critical failure threshold in any dimension. This application breaks through the traditional limitation of focusing only on rollover, constructing a full-envelope failure judgment criterion covering "side slip-side slide-rollover": Lateral yaw failure assessment (core consideration for multi-vehicle interference): This focuses on the moment when multiple vehicles are driving side-by-side or overtaking. When the flow field obstruction suddenly disappears or the wake distortion occurs, the lead vehicle will experience extreme transient lateral force changes. The system closely monitors the lead vehicle's yaw response and calculates the dynamic yaw threshold defined by the road surface adhesion limit in real time. ,in This is the acceleration due to gravity. When the real-time yaw rate is monitored... ( This is a correction factor, with a range of values. ), and the yaw rate and its derivative satisfy the same-sign divergence criterion ( ). If the duration exceeds the preset time window, it is determined to be a yaw rate divergence. When the yaw rate divergence or the transient lateral offset exceeds the safe physical boundary of the lane, it is determined to be a lateral slip failure.

[0073] Sideslip failure detection: When the main vehicle undergoes translation and yaw under the sudden change of strong aerodynamic lateral force, the system monitors the degree of lateral slip in real time. Specifically, it extracts the longitudinal velocity of the center of mass output by the dynamics solver. Lateral velocity and yaw rate Combined with the geometric distance parameters from the center of mass to the front and rear axes and The transient lateral slip ratio (i.e., slip angle, such as the front wheel slip angle) of each wheel is calculated in real time using rigid body kinematics equations. Subsequently, the system substitutes this transient lateral slip angle into a nonlinear tire dynamics model (such as the magic formula). When the system detects that the sideslip angle of any axle exceeds the current road adhesion coefficient... The limit friction envelope boundary, defined by the vertical load on the wheel, causes the tire lateral force to enter the fully saturated slip zone, which indicates that the vehicle tire has completely lost its physical tracking ability and is immediately judged as a sideslip failure.

[0074] Rollover Failure Determination: Sudden changes in lateral tilt moment under strong crosswind conditions can easily induce rollover accidents in vehicles with a high center of gravity (such as trucks). This application abandons the traditional static geometric rollover threshold and introduces the normal load transfer rate as a dynamic physical index for transient rollover determination. Specifically, within each time step of the co-simulation, the dynamic solver extracts the sum of the normal vertical loads on the left tire of the main vehicle on the same axle in real time. The sum of the normal vertical loads of the right tire This vertical load not only includes the conventional lateral mass transfer caused by centrifugal force, but also deeply couples with the nonlinear suspension compression deformation induced by the transient aerodynamic roll moment transmitted by the fluid solver. (The formula is used to...) This metric is calculated in real time. When the Lateral Load Transfer Ratio (LTR) is detected to approach or even exceed the preset extreme safety boundary (e.g., ...), When a wheel on one side of the vehicle is about to or has completely lost its normal ground support and is facing its rollover limit, the system immediately determines that the vehicle has rolled over and terminates the current iteration.

[0075] This application precisely isolates the steady-state environmental variables and vehicle-specific variables (the vehicle-specific variables strictly include the initial vehicle speed) at the instant that triggers any of the above extreme operating conditions. Dynamic axis weight matrix Reverse analysis of centroid height and wheelbase allocation mapped from the vehicle model library. With suspension roll stiffness The system uses a set of parameters representing the inherent physical boundaries of the vehicle, along with the dynamic spatial pose features of multiple vehicles, to package these into a high-dimensional feature vector. Simultaneously, the resulting joint instability probabilities (side slip probability, sideslip probability, and rollover probability are all treated as independent dimensions, forming a multi-dimensional label vector, also known as multi-dimensional dynamic instability probability) are used as supervision labels. Through tens of thousands of extreme operating conditions, a structured offline training dataset is finally output. It is used to pre-train online XGBoost models, enabling them to still output multi-dimensional risk probabilities that conform to complex multi-vehicle aerodynamics in seconds during the online phase without fluid simulation intervention.

[0076] Therefore, step 104 above can be replaced by steps 301 to 308.

[0077] Step 301: Establish a fluid computing domain covering the spatial topology of the cross-sea traffic cluster. The fluid computing domain includes the entire process of a car entering the bridge, driving on the bridge, entering the island, driving on the island, entering the tunnel, and exiting the tunnel.

[0078] Step 302: Establish a transient aerodynamic analysis model using large eddy simulation.

[0079] Step 303: In the multibody dynamics domain, instantiate in parallel to establish a nonlinear multi-degree-of-freedom model of the target vehicle and multiple adjacent vehicles.

