Method and platform for dynamically identifying accident hidden danger road section based on multi-dimensional data fusion

Abstract

The application provides an accident hidden danger road section dynamic identification method and platform based on multi-dimensional data fusion, relates to the technical field of traffic data identification, and comprises the following steps: converting a target road section area into a directed graph to determine a road network topology; generating a normal situation partition traffic trajectory through multi-dimensional road network data, judging a single vehicle trajectory and a traffic disturbance mode in combination with a real-time traffic trajectory, and generating a hidden danger identification instruction; positioning a hidden danger road section in the road network topology, performing combined inversion and state transition probability analysis based on random variables, and generating a hidden danger probability curve; performing hidden danger disturbance propagation analysis in the road network topology, determining hidden danger identification information, and performing traffic hidden danger early warning management. The application can solve the technical problem of affecting the accuracy of traffic hidden dangers in the prior art, realize dynamic hidden danger identification and prediction capability based on real-time traffic flow data and road network topology information, and achieve the technical effect of accurately identifying the space-time distribution of hidden dangers.

Description

Technical Field

[0001] This application relates to the field of traffic data recognition technology, and in particular to a method and platform for dynamic identification of accident-prone road sections based on multi-dimensional data fusion. Background Technology

[0002] With the continuous development of intelligent transportation systems, traffic safety has gradually become an important issue that urgently needs to be addressed in public transportation management and vehicle driving.

[0003] Currently, existing hazard identification technologies mainly rely on manually set rules or offline analysis-based models. These methods depend on static traffic data and simple judgment criteria, lacking the ability to monitor and respond to dynamic changes in traffic flow in real time. Furthermore, many existing technologies fail to fully consider the interaction between road network topology and traffic flow, ignoring the complex connections between roads and the changing traffic conditions across different road segments and time periods. In addition, existing systems fail to effectively analyze the propagation of disturbances in traffic flow, often unable to predict the impact of hazards on downstream road segments, leading to a failure to detect potential chain reactions in a timely manner. Therefore, these technical solutions exhibit significant limitations when addressing hazard identification problems in complex traffic scenarios.

[0004] In summary, existing technologies suffer from the inability to accurately identify the spatiotemporal distribution of potential hazards due to the failure to track and reflect the propagation process of traffic flow disturbances in real time. This, in turn, affects the accuracy and timeliness of traffic hazard identification, further impacting the timeliness and effectiveness of traffic safety management decisions and making it impossible to cope with emergencies in complex and dynamic traffic environments. Summary of the Invention

[0005] The purpose of this application is to provide a method and platform for dynamic identification of accident-prone road sections based on multi-dimensional data fusion, in order to solve the technical problems in the existing technology that fail to accurately identify the spatiotemporal distribution of hazards due to the failure to track and reflect the propagation process of traffic flow disturbances in real time, thereby affecting the accuracy and timeliness of traffic hazards, further affecting the timeliness and effectiveness of traffic safety management decisions, and making it impossible to cope with emergencies in complex and dynamic traffic environments.

[0006] In view of the above problems, this application provides a method and platform for dynamic identification of accident-prone road sections based on multi-dimensional data fusion.

[0007] Firstly, this application provides a method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion, implemented through a multi-dimensional data fusion-based dynamic identification platform for accident-prone road sections. The method includes: performing directed graph transformation on the target road section area; determining the road network topology by calibrating the risk nesting structure and hazard response mode; based on the road network topology, deploying a behavior generator and a road judgment module at the road section edge; generating normal-condition zoned traffic trajectories from the retrieved multi-dimensional road network data; combining real-time traffic trajectories to judge single-vehicle trajectories and traffic flow disturbance patterns, generating a hazard identification command; activating the hazard identification module according to the hazard identification command; locating the hazard road section in the road network topology; generating a hazard probability curve by performing random variable-based combination inversion and state transition probability analysis; combining the hazard probability curve with hazard disturbance propagation analysis in the road network topology; determining hazard identification information; and distributing it to the vehicle platform port for traffic hazard early warning management.

[0008] Preferably, the method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion further includes: constructing a behavior generator using an adversarial training method, wherein the behavior generator takes random noise, road geometric features, and environmental conditions as input and normal traffic trajectories as output; building a road judgment module based on deviation judgment of normal traffic trajectories, wherein the road judgment module includes a first judgment threshold based on single-vehicle trajectories and a second judgment threshold based on stable disturbance aggregation under the convergence of traffic flow disturbance patterns in the same time and space; deploying the behavior generator and the road judgment module on the edge of the road section and establishing a data interaction port with the vehicle network.

[0009] Preferably, the method for dynamic identification of accident hazard road sections based on multi-dimensional data fusion further includes: retrieving multi-dimensional road network data at the edge of the road section, triggering the behavior generator, generating partitioned behavior according to the road network topology, and determining the partitioned traffic trajectory, wherein each partitioned traffic trajectory is identified with road section coordinates; reading real-time traffic trajectory based on multi-dimensional road network data, performing single-vehicle trajectory deviation judgment and deviation judgment based on stable disturbance aggregation features with the partitioned traffic trajectory, and generating a hazard identification instruction, wherein the existence of at least one deviation is used as the condition for instruction generation.

[0010] Preferably, the method for dynamic identification of accident hazard road sections based on multi-dimensional data fusion further includes: constructing a hazard identification module based on the road network topology, wherein the hazard identification module includes a first positioning layer, a second inversion layer and a third analysis layer; the hazard identification module receives the hazard identification instruction, locates the instruction road section in the road network topology, and determines multiple hazard road sections; for the multiple hazard road sections, performs combination inversion based on random variables and state transition probability analysis to generate hazard probability curves.

