Traffic transportation operation monitoring early warning and decision analysis method based on vehicle infrastructure cooperation

Abstract

The invention discloses a traffic transportation operation monitoring early warning and decision analysis method based on vehicle infrastructure cooperation, and relates to the technical field of intelligent traffic. According to the method, real-time collection of dynamic traffic elements is realized through a multi-dimensional sensing network of a vehicle end, a road side and an environment and V2X communication, time-space reference unification of multi-source data is ensured, a road-vehicle-environment-event semantic network is constructed, multi-dimensional recognition of abnormal events such as accidents, congestion and severe weather is realized in combination with hierarchical feature extraction, and the method has the advantages of being high in practicability and high in practicability. The method is advantaged in that identification accuracy is improved, abnormal event propagation paths can be predicted, global road network situation prediction capability is realized, differential early warning is generated based on a comprehensive risk index, multi-level responses such as traffic signal adjustment and path planning are triggered, response time is greatly shortened, emergency response efficiency is optimized, a decision execution effect real-time feedback mechanism is established, and the method is suitable for popularization and application. The system performance is continuously optimized along with data accumulation, and the defect that a big data platform lacks an intelligent decision closed loop is avoided.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a transportation operation monitoring, early warning and decision-making analysis method based on vehicle-road collaboration. Background Art

[0002] With the rapid development of intelligent transportation systems, traditional transportation operation monitoring and decision-making technologies face many challenges. For example, the Chinese patent with announcement number CN117727183B discloses an autonomous driving safety warning method and system combined with vehicle-road collaboration. The method includes: reading the target real-time driving data, then reading the driving association state data set, obtaining the target driving visual model, and then performing static vehicle-road collaborative safety detection to obtain the target static collaborative safety detection results, and then performing dynamic vehicle-road collaborative safety detection to obtain dynamic collaborative safety detection results, generating a safety analysis report and sending it to the safety warning device for safety warning. It solves the problem that the existing method relies too much on historical data, which increases the difficulty of real-time warning, may lead to a decrease in the accuracy of the warning, and cannot be adaptively adjusted according to changes in the scene, thereby limiting the real-time nature of the warning effect. According to the dynamic collaborative safety detection results, it is judged in real time whether the vehicle has safety risks or violations, and a warning is issued, which improves the real-time and accuracy of the data.

[0003] Although the above patents improve data real-time performance and warning accuracy, they still have the following problems: 1. The existing transportation system is limited to vehicle data and the surrounding local environment data. It lacks comprehensive perception of road infrastructure, macro-road network status and environmental parameters, making it difficult to build a complete picture of traffic operation status.

[0004] 2. Risk assessment only focuses on the safety risks or violations of individual vehicles, and does not conduct multi-dimensional identification of abnormal events such as accidents, congestion, and bad weather. It is even more unable to predict the propagation path and comprehensive risks of events in the road network.

[0005] 3. Without the semantic representation and dynamic modeling of traffic status, it is impossible to quantify the impact of environmental factors on the traffic system, making it difficult to support refined traffic management and decision-making. Summary of the Invention

[0006] The purpose of the present invention is to provide a transportation operation monitoring, early warning and decision-making analysis method based on vehicle-road collaboration, using four-dimensional semantic network and hierarchical feature extraction technology, combined with graph neural network and time series data mining, to accurately identify abnormal events and predict the propagation path, build a comprehensive risk assessment model, improve the intelligent analysis capability of complex traffic scenarios, and realize efficient linkage of early warning, decision-making and execution to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions: Transportation operation monitoring, early warning, and decision-making analysis methods based on vehicle-road collaboration include: Multi-dimensional perception network construction: Build a multi-dimensional perception network to obtain real-time vehicle status, road environment and infrastructure data, and integrate the collected vehicle status, road environment and infrastructure data to generate multi-dimensional spatial data; Dynamic traffic status analysis: Perform hierarchical feature extraction on the fused multi-dimensional spatial data to construct a dynamic traffic status model. Observe in real time the degree to which traffic flow parameters deviate from normal distribution. Combine historical observation data with real-time observations to perform multi-dimensional identification of abnormal events. Comprehensive risk prediction and assessment: Based on the identification results, the identified abnormal events are analyzed for correlation, the propagation path is predicted, and the comprehensive risk index of the abnormal events is calculated according to the degree of harm and propagation risk of the abnormal events; Warning generation and decision-making: Based on the comprehensive risk index, a multi-level response mechanism is triggered, and corresponding warning information is generated. A decision plan is established and the safety of the decision plan is verified. The decision instructions are pushed to vehicles and various infrastructure through a two-way communication link.