[0080] Step 304: Read in the initial wind field boundary conditions, perform steady-state flow field calculations in the fluid computation domain, and after the flow field converges, use the converged flow field as the initial environmental variable for transient two-way coupling.

[0081] Step 305: Preset different combinations of environmental input conditions and multi-vehicle spatial flow characteristics.

[0082] Step 306: Under each combination, based on the initial environmental variables, perform a two-way coupled data interaction iterative simulation using the transient aerodynamic analysis model and the nonlinear vehicle multi-degree-of-freedom model.

[0083] Step 307: During the simulation process under each combination, when the target vehicle and the adjacent vehicle are detected to have reached the critical failure threshold in any dimension of sideslip, sideslip, or rollover, the iteration terminates, the extreme condition is determined, the state constraint vector that triggers the extreme condition is recorded, and the joint instability probability of sideslip, sideslip, and rollover is calculated.

[0084] Step 308: Take the state constraint vector that triggers the extreme working condition as input, and the joint instability probability of lateral slip, sideslip, and rollover as supervision labels to form a training dataset under extreme working conditions.

[0085] In another exemplary embodiment of this application, step 105 abandons the traditional time-consuming online hydrodynamic grid calculation and innovatively adopts an extreme gradient boosting tree as the core nonlinear mapping carrier. This network integrates multiple decision trees and utilizes second-order Taylor expansion to jointly optimize the loss function and regularization penalty term, thus possessing extremely strong high-dimensional fitting and generalization capabilities for complex physical spaces.

[0086] Benchmark risk simulation: Using a millisecond-level refresh rate, the heterogeneous physical parameters output from step one are extracted and homogenized in real time to construct a high-dimensional state constraint vector characterizing the current transient coupled environment of "wind-vehicle-road-flow". .

[0087] .

[0088] The state constraint vector Feed via offline high-fidelity physical twin dataset A fully pre-trained XGBoost network, which is abstracted as a high-dimensional nonlinear transfer function. Let the initial speed of the target vehicle be... Through forward propagation, the probability of baseline sideslip or sideslip instability risk in the initial state can be solved very quickly. (That is, the ideal failure probability assuming the target vehicle is in an absolutely isolated state, after removing the multi-vehicle flow field distortion): .

[0089] In another exemplary embodiment of this application, step 106 above involves dynamic compensation for aerodynamic penalties: to instantly restore the physical destructive force of multi-vehicle aerodynamic interference with extremely low computational overhead, a highly parameterized "analytical aerodynamic mutation penalty proxy model" is introduced. High-frequency polling of the flow state flag (State) is used to perform dynamic compensation routing (preferred embodiment). If the perceived state is State = 0 (single vehicle free flow state), it is determined that there is no significant flow field distortion in the current space, and the penalty factor is locked. .

[0090] If the perceived state is State = 1 or 2 (sparse intersection or dense following flow), a nonlinear penalty operator based on the spatial attenuation law is instantaneously activated. This operator profoundly characterizes the physical attenuation mechanism of wake and blockage: aerodynamic threat varies with spatial topological distance ( ) decays exponentially with the relative velocity difference ( It exhibits a logarithmic increase. Its characteristic equation is expressed as: .

[0091] Note: The dimensionless evolution hyperparameter in the formula... All have been rigorously calibrated through multivariate nonlinear regression fitting of cross-domain fluid dynamics interaction data.

[0092] Finally, the system performs probability compensation through multiplicative coupling, outputting the transient limit risk probability of the target vehicle under the current real physical topology. : .

[0093] In another exemplary embodiment of this application, step 107 above relates to a discrete iterative descent algorithm with constrained physical priors. A preset safety decision redline threshold representing the lower limit of management tolerance is established. (In this embodiment, the classification threshold is preset to) ).

[0094] 1) If the assessment If the current driving conditions are within the absolutely safe physical envelope, then the target safe speed is determined to be... The algorithm terminates silently.

[0095] 2) If the assessment This instantly triggers the inversion optimization mechanism of the safety boundary. Given the lack of gradients due to the non-differentiability of the tree model, this application innovatively introduces a one-way discrete stepwise heuristic search algorithm based on vehicle dynamics priors (i.e., instability risk strictly monotonically non-decreasing with vehicle speed): Automatically lock the target vehicle's legal minimum speed limit With the current speed limit Construct a dynamic constraint optimization boundary. Set a fixed discrete deceleration step size. Construct a unidirectional discrete descent sequence. The next iteration's downward test vehicle speed and the corresponding reassessment risk probability Strictly obey the following recurrence relation: ; .

[0096] In the formula, For the first The next iteration's downward test vehicle speed.