[0011] Preferably, the method for dynamic identification of accident hazard road sections based on multidimensional data fusion further includes: determining a combination of random variables through adversarial inversion based on the second inversion layer, wherein the combination of random variables is a combination of conditions that maximizes risk; determining a combination of real-time variables based on multidimensional road network data; generating a variable evolution sequence based on the linear transition from the real-time variable combination to the random variable combination; and performing Markov state transition probability analysis on the variable evolution sequence to determine the hazard probability curve.

[0012] Preferably, the method for dynamic identification of accident hazard road sections based on multi-dimensional data fusion further includes: abstracting the target road section area into a directed road graph, wherein intersections and key sections are nodes, and road sections are edges; defining a hazard response mode according to road geometry, wherein the hazard response mode includes sharp bend hazards, sight distance hazards, and road surface hazards; defining a risk nesting structure, wherein the risk nesting structure includes a regional level, a corridor level, and a point level; and identifying the directed road graph according to the risk nesting structure and the hazard response mode as the road network topology, wherein the road network topology is embedded in the behavior generator and the hazard identification module.

[0013] Preferably, the method for dynamic identification of accident hazard road sections based on multi-dimensional data fusion further includes: performing mechanical modeling of hazard disturbance propagation according to the road network topology to determine the traffic wave space, wherein the traffic wave space is jointly defined based on the disturbance propagation of vehicle interaction, density, and flow; locating the hazard source according to the hazard probability curve; and performing propagation analysis on the hazard source in the traffic wave space to determine the road network propagation heat map.

[0014] Preferably, the method for dynamic identification of accident hazard road sections based on multi-dimensional data fusion further includes: identifying the first hazard identification result according to the hazard probability curve, wherein the hazard point coordinates, confidence level, hazard type and real-time disturbance intensity are the identification elements; generating hazard identification information according to the first hazard identification result and the road network propagation heat map, and displaying the warning on the vehicle platform port.

[0015] Preferably, the method for dynamic identification of accident hazard road sections based on multi-dimensional data fusion further includes: receiving the hazard identification information at the vehicle platform port, establishing a vehicle coordinate system by retrieving real-time vehicle coordinates as the origin; generating a risk control warning instruction based on distance constraints based on the vehicle coordinate system, and executing traffic hazard warning management based on time series based on the hazard identification information.

[0016] Secondly, this application also provides a dynamic identification platform for accident-prone road sections based on multi-dimensional data fusion, used to execute the dynamic identification method for accident-prone road sections based on multi-dimensional data fusion as described in the first aspect, including: a road network topology determination module, used to perform directed graph transformation on the target road section area, and determine the road network topology by calibrating the risk nesting structure and hazard response mode; a hazard identification instruction generation module, used to deploy a behavior generator and a road judgment module on the edge side of the road section according to the road network topology, generate normal traffic trajectories by performing partitioning traffic trajectories on the retrieved multi-dimensional road network data, and judge single vehicle trajectories and traffic flow disturbance patterns by combining real-time traffic trajectories, and generate hazard identification instructions; a hazard probability curve generation module, used to activate the hazard identification module according to the hazard identification instructions, locate the hazard road section in the road network topology, and generate a hazard probability curve by performing combination inversion based on random variables and state transition probability analysis; and a hazard identification information determination module, used to perform hazard disturbance propagation analysis in the road network topology in combination with the hazard probability curve, determine hazard identification information, and send it to the vehicle platform port for traffic hazard early warning management.

[0017] The technical solution provided in this application has at least the following technical effects or advantages: by realizing the ability to dynamically identify and predict hidden dangers based on real-time traffic flow data and road network topology information, it achieves the technical effect of being able to monitor and reflect the disturbance propagation process in traffic flow in real time, accurately identify the spatiotemporal distribution of hidden dangers, and provide early warnings.

[0018] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion, as proposed in this application.

[0021] Figure 2This is a schematic diagram of the structure of the dynamic identification platform for accident-prone road sections based on multi-dimensional data fusion, as described in this application.

[0022] Explanation of reference numerals in the attached diagram: Module 1 for determining road network topology, Module 2 for generating hazard identification instructions, Module 3 for generating hazard probability curves, and Module 4 for determining hazard identification information. Detailed Implementation

[0023] This application provides a method and platform for dynamic identification of accident-prone road sections based on multi-dimensional data fusion. This addresses the technical problem in existing technologies where the failure to track and reflect the propagation process of traffic flow disturbances in real time leads to an inaccurate identification of the spatiotemporal distribution of hazards, thus affecting the accuracy and timeliness of traffic hazard identification and, consequently, the timeliness and effectiveness of traffic safety management decisions. It also fails to address the technical challenges of handling emergencies in complex and dynamic traffic environments. The application achieves dynamic hazard identification and prediction capabilities based on real-time traffic flow data and road network topology information, enabling real-time monitoring and reflection of the disturbance propagation process in traffic flow, accurate identification of the spatiotemporal distribution of hazards, and early warning.

[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0025] Example 1, please refer to the appendix. Figure 1 This application provides a method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion, which is applied to a dynamic identification platform for accident-prone road sections based on multi-dimensional data fusion. The method specifically includes the following steps:

[0026] A directed graph transformation is performed on the target road segment area, and the road network topology is determined by calibrating the risk nesting structure and hazard response mode.