[0008] Furthermore, the construction of a multi-dimensional perception network also includes: Building bidirectional communication links between vehicles and between vehicles and infrastructure based on the V2X communication components installed in the vehicles, and establishing a connection with the traffic cloud platform based on the V2X communication components; The vehicle sends vehicle status data packets to adjacent vehicles based on a two-way communication link at a preset period, and each roadside device sends event information in real time, forming a dynamic communication network between vehicles and between vehicles and various infrastructures; Determine time synchronization between vehicles and between vehicles and infrastructure based on clock synchronization mechanism; The vehicle obtains monitoring information of adjacent vehicles based on a two-way communication link, including the real-time position, movement status and intention information of adjacent vehicles, and at the same time collects traffic signal status and road environment data actively uploaded by roadside infrastructure.

[0009] Furthermore, dynamic traffic status analysis also includes the construction of a four-dimensional semantic network topology: Taking the physical entities of roads as the topological basic nodes, the structural skeleton of the road network topology map is constructed based on topological attributes and historical traffic characteristics; The real-time trajectory of the vehicle is mapped into a dynamic edge, and the vehicle position transfer within adjacent time slices is used as the edge connection relationship. Based on the edge attribute behavior characteristics, an edge set is constructed; Environmental parameters are used as weight factors in the graph structure, mapped to corresponding road nodes and vehicle behavior edges according to the spatiotemporal grid, and the identified abnormal events are embedded in the four-dimensional semantic network topology structure; At the same time, the association relationship between nodes in the four-dimensional semantic network topology structure is learned to construct a semantic representation of the traffic status.

[0010] Furthermore, the dynamic traffic state analysis and hierarchical feature extraction further includes: The road network topology is spatially discretized. Based on the physical coordinate range of the road and the lane division, the road network is divided into spatiotemporal grid units. The basic characteristics of traffic flow in each spatiotemporal grid unit are extracted to form a spatial grid traffic state vector. Extract edge attributes formed by vehicle position transfer within adjacent time slices to generate a feature sequence that represents the temporal changes in vehicle movement. At the same time, combine historical traffic data to identify the periodic patterns and sudden change characteristics of traffic flow and generate a temporal dynamic feature vector. The spatial grid traffic state vector and the temporal dynamic feature vector are weightedly fused to form a composite feature representation, and then embedded into the four-dimensional semantic network topology structure.

[0011] Furthermore, a correlation analysis is conducted based on the comprehensive risk prediction assessment, specifically including: Acquire historical abnormal event data, combine it with the multidimensional spatial data and environmental parameters in the four-dimensional semantic network topology structure, extract the propagation patterns of different types of abnormal events under different environmental parameters, and build a propagation rule knowledge base; Taking the identified abnormal events as key nodes, based on the association relationship and edge attributes between nodes in the four-dimensional semantic network topology structure, and based on the propagation rule knowledge base, the potentially affected road nodes and adjacent event nodes are screened out to construct an event evolution tree; The attributes of tree nodes include event type, spatiotemporal location, current impact level, and severity level; the attributes of tree node connection relationships include event propagation direction, propagation probability, and time delay threshold.

[0012] Furthermore, the propagation path is predicted, including: Taking the key node of the abnormal event as the target node, multi-path sampling is performed on the connection relationship attributes in the event evolution tree to generate at least one propagation path. During each sampling, the corresponding branch path is selected according to the propagation probability distribution of the current target node; Combined with the propagation time distribution of similar historical events, the time series of each target node affected in each propagation path is obtained to generate a propagation path set; Map the propagation probability distribution information of each propagation path in the propagation path set to the spatiotemporal grid unit, and cumulatively calculate the probability of each grid unit being affected in the future period to generate a risk probability cloud map for the future period; Identify the key hub nodes in the propagation path set, calculate the risk contribution of the key hub nodes in the propagation process, and prioritize each key hub node based on the risk contribution.