[0097] This process eliminates the complexities of gradient backpropagation, relying solely on the computational advantage of forward model calculations for high-frequency reassessment of the safety state. Iterative optimization continues to converge until the safety threshold is first reached in the test sequence. Mathematically, the goal is to find the optimal number of convergence steps that satisfy the constraints. : .

[0098] When the convergence condition is triggered, the iteration is truncated. The highest legal vehicle speed determined at this point is then analyzed in reverse as the critical safe vehicle speed of the high-risk vehicle under the current transient multiphysics coupling. : .

[0099] Therefore, the implementation process of step 107 above can be summarized as follows: construct the target vehicle's legal minimum speed limit and the current initial observed vehicle speed as a dynamic constraint optimization boundary; construct a unidirectional discrete descent sequence from within the dynamic constraint optimization boundary according to a fixed discrete deceleration step size; sequentially select a test speed from the unidirectional discrete descent sequence, and then apply the formula... The process is iterated to obtain the reassessment risk probability for each iteration. When the reassessment risk probability first equals the safety decision red line threshold, the iteration stops, and the downward test speed used in the current iteration is determined as the critical safe speed.

[0100] The "analyzed aerodynamic penalty proxy model" and "discrete iterative descent algorithm with limited physical priors" established in this application transform the heuristic dynamic adjustment process of safe vehicle speed into rigorous mathematical control logic.

[0101] Steps 105 to 107 above constitute online real-time prediction and inversion of safe vehicle speed.

[0102] In another exemplary embodiment of this application, step 108 involves hierarchical decision and control instruction generation. This step is responsible for reversing the parsed physical safety boundary (critical safe speed). This is transformed into concrete, executable spatial cooperative control semantics for multi-agent traffic flow. The system is based on real-time vehicle speed. With critical safe speed Based on the deviation gradient, a three-level dynamic risk envelope is constructed, and differentiated control instructions are adaptively calculated: 1. Low-risk envelope (steady-state cruise zone): When the observed state (real-time speed of the target vehicle) meets the following conditions... Time (in the formula) For example, take the preset safety margin coefficient. The system determines that the target vehicle's current dynamic response is within the steady-state safety domain. The control logic keeps the system silent, without generating new multi-dimensional cooperative control commands, or only generating "adaptive cruise control suggestion commands" to maintain the current vehicle speed, maximizing the release of the infrastructure traffic potential of the cross-sea traffic cluster while ensuring basic safety.

[0103] 2. Medium-risk envelope (cooperative intervention zone): When the observed state approaches the critical boundary, i.e. First, an instantaneous forced speed limit control primitive is generated to forcibly restrict the target vehicle's desired longitudinal speed to a specific value. Then, the flow regime flags obtained in step one are deeply fused for spatial intervention. If the current state is State = 1 (sparse intersection flow) and a significant overtaking intention from the following vehicle is detected, a "spatial arrangement constraint instruction" is forcibly calculated and superimposed to generate—that is, issuing "parallel overtaking is strictly prohibited," "lane keeping is mandatory," and "reconstructing and maintaining a longitudinal aerodynamic safe distance of at least..." The system sends a collaborative message to the target vehicle. Finally, if the roadside sensor confirms that the target vehicle is currently in the windward lane of the cross-sea bridge and the crosswind area is extremely large, an additional "recommended trajectory instruction" is generated to change lanes to the leeward lane, actively utilizing the bridge's own wind-resistant flow field distortion zone to achieve passive risk avoidance.

[0104] The "overtaking intention" is not a subjective prediction, but rather a continuous trajectory matrix based on the fusion output of millimeter-wave radar and a visual camera. The millimeter-wave radar detects at high frequency that the relative speed difference between the following vehicle and the target vehicle is greater than 0 (i.e., the following vehicle's speed is greater than the target vehicle's), and the transient collision time (TTC) calculated from the relative distance and relative speed is less than a preset overtaking and following safety threshold, indicating that the following vehicle is in a state of aggressive longitudinal approach. Simultaneously, the visual camera, through lane topology tracing, detects a continuous lateral shift in the center coordinates of the following vehicle's two-dimensional bounding box. When the lateral movement rate of the following vehicle... When the lateral deviation threshold is exceeded and the relative lateral distance from the center line of the target overtaking lane shows a monotonically decreasing trend within a set time window (i.e., the vehicle is leaning towards the outer lane or crossing the line), it indicates that a lateral lane change action has been initiated. When the fusion perception system simultaneously captures the above-mentioned longitudinal approach and lateral expansion objective physical quantities exceeding the threshold within N consecutive frames, it determines in engineering logic that "the following vehicle has a significant relative overtaking intention".