[0027] Furthermore, this application also includes: abstracting the target road segment area into a directed road graph, wherein intersections and key sections are nodes, and road segments are edges; defining a hazard response mode based on road geometry, wherein the hazard response mode includes sharp bend hazards, sight distance hazards, and road surface hazards; defining a risk nesting structure, wherein the risk nesting structure includes a regional level, a corridor level, and a point level; and identifying the directed road graph according to the risk nesting structure and the hazard response mode as the road network topology, wherein the road network topology is embedded in the behavior generator and the hazard identification module.

[0028] Specifically, abstracting the target road segment area into a directed road graph refers to formally representing the road structure in real space using graph theory. By constructing a topological structure with directional attributes, mathematical modeling of traffic flow organization and spatial connectivity is achieved. Intersections and key sections serve as nodes to represent the locations of traffic flow convergence, divergence, and state transitions, while road segments serve as edges to represent the directed connectivity between nodes and the traffic flow propagation path. Through the association between nodes and edges, a computable road structure model is formed, providing a structured foundation for subsequent risk analysis and trajectory mapping.

[0029] Furthermore, according to the road geometry definition of the hazard response model, it refers to the categorized modeling of structural features that may induce accident risks based on geometric attributes such as road horizontal and vertical cross-sectional parameters, radius of curvature, slope changes, sight distance conditions, and pavement structural characteristics. Among them, sharp curve type hazards represent lateral stability risks caused by abrupt changes in radius of curvature or the superposition of continuous curves; sight distance type hazards represent perceptual risks caused by insufficient driver visibility due to obstruction, longitudinal slope superposition, or the influence of structures; and pavement type hazards represent handling risks caused by a decrease in pavement adhesion coefficient, structural damage, or material changes. Through the above classification, a patterned expression of the structural risk triggering mechanism is formed.

[0030] Subsequently, a risk nesting structure is defined, which refers to establishing a hierarchical structural model at the spatial scale and risk impact scope level. The risk impact scope is divided into regional level, corridor level, and location level. Among them, the regional level is used to describe the comprehensive risk background formed by macro traffic organization and overall operating environment, the corridor level is used to depict the risk transmission characteristics presented by specific traffic channels or continuous road segments, and the location level is used to represent the high-concentration risk sources generated by local key locations. Through the hierarchical nesting relationship, the risk is expressed progressively from macro to micro.

[0031] Furthermore, based on the risk nesting structure and hazard response mode, the directed road graph is labeled as a road network topology. This means that risk level attributes and hazard category attributes are added to the nodes and edges of the graph structure, so that the directed road graph not only retains spatial connection relationships, but also carries risk semantic information, forming a road network topology structure with dual expression capabilities of structural attributes and risk attributes. The road network topology is then embedded into the behavior generator and hazard identification module, so that traffic trajectory generation, deviation judgment, probability inversion and disturbance propagation analysis are all based on unified topological semantics as a constraint basis, realizing the coupling of structural semantics and algorithm model.

[0032] Based on the road network topology, a behavior generator and a road judgment module are deployed at the edge of the road segment. By generating normal traffic trajectories in the multi-dimensional road network data, and combining the real-time traffic trajectories, the system judges the trajectory of individual vehicles and traffic flow disturbance patterns, and generates hazard identification instructions.

[0033] Furthermore, this application also includes: constructing a behavior generator using an adversarial training method, wherein the behavior generator takes random noise, road geometric features, and environmental conditions as input and normal traffic trajectory as output; building a road judgment module based on deviation judgment of normal traffic trajectory, wherein the road judgment module includes a first judgment threshold based on single vehicle trajectory and a second judgment threshold based on stable disturbance aggregation under the convergence of traffic disturbance patterns in the same time and space; deploying the behavior generator and the road judgment module on the edge of the road segment and establishing a data interaction port with the vehicle network.

[0034] Furthermore, this application also includes: retrieving multi-dimensional road network data at the road segment edge, triggering the behavior generator, generating partitioned behavior based on the road network topology, and determining the partitioned traffic trajectory, wherein each partitioned traffic trajectory is identified by road segment coordinates; reading real-time traffic trajectories based on multi-dimensional road network data, performing single-vehicle trajectory deviation judgment and deviation judgment based on stable disturbance aggregation features with the partitioned traffic trajectories, and generating a hazard identification instruction, wherein the existence of at least one deviation is used as the instruction generation condition.

[0035] Specifically, constructing a behavior generator using an adversarial training approach involves introducing a game-theoretic optimization mechanism between the generator network and the discriminator network during the model training phase. Random noise is used as a latent variable input, and a joint feature vector is constructed by combining road geometric features and environmental conditions. The generator network outputs a distribution of normal traffic trajectory data that conforms to statistical laws. After training, only the generator network structure is retained as the behavior generator, which is used to generate corresponding normal traffic trajectories based on the input road structure parameters and environmental conditions during real-time operation. Random noise is used to simulate random fluctuations in traffic behavior, road geometric features are used to characterize curvature, slope, lane width, and linear structure attributes, environmental conditions are used to characterize external influencing factors such as weather, lighting, and traffic control status, and normal traffic trajectories are used to represent the expected motion path and speed evolution sequence of vehicles under accident-free conditions.

[0036] Furthermore, building a road judgment module based on deviation judgment of normal traffic trajectories refers to constructing a conventional network model or rule judgment structure independent of the adversarial training structure. This involves comparing and analyzing real-time collected traffic trajectory data with normal traffic trajectories output by the behavior generator, and identifying abnormal deviation behaviors by setting judgment thresholds. Specifically, the first judgment threshold based on a single vehicle trajectory is used to measure the deviation of a single vehicle from the normal trajectory in terms of position, speed, or acceleration. When the deviation exceeds a preset threshold, it is judged as an individual anomaly. The second judgment threshold based on stable disturbance aggregation under the convergence of traffic flow disturbance patterns in the same time and space is used to statistically aggregate the disturbance characteristics of multiple vehicles in the same time window and spatial grid. By analyzing the disturbance amplitude, directional consistency, and duration, a stable disturbance aggregation index is formed. When the aggregation result exceeds the group threshold, it is judged as an abnormal state at the traffic flow level.