[0013] Furthermore, the key hub nodes in the propagation path set are identified, including: Extract node attributes and topological structure information of each road node in the propagation path set in the four-dimensional semantic network topological structure; Based on the node attributes and topological structure information, the basic flow characteristics of each road node and the flow transmission sensitivity between road nodes are obtained to determine the flow weight coefficient and flow transmission weight coefficient; According to the preset law enforcement management method information corresponding to each road node, the law enforcement management weight of the road node is determined based on the management level corresponding to the law enforcement management method; The law enforcement management weight, flow weight coefficient and flow conduction weight coefficient are comprehensively calculated to obtain the road topology weight value, which is compared with the preset weight threshold to screen out road nodes with a value higher than the preset weight threshold as key hub nodes.

[0014] Furthermore, a multi-level response mechanism includes: Establish mapping rules between warning levels and response strategies. For example, if the warning level is low risk, text warnings will be pushed to surrounding vehicles, and cautious driving prompts will be given through the in-vehicle navigation. If the warning level is medium risk, the timing optimization of roadside traffic lights will be triggered simultaneously, and route suggestions will be sent to the logistics fleet. If the warning level is high risk, cross-regional emergency linkage will be initiated, and event details and disposal plans will be pushed to the traffic management center. At the same time, speed limit instructions will be enforced through a two-way communication link.

[0015] Furthermore, before fusing the collected vehicle status, road environment and infrastructure data, the method further includes: performing noise reduction on image data included in the road environment data; The denoising of the image data in the road environment data includes: Arbitrarily obtain an image from the road environment data as the image to be denoised; Performing grayscale processing on the image to be denoised to obtain a grayscale image; Randomly select a pixel point in the grayscale image as the first pixel point; determine the target area with the first pixel point as the center and a preset distance as the radius; Calculating the difference between the first pixel and the other pixels in the target area except the first pixel to obtain a plurality of difference values; comparing the plurality of difference values ​​with a preset difference threshold, and marking the pixels whose difference values ​​are greater than or equal to the preset difference threshold as abnormal; Traverse all pixels in the grayscale image and count the number of abnormal marks for each pixel; Comparing the number of abnormal marks with a preset abnormal mark threshold, taking pixels whose number of abnormal marks is greater than or equal to the preset abnormal mark threshold as first abnormal pixel points, to obtain a plurality of first abnormal pixel points; Perform edge recognition on the grayscale image to determine the edge pixels in the grayscale image; Subtracting a plurality of first abnormal pixel points from the edge pixel points to obtain a plurality of second abnormal pixel points; Calculating the discreteness of each second abnormal pixel point respectively, and comparing the discreteness with a preset discreteness threshold, taking the pixel point when the discreteness is greater than or equal to the preset discreteness threshold as a third abnormal pixel point, to obtain a plurality of third abnormal pixel points; Deleting several third abnormal pixels to obtain a denoised grayscale image; Traverse all images in the road environment data to obtain the denoised road environment data.

[0016] Furthermore, based on the degree of harm and the risk of spread of the abnormal event, the comprehensive risk index of the abnormal event is calculated, including steps 1-2: Step 1: Obtain the average risk value of vehicle a in historical driving data and the propagation risk value of abnormal events on the driving path of vehicle a. Based on the average risk value, the propagation risk value, and the expected risk value of the abnormal event, determine the comprehensive risk index of the abnormal event for vehicle a. ; in, represents the comprehensive risk index of the abnormal event to vehicle a; n represents the total number of risk factors in the abnormal event; Represents the average risk value of vehicle a in historical driving data; represents the risk value of the abnormal event spreading to the driving path of the vehicle a; Represents the risk level value of the i-th risk factor type among n risk factors; The probability density function of the risk of occurrence of the i-th risk factor type; It represents the cumulative probability of the occurrence of risk of the i-th risk factor in the next unit time interval starting from time t; It represents the total cumulative probability of the occurrence of risk of the i-th risk factor in all future time starting from time t; Step 2: Determine the comprehensive risk index of the abnormal event based on the comprehensive risk index of the abnormal event to vehicle a and the total number of vehicles in the driving path of vehicle a; ; in, Indicates the comprehensive risk index of abnormal events; Indicates the total number of vehicles in the driving path of vehicle a.