[0105] 3. High-risk envelope (extreme blocking zone): The predicted final risk probability of a target vehicle at extremely low legal speeds due to extremely light load, excessive height, or encountering extreme gusts of wind. There is still a high probability that the safety threshold will be breached. When it is determined that conventional speed reduction is insufficient to curb the physical tendency of skidding / rollover, a highest-priority "absolute prohibition" control command will be generated, or an emergency escape route will be forcibly calculated to guide the target vehicle to the nearest emergency escape lane / windbreak island for parking, thus completely breaking the chain of disaster in physical space.

[0106] In another exemplary embodiment of this application, step 109 above involves VMS-based visual target addressing (for non-connected vehicles): First, the generated integrated cooperative control semantics (critical safe speed) are... The system deeply reconstructs the data into a vectorized visual data stream containing "unique license plate ID + strong warning-level spatial constraint semantics," considering factors such as lane preference and anti-parallelism spatial constraints. Then, the edge scheduling node uses the real-time dynamic spatiotemporal coordinates of the target high-risk vehicle (…). It incorporates a dynamic heading and line-of-sight model. This dynamic heading and line-of-sight model is based on the target vehicle's current longitudinal speed. The feedforward quantity is used to calculate the dynamic pre-aiming range. (in For aiming time gain, (This is the minimum safe sight distance constant), combined with the vehicle's transient yaw angle. This meticulously constructs the vehicle's expected driving trajectory and the driver's effective forward visual cone area within a very short timeframe. The specific process involves: first, obtaining the target vehicle's current physical centroid coordinates. Longitudinal speed With transient yaw angle Based on aiming time gain Calculate the dynamic pre-aiming distance Then, assume that the target vehicle maintains a transient yaw rate within a preset time window in the future. Based on the CTRV (Constant Turning Rate and Speed) kinematic model, the expected trajectory curve of the target vehicle is generated by integration. :when At that time, its trajectory coordinates satisfy ,as well as , where is exactly the currently obtained physical centroid coordinate of the target vehicle. Further considering the driver's viewpoint offset, taking the transient yaw heading angle as the central visual axis, the effective central viewing angle of human vision is introduced ( is a constant, which is calibrated based on the visual limit in traffic ergonomics that allows the driver to clearly read roadside information without turning the head, and usually has a value range of ±15° to ±30° on both sides of the vehicle heading centerline) to define the left and right visual boundary rays, and taking the dynamic preview sight distance as the truncation radius, a fan-shaped effective forward visual cone polygon is enclosed and generated in the global two-dimensional coordinate system . Finally, the set of candidate VMS physical nodes on the front road section is traversed, and the intersection operation is performed through the spatial polygon point inclusion operator. The system retrieves the VMS node that not only falls within the effective forward visual cone polygon but also has the smallest vertical projection distance to the expected trajectory curve . This node is identified as the optimal visual targeting physical screen, and the system immediately directs the customized warning data to it for targeted distribution.

[0107] Based on the solved mapping between the trajectory and the visual cone area, the optimal variable message sign physical node in front of the driving trajectory of the vehicle is accurately locked and occupied. Finally, the VMS node performs directional visual projection with high brightness and high-frequency flashing (for example, targeted display: "Min A·XXXXX, extreme crosswind risk, mandatory speed limit 40, no parallel overtaking"). By forcibly awakening the visual cognitive center of the human driver, dynamic redistribution of road surface physical right-of-way and flexible reconstruction of spatial safety topology are realized.

[0108] Therefore, the process of step 109 described above can be summarized as: reconstructing the multi-dimensional cooperative control command into a vectorized visual data stream with unique license plate ID and warning-level spatial constraint semantics; according to the current longitudinal speed of the target vehicle, using the formula to solve the dynamic preview sight distance; according to the dynamic preview sight distance and the transient yaw heading angle of the target vehicle, constructing the expected driving trajectory of the target vehicle within the future preset time and the effective forward visual cone area of the driver; according to the expected driving trajectory and the effective forward visual cone area, locking and occupying the optimal variable message sign in front of the driving trajectory of the target vehicle; controlling the optimal variable message sign to execute the directional visual projection of the vectorized visual data stream.