[0037] Subsequently, the behavior generator and road judgment module are deployed at the edge of the road segment and a data interaction port is established with the vehicle network. This means that edge computing units with computing capabilities are deployed at the data collection nodes near the target road segment. The behavior generator is used to generate normal traffic trajectory benchmarks in real time, while the road judgment module is used to perform real-time deviation judgment. By establishing a two-way data interface with the vehicle network communication system, real-time transmission and feedback of vehicle position, speed, environmental status and control information are realized, thereby forming a low-latency hazard triggering mechanism and data closed loop.

[0038] The process involves retrieving multidimensional road network data from the edge of a road segment, triggering a behavior generator, and generating zoned behavior based on the road network topology to determine the zoned traffic trajectory. This means that edge computing nodes deployed near the target road segment read multidimensional road network data, including road structure parameters, historical traffic flow statistics, environmental status information, and real-time perception data. Using the constructed road network topology as a spatial constraint framework, the target road segment area is spatially partitioned according to nodes, edges, and risk levels. The partitioned units are then used as boundary conditions for behavior generation and input to the behavior generator to generate a sequence of normal traffic operation trajectories for the corresponding partitions. The multidimensional road network data provides multi-source heterogeneous feature input, the road network topology defines the generation range and spatial semantic relationships, the zoned behavior generation forms differentiated traffic behavior distribution models within different spatial units, and the zoned traffic trajectory represents the expected vehicle path, speed curve, and temporal evolution characteristics within the corresponding road segment coordinate range. Furthermore, the addition of road segment coordinates achieves a precise mapping between the trajectory and spatial location.

[0039] Furthermore, based on the real-time traffic trajectory read from the multi-dimensional road network data, deviation judgments are made for single-vehicle trajectories and deviation judgments based on stable disturbance aggregation features, generating hazard identification instructions. This means that real-time vehicle operation data is acquired in the same edge computing node, and the real-time traffic trajectory is mapped to the corresponding partition according to spatial coordinates, and compared and analyzed with the partition traffic trajectory. Among them, single-vehicle trajectory deviation judgment is used to calculate the degree of deviation of a single vehicle from the partition traffic trajectory in dimensions such as position offset, speed difference, acceleration change rate, or heading angle change. Stable disturbance aggregation features are used to statistically summarize the disturbance amplitude, directional consistency, and duration of multiple vehicles in the same spatiotemporal grid to form a group disturbance intensity index. When any one of the single-vehicle deviation or group disturbance aggregation index exceeds a preset threshold, the existence of at least one deviation is used as the condition for instruction generation, and a hazard identification instruction is output to trigger the subsequent hazard location and probability analysis process.

[0040] According to the hazard identification instruction, the hazard identification module is activated to locate the hazard road segment in the road network topology, and a hazard probability curve is generated by performing a combination inversion based on random variables and state transition probability analysis.

[0041] Furthermore, this application also includes: constructing a hazard identification module based on the road network topology, wherein the hazard identification module includes a first positioning layer, a second inversion layer and a third analysis layer; the hazard identification module receives the hazard identification instruction, locates the instruction road segment in the road network topology, and identifies multiple hazard road segments; for the multiple hazard road segments, it performs a combination inversion based on random variables and a state transition probability analysis to generate a hazard probability curve.

[0042] Furthermore, this application also includes: determining a combination of random variables through adversarial inversion based on the second inversion layer, wherein the combination of random variables is a combination of conditions that maximizes risk; determining a combination of real-time variables based on multi-dimensional road network data, generating a variable evolution sequence based on the linear transition from the real-time variable combination to the random variable combination; and performing Markov state transition probability analysis on the variable evolution sequence to determine the hazard probability curve.

[0043] Specifically, the hazard identification module is activated according to the hazard identification command, and the hazard road segment is located in the road network topology. The hazard identification module is constructed based on the road network topology. This means that after the hazard identification command is generated on the edge side, the analysis system coupled with the road network topology structure is started, and the hazard identification module is constructed using the road network topology, which includes node, edge and risk level identification information, as a spatial index and semantic constraint framework. The hazard identification module is internally divided into a first positioning layer, a second inversion layer and a third analysis layer. The first positioning layer is used to complete spatial location matching and candidate road segment screening according to the command information. The second inversion layer is used to construct a variable relationship model and perform risk condition inversion calculation. The third analysis layer is used to perform probability calculation and trend evaluation on the variable evolution results. The hierarchical structure realizes a progressive processing flow from spatial positioning to risk quantification.

[0044] Then, the hazard identification module receives the hazard identification instruction, locates the instruction road segment in the road network topology, and identifies multiple hazard road segments. This means that the first positioning layer reads the road segment coordinates, deviation type, and disturbance intensity contained in the hazard identification instruction, performs node matching and edge mapping operations in the road network topology, maps the spatial location corresponding to the instruction to specific road segment units, and extends it to adjacent or related road segments at the same level in combination with the risk nesting hierarchy relationship, thereby forming a set of multiple candidate hazard road segments for subsequent risk inversion and probability analysis.