[0017] Compared with the prior art, the present invention has the following beneficial effects: Through the multi-dimensional perception network of the vehicle side, road side and environment and V2X communication, real-time collection of dynamic traffic elements is achieved, the spatiotemporal benchmark of multi-source data is ensured, and a "road-vehicle-environment-event" semantic network is constructed. Combined with hierarchical feature extraction, multi-dimensional identification of abnormal events such as accidents, congestion, and bad weather is achieved, the recognition accuracy is improved, the propagation path of abnormal events can be predicted, and the global road network situation prediction capability is possessed. Differentiated warnings are generated based on comprehensive risk indexes, and multi-level responses such as traffic signal adjustment and route planning are triggered, which greatly shortens the response time, optimizes the efficiency of emergency response, and establishes a real-time feedback mechanism for decision execution effects. System performance is continuously optimized with data accumulation, avoiding the defect of the big data platform lacking an intelligent decision-making closed loop. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the transportation operation monitoring, early warning and decision analysis method of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] To address the problems of existing technologies being limited to vehicles and their surroundings, lacking multi-dimensional identification and propagation prediction in risk assessment, lacking differentiated response and cross-regional coordination in decision-making mechanisms, and lacking semantic representation of traffic status, making it difficult to build a complete situational picture and support refined management, please refer to Figure 1 , this embodiment provides the following technical solutions: Transportation operation monitoring, early warning, and decision-making analysis methods based on vehicle-road collaboration include: Multi-dimensional perception network construction: Build a multi-dimensional perception network covering the vehicle side, road side, and environment to obtain real-time vehicle status, road environment, and infrastructure data. This data is then integrated to generate multi-dimensional spatial data with timestamps, enabling real-time collection of dynamic traffic elements and the unification of spatiotemporal benchmarks. Dynamic traffic status analysis: Perform hierarchical feature extraction on the fused multidimensional spatial data to construct a dynamic traffic status model. This allows real-time observation of the degree to which traffic flow parameters deviate from normal distributions. Combining historical observation data with real-time observations, this allows for multi-dimensional identification of abnormal events such as accidents, congestion, and severe weather. Comprehensive risk prediction and assessment: Based on the identification results, the identified abnormal events are analyzed for correlation, the propagation path is predicted, and the comprehensive risk index of the abnormal events is calculated according to the degree of harm and propagation risk of the abnormal events; Warning generation and decision-making: A multi-level response mechanism is triggered based on a comprehensive risk index, generating warning information that includes event type, impact scope, and disposal recommendations. A multi-objective optimization model is constructed that considers traffic efficiency, safety risks, and energy costs. A multi-agent reinforcement learning (MARL) algorithm is used to establish differentiated decision-making solutions, which are then safety-verified. A real-time feedback mechanism for decision execution effectiveness is established to enable coordinated updates of cross-regional decision-making models and improve global adaptability. Decision instructions are then pushed to vehicles and infrastructure via a two-way communication link. In this embodiment, the multi-level response mechanism includes: Establish mapping rules between warning levels and response strategies. For example, if the warning level is low risk, text warnings (such as "the road ahead is slippery") will be pushed to surrounding vehicles, and cautious driving prompts will be given through the in-vehicle navigation. If the warning level is medium risk, the timing of roadside traffic lights will be optimized (such as extending the green light to alleviate congestion) and route suggestions will be sent to the logistics fleet. If the warning level is high risk, cross-regional emergency linkage will be initiated, and event details and disposal plans will be pushed to the traffic management center. At the same time, speed limit instructions will be enforced through a two-way communication link (such as a 40km / h speed limit on the accident section).

[0021] In this embodiment, the construction of a multi-dimensional perception network realizes the fusion of multi-source data and the unification of time and space, and constructs a multi-dimensional fusion perception system of vehicle-road-environment. Dynamic traffic status analysis can identify abnormal events in multiple dimensions. Hierarchical feature extraction and multi-dimensional anomaly recognition models can realize parallel detection of accidents, congestion, and severe weather. Comprehensive risk prediction and assessment can accurately predict the propagation path and risk, solve the one-sided problem of traditional risk assessment, and predict the expansion path of the impact range of events. The multi-level response mechanism in early warning generation and decision-making can take differentiated measures according to the risk level. The application of multi-objective optimization model and MARL algorithm makes decision-making more scientific, and cross-regional collaborative updates improve global adaptability.