[0109] In another exemplary embodiment of the present application, the above step 110 involves physical direct connection based on C-V2X (Cellular Vehicle-to-Everything, cellular vehicle-to-everything) (for intelligent connected vehicles): first, the roadside edge computing unit converts the exclusive critical safe vehicle speed of the target vehicle The system incorporates a multi-vehicle spatial arrangement constraint matrix, encapsulated into low-latency control messages conforming to the underlying communication protocol stack (such as the BSM / SPaT standard protocol). Millisecond-level point-to-point broadcasting is implemented via PC5 direct air interface communication, and the target vehicle's onboard communication terminal confirms receipt. The control messages are then seamlessly transmitted to the intelligent vehicle's underlying domain controller. After parsing the instructions, the vehicle's underlying control module directly takes over or intervenes at a higher authority in the vehicle's longitudinal throttle opening / brake hydraulic pressure, as well as the lateral electric power steering system's damping torque. This achieves a leapfrog safety closed loop from "passive reception of visual warnings" to "active machine-level physical risk avoidance," nipping wind-related driving risks in the bud at the millisecond level of underlying physical response.

[0110] The above steps 109 and 110 innovatively construct a dual-channel, asymmetric cyber-physical fusion intervention network based on "macroscopic visual addressing" and "microscopic wire control direct connection", ensuring that the calculated multi-dimensional collaborative control commands can be accurately delivered and rigidly executed.

[0111] This application aims to break through the technological limitations of existing traffic control systems in high-wind areas by establishing a robust patent protection network through technological innovations in the following four dimensions: 1. Key protected bridges' intelligent traffic hierarchical management and control methods and procedures.

[0112] 2. This application focuses on protecting the spatial topology parameters of multiple vehicles analyzed by radar / vision ( The vehicle dynamic ontology features (axle load distribution, centroid height prediction) extracted by WIM are synchronously registered in a unified spatiotemporal coordinate system, and micro-meteorological data and road surface friction coefficient are integrated to construct a high-dimensional transient state constraint vector. The system-level underlying perception data bus architecture.

[0113] 3. This application focuses on protecting the use of exponentially decaying "penalty factors" in the forward inference of nonlinear machine learning networks, based on real-time traffic flow dynamic activation. "Based on the physical mapping mechanism of targeted correction of basic risk probabilities; and overcoming the non-differentiability limitation of tree models, a one-way discrete step-by-step search algorithm based on physical safety priors is used to reverse-close the loop and find the critical safe speed specific to different vehicle characteristics under the constraint of the safety red line." The core mathematical control logic.

[0114] 4. This application focuses on protecting the technical solutions for adaptively calculating horizontal / vertical spatial arrangement constraint commands such as "anti-parallel overtaking" and "forced vehicle spacing reconstruction" based on flow characteristics and risk gradients; and the cyber-physical systems (CPS) that utilize VMS and C-V2X / OBU to implement dual-channel, asymmetric differentiated closed-loop intervention.

[0115] 5. As the underlying data foundation of this application, the key protection focuses on the data interaction and dynamic mesh boundary reconstruction technology of aerodynamic load and pose response performed within a very small discrete time step through the fluid computation domain and multibody dynamics domain. This technology automatically exhaustively enumerates and removes critical instability features under extreme multi-vehicle aerodynamic interference conditions, thereby legally and efficiently constructing a high-fidelity nonlinear mapping training set. Engineering-based virtual reconstruction methods.

[0116] Compared with existing traditional traffic control technologies that heavily rely on macroscopic wind speed lookup tables, this application, through system-level architecture reconstruction and underlying algorithm innovation, has produced the following extremely significant and unexpected technical effects: 1. Improved the accuracy of risk warning in complex traffic environments. Unlike traditional single-dimensional wind speed limits, this application integrates environmental wind field, road friction, dynamic axle load, center of gravity height and other intrinsic characteristics to solve the problem of different risk control for vehicles with different load characteristics under the same wind conditions, and improves the targeting.

[0117] 2. It makes up for the lack of control over aerodynamic interference of multiple vehicles. By introducing a flow pattern discrimination mechanism and aerodynamic change penalty factor based on the spatiotemporal topology of multiple vehicles, it quantifies the additional risks caused by wake distortion and aerodynamic shielding effects. It can proactively implement spatial collaborative intervention before parallel or overtaking conditions occur, reducing the probability of instability accidents caused by aerodynamic changes.

[0118] 3. Balancing prediction accuracy and real-time online computation, a decoupled architecture is adopted, which uses offline physical co-simulation to generate datasets to train online machine learning models. This preserves the accuracy of fluid dynamics calculations in physically mapping extreme conditions while avoiding the problem of excessively long online simulation times, thus meeting the millisecond-level response requirements of traffic control systems.

[0119] 4. Differentiated closed-loop management for mixed traffic flow has been achieved. At the execution layer, a dual-channel system based on VMS macroscopic visual guidance and C-V2X microscopic communication intervention has been constructed. This system can not only provide accurate visual cues to traditional vehicles, but also directly transmit control messages to intelligent connected vehicles to assist in dynamic takeover, thereby improving the applicability and collaborative control efficiency of the system.