[0045] Subsequently, for multiple hazardous road sections, based on the second inversion layer, adversarial inversion is used to determine the combination of random variables. This refers to constructing a set of random variables in the second inversion layer of the hazard identification module, including uncertain factors such as traffic flow density, speed fluctuation amplitude, headway, road adhesion coefficient, meteorological influence factors, and structural geometric parameters. An adversarial optimization mechanism is introduced, and an objective function is constructed to make the risk indicators tend to extreme values. Iterative search and parameter back-calculation are performed in the variable space to obtain the combination of variable values ​​that maximizes the risk level under constraints. Among them, the combination of random variables is used to represent the set of multidimensional influencing factors with uncertainty in a statistical sense, and the combination of risk maximization conditions is used to characterize the parameter configuration structure under the most unfavorable state of potential accident risk.

[0046] Furthermore, based on multidimensional road network data, real-time variable combinations are determined. A variable evolution sequence is generated based on the linear migration from real-time variable combinations to random variable combinations. This involves extracting traffic flow state parameters, environmental state parameters, and road structure parameters from multidimensional road network data at the current moment to form real-time variable combinations. A mapping path from these real-time variable combinations to risk-maximizing condition combinations is constructed in the variable space. The changing trends of each variable at continuous time steps are calculated using a linear migration model, resulting in a variable evolution sequence that gradually approaches the risk extreme state over time. Here, linear migration describes the process of proportional adjustment of variables within the constraint space, and the variable evolution sequence represents the continuous change trajectory of multidimensional variables in the time dimension.

[0047] Subsequently, Markov state transition probability analysis is performed on the variable evolution sequence to determine the hazard probability curve. This involves discretizing the variable evolution sequence into several risk state levels, constructing a state-space model, and calculating the state transition probability matrix between adjacent time steps based on the Markov assumption. Through recursive calculation, the probability distribution of different risk states at each future time node is obtained, and the probability values ​​corresponding to high-risk states are arranged in chronological order to form the hazard probability curve. Among these, Markov state transition probability analysis is used to characterize the random transition characteristics of risk states over time, and the hazard probability curve is used to intuitively express the evolution trend of the probability of risk occurrence on the time axis.

[0048] Based on the aforementioned hazard probability curve, hazard disturbance propagation analysis is performed in the road network topology to determine hazard identification information, which is then distributed to the vehicle platform port for traffic hazard early warning management.

[0049] Furthermore, this application also includes: performing mechanical modeling of hazard disturbance propagation based on the road network topology to determine the traffic wave space, wherein the traffic wave space is jointly defined based on the disturbance propagation of vehicle interaction, density, and flow; locating hazard sources based on the hazard probability curve; and performing propagation analysis on the hazard sources in the traffic wave space to determine the road network propagation heatmap.

[0050] Furthermore, this application also includes: identifying the first hazard identification result based on the hazard probability curve, wherein the hazard point coordinates, confidence level, hazard type and real-time disturbance intensity are the identification elements; generating hazard identification information based on the first hazard identification result and the road network propagation heat map, and displaying the warning on the vehicle platform port.

[0051] Furthermore, this application also includes: receiving the hazard identification information at the vehicle platform port, establishing a vehicle coordinate system by retrieving real-time vehicle coordinates as the origin; generating a risk control warning instruction based on distance constraints for the hazard identification information based on the vehicle coordinate system, and executing traffic hazard warning management based on time series.

[0052] Specifically, based on the road network topology, a dynamic model of the propagation of hazard disturbances is established to determine the traffic wave space. This involves introducing traffic flow continuity equations and dynamic relationships into the road network topology, which includes node connection relationships and risk level identification information. The physical mechanisms of disturbances in traffic flow caused by hazard triggering, such as sudden speed changes, density variations, and spacing compression, are modeled to construct a mathematical space structure describing the propagation path and intensity of disturbances in the road network. The traffic wave space represents the dynamic propagation field of disturbances spreading along or against the traffic flow direction under the constraints of vehicle interaction mechanisms. Vehicle interaction is used to characterize the coupling relationship between following behavior, lane-changing behavior, and braking response. The propagation of density and flow disturbances describes the transmission process of traffic flow parameter changes in time and space. For example, disturbances can spread downstream along the traffic flow direction and naturally decay under the condition of no continuous excitation, or form a continuous amplification effect under the influence of continuous risk factors such as road defect source terms. Through the above joint definition, a traffic wave space model with propagation direction, propagation speed, and decay characteristics is formed.

[0053] Furthermore, locating the source of a hazard based on the probability curve of the hazard refers to analyzing the peak position and growth trend of the probability curve of the hazard in the time dimension, and combining the spatial mapping relationship to map the time nodes of significant probability increases to the coordinates of specific road segments in the road network topology. By tracing back the evolution path of variables, the initial trigger position of the disturbance is determined, thereby identifying the spatial unit where the risk first appears or the risk intensity is highest, forming the hazard source location result, which is used to represent the starting point or continuous excitation point of disturbance propagation.

[0054] Subsequently, the propagation analysis of potential hazards in the traffic wave space is performed to determine the road network propagation heat map. This involves using potential hazards as initial conditions, calculating the propagation path, intensity, and time delay of disturbances between different road segment units within the constructed traffic wave space model, solving the disturbance propagation equation to obtain the risk superposition value at different spatial nodes, and mapping the calculation results to the road network topology to form a visual distribution map. The road network propagation heat map is used to represent the degree of disturbance impact on each road segment using color gradients or intensity levels, realizing a spatial expression of the scope and intensity of the hazard's impact.