[0022] In this embodiment, the multi-dimensional perception network construction further includes: Building bidirectional communication links between vehicles and between vehicles and infrastructure based on the V2X communication components installed in the vehicles, and establishing a connection with the traffic cloud platform based on the V2X communication components; The vehicle sends vehicle status data packets to adjacent vehicles based on a two-way communication link at a preset period. These packets contain "beacon frames" such as location, speed, and heading angle. Roadside equipment transmits real-time information on traffic signal status, road construction / congestion, and other events, forming a dynamic communication network between vehicles and between vehicles and infrastructure. Based on the IEEE1588 clock synchronization protocol or the GPS Beidou satellite clock synchronization mechanism, the time synchronization between vehicles and between vehicles and various infrastructures is determined, achieving nanosecond-level time synchronization for all devices; The vehicle obtains monitoring information of adjacent vehicles based on a two-way communication link, including the real-time position, movement status and intention information of adjacent vehicles, and at the same time collects traffic signal status and road environment data actively uploaded by roadside infrastructure.

[0023] In this embodiment, dynamic traffic status analysis also includes constructing a four-dimensional semantic network topology structure of "road-vehicle-environment-event": Taking road physical entities (such as road sections and intersections) as topological basic nodes, the structural skeleton of the road network topology map is constructed based on topological attributes (coordinate range, number of lanes, speed limit) and historical traffic characteristics (average daily flow, speed distribution); The real-time vehicle trajectories are mapped into dynamic edges, with the vehicle position transfer within adjacent time slices as the edge connection relationship. Based on the edge attributes including speed, acceleration, steering intention and other behavioral characteristics, an edge set reflecting the dynamic interaction of traffic flow is constructed. Environmental parameters (precipitation intensity, visibility, and road surface temperature) are used as weight factors in the graph structure and mapped to corresponding road nodes and vehicle behavior edges according to the spatiotemporal grid. This allows for a quantitative representation of the impact of environmental factors on traffic conditions. Identified abnormal events are then embedded in a four-dimensional semantic network topology, with node attributes including event type, impact range, and timeliness parameters. At the same time, the graph neural network (GNN) is used to learn the association relationship between nodes in the four-dimensional semantic network topology structure to construct a semantic representation of the traffic status.

[0024] In this embodiment, a dynamic communication network is constructed based on V2X communication components, and nanosecond-level time synchronization is achieved by combining IEEE1588 or satellite clock synchronization, breaking through the problems of traditional communication delay and time deviation, significantly enhancing the real-time and coordination of data, and visually presenting the traffic status through the four-dimensional semantic network topology structure of "road-vehicle-environment-event". It can more accurately depict the dynamic interaction of traffic flow and the influence of environmental factors, more comprehensively identify abnormal events, improve the accuracy of understanding complex traffic scenarios, reduce the prediction error of event propagation path, and provide more intuitive and accurate semantic representation for risk assessment and decision-making.

[0025] In this embodiment, the dynamic traffic state analysis and hierarchical feature extraction further includes: The road network topology is spatially discretized. Based on the physical coordinate range of the road and the lane division, the road network is divided into spatiotemporal grid cells containing attributes such as geographic coordinates, speed limits, and number of lanes. The basic traffic flow characteristics (volume, occupancy, average speed) within each spatiotemporal grid cell are extracted to form a spatial grid traffic state vector. Extract edge attributes formed by vehicle position transfer within adjacent time slices to generate a feature sequence that represents the temporal changes in vehicle movement. At the same time, combine historical traffic data to identify the periodic patterns and sudden change characteristics of traffic flow and generate a temporal dynamic feature vector. The spatial grid traffic state vector and the temporal dynamic feature vector are weightedly fused to form a composite feature representation containing spatiotemporal context, highlighting the characteristic influence of key spatiotemporal regions (such as the real-time impact weight of the accident section on the surrounding road network), and embedded in the four-dimensional semantic network topology structure.

[0026] In this embodiment, the spatiotemporal grid unit division and feature fusion strategy discretizes the road network space, combines the vehicle motion time series characteristics with the historical data patterns, and constructs a composite feature representation containing spatiotemporal context. It can accurately capture the impact of key areas and embed it into a four-dimensional semantic network topology structure to achieve a deep fusion analysis of spatial distribution and temporal dynamic changes, so that the granularity of traffic flow basic feature extraction is refined to the lane level. Combined with the recognition patterns of historical data, it can greatly predict sudden changes in traffic flow in advance. The composite feature representation effectively improves the accuracy of the impact weights of key areas such as accident sections, thereby improving the timeliness and accuracy of traffic status analysis.