[0120] Based on the same inventive concept, this application also provides a smart traffic classification and control system for cross-sea traffic clusters in strong wind areas, used to implement the aforementioned smart traffic classification and control method for cross-sea traffic clusters in strong wind areas. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the smart traffic classification and control system for cross-sea traffic clusters in strong wind areas provided below can be found in the limitations of the smart traffic classification and control method for cross-sea traffic clusters in strong wind areas described above, and will not be repeated here.

[0121] In one exemplary embodiment, such as Figure 3 As shown, a smart traffic hierarchical control system for cross-sea traffic clusters in strong wind areas is provided, comprising: a holographic perception input layer, an offline model calculation layer, an online risk prediction layer, a hierarchical decision-making logic layer, and a collaborative release execution layer.

[0122] The holographic perception input layer is used for spatiotemporal registration and feature extraction of multi-source heterogeneous physical parameters; the offline model computation layer is used for digital twin and dataset generation based on fluid-structure interaction; the online risk prediction layer is used for real-time online safe vehicle speed prediction and inversion; the hierarchical decision-making logic layer is used for hierarchical decision-making and control command generation; and the collaborative release execution layer is used for collaborative information release.

[0123] The online risk prediction layer is deployed on the online edge computing nodes or cloud decision-making center of the cross-sea navigation cluster, forming the "real-time inference engine" of the entire intelligent management and control system.

[0124] The hierarchical decision-making logic layer, acting as the "control law synthesizer" of the entire intelligent management and control system, is responsible for reversing the analysis of the physical security boundary derived in step three. This translates into concrete, executable spatial collaborative control semantics for multi-agent traffic flow.

[0125] The collaborative release execution layer, as the "execution end" projected from the cloud-based collaborative control law to the physical world, aims to overcome the engineering reality of uneven vehicle connectivity penetration rates in cross-sea traffic clusters.

[0126] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores multi-dimensional collaborative control instructions. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a smart passage hierarchical management method for cross-sea passage clusters in strong wind areas.

[0127] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment to which the present application is applied. Specific computer equipment may include, for example, [the following is a list of possible additional structures]. Figure 4 The embodiments show more or fewer components, combinations of certain components, or different component arrangements. In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, which the processor executes to implement the steps in the above-described method embodiments.

[0128] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0129] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0130] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0131] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0132] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0133] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for intelligent hierarchical control of cross-sea traffic clusters in strong wind areas, characterized in that, include: Acquire multi-source heterogeneous perception features of target vehicles when they are traveling in cross-sea traffic clusters; The multi-source heterogeneous sensing features include environmental domain features, vehicle ontology domain features, and vehicle cooperative domain features; Based on the characteristics of the vehicle cooperative domain, a spatial flow state discrimination mechanism based on physical topology is used to perform quantitative flow state discrimination and determine the flow state flag. After spatiotemporal registration of the multi-source heterogeneous sensing features, they are reconstructed together with the flow state flags into a state constraint vector; By using bidirectional time-varying coupled simulation of computational fluid dynamics and multibody dynamics, a training dataset under extreme conditions is obtained, and the extreme gradient boosting tree is pre-trained using the training dataset to obtain a pre-trained extreme gradient boosting tree. The reconstructed state constraint vector is input into the pre-trained extreme gradient boosting tree, and the joint instability probability in the initial state is output. Based on the flow regime flags, an analytical aerodynamic mutation penalty surrogate model is used to determine the penalty factor, which is then multiplicatively coupled with the joint instability probability in the initial state to obtain the transient limit risk probability of the target vehicle. The analytical aerodynamic mutation penalty surrogate model is: Characteristic Equation In the formula, As a penalty factor, All are dimensionless evolutionary hyperparameters. The topological distance in two-dimensional space. The difference is the relative speed. If the transient limit risk probability is greater than the safety decision red line threshold, a one-way discrete step forward heuristic search algorithm based on vehicle dynamics prior is used to determine the critical safe vehicle speed. Based on the deviation gradient between the real-time vehicle speed and the critical safe vehicle speed of the target vehicle, the multi-dimensional cooperative control command is adaptively calculated. If the target vehicle is a non-connected vehicle, then according to the multi-dimensional collaborative control instructions, the target vehicle will be subjected to macroscopic visual target addressing and flexible control based on variable message signs. If the target vehicle is a connected vehicle, then according to the multi-dimensional collaborative control command, the target vehicle will be subjected to rigid takeover of the micro-level drive-by-wire chassis based on cellular vehicle networking.

2. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, Based on the characteristics of the vehicle cooperative domain, a physical topology-based spatial flow regime discrimination mechanism is used to perform quantitative flow regime discrimination and determine flow regime flags, including: If the target vehicle and all adjacent vehicles within the perception range satisfy the following conditions: or If the target vehicle is outside the aerodynamic wake and obstruction zone, the flow status flag is set to 0; where, For the target vehicle Adjacent vehicles within the perception range The relative longitudinal distance between vehicles, For vertical security threshold, For the target vehicle Adjacent vehicles within the perception range The relative lateral spacing, The threshold for lateral aerodynamic influence; If the target vehicle and any adjacent vehicle within the perception range satisfy the following conditions: , and This indicates that the target vehicle is undergoing a high-speed parallel overtaking or passing maneuver with an adjacent vehicle, and the flow status flag is set to 1; where, For the target vehicle Adjacent vehicles within the perception range The relative speed difference The threshold for abrupt changes in relative velocity; If the target vehicle and any adjacent vehicle within the perception range satisfy the following conditions: , and This indicates that the convoy follows the car, and the following car remains in the turbulent wake of the preceding car for a preset period of time. The flow status flag is set to 2.

3. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, Through bidirectional time-varying coupled simulations of computational fluid dynamics and multibody dynamics, a training dataset under extreme conditions was obtained, including: Establish a fluid computing domain that covers the spatial topology of the cross-sea traffic cluster; the fluid computing domain includes the entire process of a car entering the bridge, driving on the bridge, entering the island, driving on the island, entering the tunnel, and exiting the tunnel. A transient aerodynamic analysis model was established using large eddy simulation. In the multibody dynamics domain, a nonlinear multi-degree-of-freedom model of the target vehicle and multiple adjacent vehicles is established by parallel instantiation. The initial wind field boundary conditions are read in, and steady-state flow field calculations are performed in the fluid computation domain. After the flow field converges, the converged flow field is used as the initial environmental variable for transient two-way coupling. Preset different combinations of environmental input conditions and multi-vehicle spatial flow characteristics; Under each combination, simulations are performed using a two-way coupled data interaction iteration model based on the initial environmental variables and a transient aerodynamic analysis model and a nonlinear vehicle multi-degree-of-freedom model. During the simulation of each combination, when the target vehicle and the adjacent vehicle are detected to have reached the critical failure threshold of any dimension of sideslip, sideslip, or rollover, the iteration terminates, the extreme condition is determined to have occurred, the state constraint vector that triggers the extreme condition is recorded, and the joint instability probability of sideslip, sideslip, and rollover is calculated. The state constraint vector that triggers the extreme working condition is used as input, and the joint instability probability of lateral slip, sideslip, and rollover is used as supervision label to form a training dataset under extreme working conditions.

4. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, Based on the aforementioned flow regime flags, the penalty factor is determined using an analytical aerodynamic mutation penalty surrogate model, including: If the flow regime flag is 0, it is determined that there is no flow field distortion in the current space, and the penalty factor is locked to 1. If the flow regime flag is 1 or 2, then the characteristic equation is used. Determine the penalty factor.

5. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, Also includes: If the transient limit risk probability is less than or equal to the safety decision red line threshold, then the current driving condition is determined to be within the safe physical envelope, and the target vehicle's current initial observed speed is the target safe speed.

6. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, If the transient limit risk probability is greater than the safety decision red line threshold, a one-way discrete step-by-step heuristic search algorithm based on vehicle dynamics priors is used to determine the critical safe speed, including: The legal minimum speed limit of the target vehicle and the current initial observed vehicle speed are used to construct a dynamic constraint optimization boundary. A unidirectional discrete descent sequence is constructed within the dynamic constraint optimization boundary using a fixed discrete descent step size. From the unidirectional discrete descent sequence, sequentially select a test speed for the downward movement, and then follow the formula... The process is iterated to obtain the reassessment risk probability for each iteration; where, For the first The probability of reassessment risk in the next iteration. The nonlinear transfer function is abstracted from the pre-trained extreme gradient boosting tree. For the state constraint vector, For the first The test speed for the next iteration of the downward probe. As a penalty factor; When the reassessed risk probability first equals the safety decision red line threshold, the iteration stops, and the downward test speed used in the current iteration is determined as the critical safe speed.

7. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, Based on the deviation gradient between the target vehicle's real-time speed and the critical safe speed, multi-dimensional cooperative control commands are adaptively calculated, including: When the real-time speed of the target vehicle meets When the target vehicle's current dynamic response is determined to be within the steady-state safety domain, no new multi-dimensional cooperative control commands are generated, or only an adaptive cruise control suggestion command to maintain the current vehicle speed is generated; where, The real-time speed of the target vehicle. The preset safety margin coefficient, The critical safe speed; When the real-time speed of the target vehicle meets First, a mandatory speed limit control primitive is generated to force the target vehicle's longitudinal desired speed to be clamped at the critical safe speed. Then, when the flow flag is equal to 1 and the vehicle behind is detected to have the intention to overtake, a collaborative message of spatial arrangement constraint instruction is generated. Finally, if the roadside perception confirms that the target vehicle is currently in the windward lane of the cross-sea bridge and the crosswind area is greater than the area threshold, an additional recommended trajectory instruction to change lanes to the leeward lane is generated to actively utilize the windproof flow field distortion area of ​​the cross-sea bridge itself to achieve passive risk avoidance. Predict the transient limit risk probability of the target vehicle at the legal minimum speed limit. If the transient limit risk probability at the legal minimum speed limit is still greater than the safety decision red line threshold, generate the highest priority absolute prohibition control command, or forcibly solve the avoidance path and guide the target vehicle to drive into the nearest emergency escape zone / windbreak island for parking.

8. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, Following multi-dimensional collaborative control commands, macroscopic visual targeting and flexible control of the target vehicle based on variable message signs are performed, including: Multidimensional collaborative control commands are reconstructed into a vectorized visual data stream containing a unique license plate ID and warning-level spatial constraint semantics; Based on the target vehicle's current longitudinal speed, using the formula Calculate the dynamic pre-aiming range; where, For dynamic pre-aiming of the line of sight, For aiming time gain, The current longitudinal speed of the target vehicle, It is the minimum safe sight distance constant; Based on the dynamic pre-aiming sight distance and the transient yaw angle of the target vehicle, the expected driving trajectory of the target vehicle and the driver's effective forward visual cone area are constructed within a preset time period in the future. Based on the expected driving trajectory and the effective forward visual cone area, lock onto and seize the best variable message sign in front of the target vehicle's driving trajectory. The optimal variable information board is controlled to perform directional visual projection of the vectorized visual data stream.

9. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, According to the multi-dimensional collaborative control commands, the target vehicle undergoes micro-level drive-by-wire chassis rigid takeover based on cellular vehicle-to-everything (V2X) technology, including: The critical safe speed of the target vehicle and the multi-dimensional collaborative control commands are encapsulated into control messages that conform to the underlying communication protocol stack. Through PC5 direct air interface communication, control messages are broadcast point-to-point at the millisecond level and received by the vehicle's onboard communication terminal, and finally passed through to the target vehicle's underlying domain controller. After parsing the control message in the lower-level domain controller and obtaining the control command, it directly takes over or intervenes in the longitudinal motion control quantity of the target vehicle and the damping torque of the lateral electric power steering system; the longitudinal motion control quantity is the longitudinal throttle opening or the longitudinal brake hydraulic pressure.

10. The intelligent traffic hierarchical management method for cross-sea traffic clusters in strong wind areas according to claim 1, characterized in that, Also includes: By using the real-time data quality monitoring module deployed on the roadside edge computing node, the online availability assessment and health verification of the acquired multi-source heterogeneous sensing features are performed. When the multi-source heterogeneous sensing features are detected to have missing data, outlier overflow, or time domain refresh delay, the three-layer cascaded fault tolerance and degradation control strategy is adaptively activated. The three-tiered fault-tolerant degradation control strategy includes: a first-tier ontology perception degradation strategy, a second-tier environment perception degradation strategy, and a third-tier all-hazard red line cutoff strategy. Ontology perception degradation strategy: If the dynamic axle redistribution matrix or centroid height data in the vehicle ontology domain features are missing, the vehicle model physical prior database is called to retrieve the vehicle model classification corresponding to the target vehicle, extract the maximum mass envelope under full load and the highest centroid envelope under no load corresponding to the vehicle model classification, and use them as alternative physical parameters to reconstruct the state constraint vector. Environmental perception degradation strategy: If the refresh delay of the effective environmental wind speed data in the environmental domain features exceeds the preset time threshold, the maximum effective gust peak value within the current preset sliding window time will be automatically locked and used as the input environmental wind speed at the current time to reconstruct the state constraint vector. Full-hazard red line cutoff strategy: If the offline rate of global roadside perception devices in the roadside edge computing node is greater than the preset number of red lines, or the online inference time of the extreme gradient boosting tree exceeds the timeliness threshold, then actively exit the macroscopic visual targeted addressing based on variable message signs and the microscopic wire-controlled chassis rigid takeover based on cellular vehicle networking, and directly downgrade and issue a macroscopic static speed limit red line command for the entire road section or physically shut down the right of passage of the bridge.

Images