[0055] The identification result of the first hidden danger based on the hidden danger probability curve refers to the threshold judgment and peak analysis of the risk probability value corresponding to each time node in the hidden danger probability curve, mapping the time segment where the probability exceeds the preset judgment standard to the specific spatial location in the road network topology to form the coordinates of the hidden danger point, and determining the confidence level by combining the numerical stability index and confidence interval calculation results output by the probability calculation model, and determining the hidden danger type based on the aforementioned hidden danger response mode classification results, and determining the real-time disturbance intensity by combining the local disturbance amplitude and disturbance growth rate obtained by traffic wave spatial calculation, thus constituting a structured first hidden danger identification result including spatial location, risk confidence level, risk category and disturbance energy level; where the hidden danger point coordinates are used to represent the precise spatial location of the risk occurrence, the confidence level is used to represent the reliability of the identification result, the hidden danger type is used to distinguish the risk category under different structures or operating mechanisms, and the real-time disturbance intensity is used to quantify the degree of impact on the current traffic flow.

[0056] Furthermore, based on the first hazard identification result and the road network propagation heat map, hazard identification information is generated and displayed as a warning on the vehicle platform port. This means that the first hazard identification result, which includes the coordinates of the hazard point, confidence level, hazard type, and real-time disturbance intensity, is fused with the risk diffusion range and intensity distribution of the corresponding road segment in the road network propagation heat map to form a comprehensive data structure containing predicted information on risk level, impact range, propagation direction, and duration. This data structure is then sent to the vehicle platform port through a communication interface and presented as a warning in the vehicle display system in a graphical or symbolic manner, thereby providing risk warnings and route decision assistance for driving behavior.

[0057] The vehicle platform receives hazard identification information and establishes a vehicle coordinate system by retrieving real-time vehicle coordinates as the origin. This means that the vehicle terminal system obtains hazard identification information, including hazard point coordinates, risk level, propagation range, and disturbance intensity, through the communication interface. At the same time, it calls the real-time vehicle coordinates output by the vehicle positioning module, sets the real-time vehicle coordinates as the coordinate origin, and establishes a local spatial reference frame by combining the vehicle's heading angle and driving direction, forming a vehicle coordinate system centered on the vehicle's current position. The vehicle coordinate system is used to convert the risk position in the external road network space into a relative position expression relative to the vehicle's motion state, thereby realizing the dynamic correlation between spatial risk information and vehicle operating status.

[0058] Furthermore, based on the vehicle coordinate system, distance-constrained risk control warning commands are generated from the hazard identification information, and time-series-based traffic hazard warning management is implemented. This involves calculating the longitudinal distance, lateral offset, and predicted arrival time of the hazard point relative to the origin in the vehicle coordinate system, comparing the calculation results with preset distance and time thresholds, and generating different levels of risk control warning commands according to distance constraint rules. Among these, the risk control warning commands are used to define the prompting method, prompting intensity, and intervention level, while the time-series warning management is used to perform rolling predictions and dynamic updates of the risk proximity within several future time steps according to the vehicle's movement speed and risk propagation trend, thereby realizing a continuous and phased traffic hazard warning control process.

[0059] In summary, the method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion provided in this application has the following technical effects: by realizing the ability to dynamically identify and predict hazards based on real-time traffic flow data and road network topology information, it achieves the technical effects of being able to monitor and reflect the disturbance propagation process in traffic flow in real time, accurately identify the spatiotemporal distribution of hazards, and provide early warnings.

[0060] Example 2: Based on the same inventive concept as the method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion in the foregoing examples, this application also provides a platform for dynamic identification of accident-prone road sections based on multi-dimensional data fusion. Please refer to the appendix. Figure 2 The system includes: a road network topology determination module 1, used to perform directed graph transformation on the target road segment area, and determine the road network topology by calibrating the risk nesting structure and hazard response mode; a hazard identification instruction generation module 2, used to deploy a behavior generator and a road judgment module on the edge side of the road segment based on the road network topology, generate hazard identification instructions by performing normal-condition partitioned traffic trajectory generation on the retrieved multi-dimensional road network data, and judge single-vehicle trajectory and traffic flow disturbance mode by combining real-time traffic trajectory; a hazard probability curve generation module 3, used to activate the hazard identification module according to the hazard identification instruction, locate the hazard road segment in the road network topology, and generate a hazard probability curve by performing combination inversion based on random variables and state transition probability analysis; and a hazard identification information determination module 4, used to perform hazard disturbance propagation analysis in the road network topology based on the hazard probability curve, determine hazard identification information, and send it to the vehicle platform port for traffic hazard early warning management.

[0061] Furthermore, the dynamic identification platform for accident hazard road sections based on multi-dimensional data fusion is also used for: constructing a behavior generator using an adversarial training method, wherein the behavior generator takes random noise, road geometric features, and environmental conditions as input and normal traffic trajectories as output; building a road judgment module based on deviation judgment of normal traffic trajectories, wherein the road judgment module includes a first judgment threshold based on single vehicle trajectory and a second judgment threshold based on stable disturbance aggregation under the convergence of traffic disturbance patterns in the same time and space; deploying the behavior generator and the road judgment module on the edge of the road section and establishing a data interaction port with the vehicle network.

[0062] Furthermore, the dynamic identification platform for accident hazard road sections based on multi-dimensional data fusion is also used for: retrieving multi-dimensional road network data at the edge of the road section, triggering the behavior generator, generating partitioned behavior based on the road network topology, and determining the partitioned traffic trajectory, wherein each partitioned traffic trajectory is identified with road section coordinates; reading real-time traffic trajectory based on multi-dimensional road network data, performing single-vehicle trajectory deviation judgment and deviation judgment based on stable disturbance aggregation features with the partitioned traffic trajectory, and generating a hazard identification instruction, wherein the existence of at least one deviation is used as the condition for instruction generation.

[0063] Furthermore, the dynamic identification platform for accident hazard road sections based on multi-dimensional data fusion is also used for: constructing a hazard identification module based on the road network topology, wherein the hazard identification module includes a first positioning layer, a second inversion layer, and a third analysis layer; the hazard identification module receives the hazard identification instruction, locates the instruction road section in the road network topology, and identifies multiple hazard road sections; for the multiple hazard road sections, it performs a combination inversion based on random variables and a state transition probability analysis to generate a hazard probability curve.

[0064] Furthermore, the dynamic identification platform for accident hazard road sections based on multidimensional data fusion is also used for: determining a combination of random variables through adversarial inversion based on the second inversion layer, wherein the combination of random variables is a combination of conditions that maximizes risk; determining a real-time combination of variables based on multidimensional road network data; generating a variable evolution sequence based on the linear transition from the real-time variable combination to the random variable combination; and performing Markov state transition probability analysis on the variable evolution sequence to determine the hazard probability curve.

[0065] Furthermore, the dynamic identification platform for accident hazard road sections based on multi-dimensional data fusion is also used to: abstract the target road section area into a directed road graph, wherein intersections and key sections are nodes and road sections are edges; define hazard response modes according to road geometry, wherein the hazard response modes include sharp bend hazards, sight distance hazards, and road surface hazards; define a risk nesting structure, wherein the risk nesting structure includes regional level, corridor level, and point level; and identify the directed road graph according to the risk nesting structure and the hazard response modes as the road network topology, wherein the road network topology is embedded in the behavior generator and the hazard identification module.

[0066] Furthermore, the dynamic identification platform for accident hazard road sections based on multi-dimensional data fusion is also used for: performing mechanical modeling of hazard disturbance propagation according to the road network topology, determining the traffic wave space, wherein the traffic wave space is jointly defined based on the disturbance propagation of vehicle interaction, density, and flow; locating the hazard source according to the hazard probability curve; and performing propagation analysis on the hazard source in the traffic wave space to determine the road network propagation heat map.

[0067] Furthermore, the dynamic identification platform for accident hazard road sections based on multi-dimensional data fusion is also used to: identify the first hazard identification result according to the hazard probability curve, wherein the hazard point coordinates, confidence level, hazard type and real-time disturbance intensity are the identification elements; generate hazard identification information according to the first hazard identification result and the road network propagation heat map, and display the warning on the vehicle platform port.

[0068] Furthermore, the dynamic identification platform for accident hazard road sections based on multi-dimensional data fusion is also used for: receiving the hazard identification information at the vehicle platform port, establishing a vehicle coordinate system by retrieving real-time vehicle coordinates as the origin; generating risk control and early warning instructions based on distance constraints for the hazard identification information based on the vehicle coordinate system, and executing traffic hazard early warning management based on time series.

[0069] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The method and specific examples of dynamic identification of accident hazard road sections based on multi-dimensional data fusion in the aforementioned embodiment 1 are also applicable to the dynamic identification platform of accident hazard road sections based on multi-dimensional data fusion in this embodiment. Through the foregoing detailed description of the method of dynamic identification of accident hazard road sections based on multi-dimensional data fusion, those skilled in the art can clearly understand the dynamic identification platform of accident hazard road sections based on multi-dimensional data fusion in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0070] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0071] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion, characterized in that, The method includes: A directed graph transformation is performed on the target road segment area, and the road network topology is determined by calibrating the risk nesting structure and hazard response mode; Based on the road network topology, a behavior generator and a road judgment module are deployed at the edge of the road segment. By generating normal traffic trajectories in the partitions of the retrieved multi-dimensional road network data, and combining the real-time traffic trajectories to judge the trajectory of a single vehicle and the traffic flow disturbance pattern, a hazard identification instruction is generated. According to the hazard identification instruction, the hazard identification module is activated to locate the hazard road segment in the road network topology and generate a hazard probability curve by performing a combination inversion based on random variables and state transition probability analysis. Based on the aforementioned hazard probability curve, hazard disturbance propagation analysis is performed in the road network topology to determine hazard identification information, which is then distributed to the vehicle platform port for traffic hazard early warning management. Based on the road network topology, a behavior generator and a road determination module are deployed at the edge of the road segment, including: An adversarial training method is used to construct a behavior generator, wherein the behavior generator takes random noise, road geometric features and environmental conditions as input and normal traffic trajectory as output. A road judgment module is built based on deviation judgment of normal traffic trajectory. The road judgment module includes a first judgment threshold based on single vehicle trajectory and a second judgment threshold based on stable disturbance aggregation under the convergence of traffic disturbance patterns in the same time and space. The behavior generator and road judgment module are deployed at the edge of the road segment to establish a data interaction port with the vehicle network. Generate hazard identification instructions, including: Multi-dimensional road network data is retrieved from the edge of the road segment, triggering the behavior generator to generate partitioned behaviors based on the road network topology and determine the traffic trajectory of each partition. Each partition traffic trajectory is identified by the road segment coordinates. Real-time traffic trajectories are read from multi-dimensional road network data, and deviation judgments based on single-vehicle trajectory and stable disturbance aggregation features are made between the traffic trajectories of the partition and the deviation judgments. Hazard identification instructions are generated, wherein the existence of at least one deviation is used as the condition for instruction generation. Determine the road network topology, including: The target road segment area is abstracted as a directed road graph, where intersections and key sections are nodes and road segments are edges; According to the road geometry definition of the hazard response mode, the hazard response mode includes sharp curve type hazard, sight distance type hazard and road surface type hazard; Define a risk nesting structure, wherein the risk nesting structure includes a regional level, a corridor level, and a point level; Based on the risk nesting structure and the hazard response mode, the directed road graph is identified as the road network topology, wherein the road network topology is embedded in the behavior generator and the hazard identification module.

2. The method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion as described in claim 1, characterized in that, Perform combination inversion and state transition probability analysis based on random variables to generate hazard probability curves, including: Based on the road network topology, a hazard identification module is constructed, which includes a first positioning layer, a second inversion layer and a third analysis layer. The hazard identification module receives the hazard identification instruction, locates the instruction road segment in the road network topology, and identifies multiple hazard road segments. For the multiple hazardous road sections, a combination inversion based on random variables and a state transition probability analysis are performed to generate hazardous probability curves.

3. The method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion as described in claim 2, characterized in that, Performing combination inversion and state transition probability analysis based on random variables includes: Based on the second inversion layer, a combination of random variables is determined through adversarial inversion, wherein the combination of random variables is a combination of conditions that maximizes risk; Based on multidimensional road network data, real-time variable combinations are determined, and variable evolution sequences are generated based on the linear transition from real-time variable combinations to random variable combinations. Markov state transition probability analysis was performed on the variable evolution sequence to determine the hazard probability curve.

4. The method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion as described in claim 1, characterized in that, The analysis of hazard disturbance propagation in the road network topology includes: Based on the road network topology, a mechanical model of the propagation of potential hazards is performed to determine the traffic wave space, wherein the traffic wave space is jointly defined based on the disturbance propagation of vehicle interactions, density, and flow. Based on the aforementioned hazard probability curve, locate the source of the hazard; In the traffic wave space, the propagation analysis of the potential hazard sources is performed to determine the road network propagation heat map.

5. The method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion as described in claim 4, characterized in that, Determine hazard identification information, including: Based on the aforementioned hazard probability curve, the first hazard identification result is marked, wherein the hazard point coordinates, confidence level, hazard type and real-time disturbance intensity are the marking elements. Based on the first hazard identification result and the road network propagation heat map, hazard identification information is generated and displayed as a warning on the vehicle platform port.

6. The method for dynamic identification of accident-prone road sections based on multi-dimensional data fusion as described in claim 1, characterized in that, Traffic hazard early warning management is delegated to the vehicle platform port, including: The vehicle platform port receives the hazard identification information and establishes a vehicle coordinate system by retrieving real-time vehicle coordinates as the origin. Based on the vehicle coordinate system, a risk control and early warning instruction based on distance constraints is generated from the hazard identification information, and a traffic hazard early warning management based on time series is executed.

7. A dynamic identification platform for accident-prone road sections based on multi-dimensional data fusion, characterized in that, The steps for implementing the dynamic identification method for accident-prone road sections based on multi-dimensional data fusion as described in any one of claims 1 to 6 include: The road network topology determination module is used to perform directed graph transformation on the target road segment area and determine the road network topology by calibrating the risk nesting structure and the hidden danger response mode. The hazard identification instruction generation module is used to deploy a behavior generator and a road judgment module on the edge side of the road segment based on the road network topology. It generates a hazard identification instruction by generating normal traffic trajectories in the partitions of the retrieved multi-dimensional road network data and judging the single vehicle trajectory and traffic flow disturbance pattern in combination with the real-time traffic trajectory. The hazard probability curve generation module is used to activate the hazard identification module according to the hazard identification instruction, locate the hazard road segment in the road network topology, and generate the hazard probability curve by performing a combination inversion based on random variables and state transition probability analysis. The hazard identification information determination module is used to combine the hazard probability curve to perform hazard disturbance propagation analysis in the road network topology, determine hazard identification information, and distribute it to the vehicle platform port for traffic hazard early warning management. Furthermore, the dynamic identification platform for accident-prone road sections based on multi-dimensional data fusion is also used for: An adversarial training method is used to construct a behavior generator, wherein the behavior generator takes random noise, road geometric features and environmental conditions as input and normal traffic trajectory as output. A road judgment module is built based on deviation judgment of normal traffic trajectory. The road judgment module includes a first judgment threshold based on single vehicle trajectory and a second judgment threshold based on stable disturbance aggregation under the convergence of traffic disturbance patterns in the same time and space. The behavior generator and road judgment module are deployed at the edge of the road segment to establish a data interaction port with the vehicle network. Multi-dimensional road network data is retrieved from the edge of the road segment, triggering the behavior generator to generate partitioned behaviors based on the road network topology and determine the traffic trajectory of each partition. Each partition traffic trajectory is identified by the road segment coordinates. Real-time traffic trajectories are read from multi-dimensional road network data, and deviation judgments based on single-vehicle trajectory and stable disturbance aggregation features are made between the traffic trajectories of the partition and the deviation judgments. Hazard identification instructions are generated, wherein the existence of at least one deviation is used as the condition for instruction generation. The dynamic identification platform for accident-prone road sections based on multi-dimensional data fusion is also used for: The target road segment area is abstracted as a directed road graph, where intersections and key sections are nodes and road segments are edges; According to the road geometry definition of the hazard response mode, the hazard response mode includes sharp curve type hazard, sight distance type hazard and road surface type hazard; Define a risk nesting structure, wherein the risk nesting structure includes a regional level, a corridor level, and a point level; Based on the risk nesting structure and the hazard response mode, the directed road graph is identified as the road network topology, wherein the road network topology is embedded in the behavior generator and the hazard identification module.

Images