[0027] In this embodiment, the comprehensive risk prediction assessment performs correlation analysis, specifically including: Obtain historical abnormal event data (e.g., the diffusion patterns of the past 1,000 accidents), combine it with the multidimensional spatial data and environmental parameters in the four-dimensional semantic network topology, and use a time series data mining algorithm to extract the propagation patterns of different types of abnormal events (accidents, congestion, bad weather) under different environmental parameters (e.g., rainy days, nighttime), and build a propagation rule knowledge base. The rule knowledge base includes the mapping relationship between event type, environmental conditions, and propagation path; Taking the identified abnormal events as key nodes, based on the association relationship and edge attributes between nodes in the four-dimensional semantic network topology structure, and based on the propagation rule knowledge base, the potentially affected road nodes and adjacent event nodes are screened out to construct an event evolution tree; The tree node attributes include event type, spatiotemporal location, current impact, and severity. The tree node connection relationship attributes include event propagation direction, propagation probability, and time delay threshold (e.g., the probability of congestion propagating from section A to section B is 0.7, and the delay time is 5 minutes). In this embodiment, the initial propagation probability of each edge is calculated through an association rule mining algorithm, and the time delay in the edge attribute is determined by combining the average propagation time of historical data statistics. At the same time, the current road network traffic status parameters (such as real-time traffic volume and vehicle speed distribution) and the basic characteristics of traffic flow in the spatiotemporal grid unit are used as constraints to prune the branches of the evolution tree and eliminate low-probability propagation paths.

[0028] In this embodiment, predicting the propagation path specifically includes: Taking the key node of the abnormal event as the target node, multi-path sampling is performed on the connection relationship attributes in the event evolution tree to generate at least one propagation path. During each sampling, the corresponding branch path is selected according to the propagation probability distribution of the current target node; Combined with the propagation time distribution of similar historical events, the time series of each target node affected in each propagation path is obtained through the time series interpolation algorithm, generating a propagation path set with time and space dimensions. The propagation probability distribution information of each propagation path in the propagation path set is mapped to the spatiotemporal grid unit, and the probability of each grid unit being affected in the future period is accumulated and calculated to generate a risk probability cloud map for the future period, which intuitively shows the probability of different areas being affected at different time points; Identify the key hub nodes in the propagation path set, calculate the risk contribution of the key hub nodes in the propagation process (for example, if the node is blocked, the road network capacity will drop by 40%), and prioritize each key hub node based on the risk contribution as the priority treatment object for subsequent response.

[0029] In this embodiment, identifying key hub nodes in the propagation path set specifically includes: Extract node attributes and topological structure information of each road node in the propagation path set in the four-dimensional semantic network topological structure; In this embodiment, the traffic flow corresponding to the key time nodes is obtained from the road node attributes. The key time nodes include the time when the abnormal event occurs and the period when similar events are most likely to occur in history. In this embodiment, topological structure information is extracted, including the number of road nodes associated with the road node and the percentage change in traffic volume of the associated road nodes when the traffic volume of the road node increases by 1%. This data is obtained by analyzing the historical propagation data of edge attributes in the event evolution tree. Based on the node attributes and topological structure information, the basic flow characteristics of each road node and the flow transmission sensitivity between road nodes are obtained to determine the flow weight coefficient and flow transmission weight coefficient; In this embodiment, the average traffic flow of each road node at the key time node is used as the basic traffic feature. Based on the percentage change of traffic volume of each associated road node, the average percentage change of the associated road nodes corresponding to the traffic flow change of the road node is calculated to quantify the traffic transmission sensitivity between road nodes. In this embodiment, the traffic flow average value of each road node is calculated by comparing it with a preset traffic flow reference value (such as the average traffic flow of the road network or the designed traffic flow of the road section) to obtain a traffic flow weight coefficient, which reflects the relative importance of the traffic flow of the road node; In this embodiment, the average value of the change percentage of the associated road nodes of each road node is normalized to obtain the flow conduction weight coefficient, which represents the diffusion capacity of the influence of the road node on the surrounding road network; Based on the preset law enforcement management information corresponding to each road node (such as daily law enforcement frequency and emergency response priority), the law enforcement management weight of the road node is determined based on the management level corresponding to the law enforcement management method through a mapping function. Road nodes with high law enforcement priority are given higher weights, reflecting the relationship between management resource investment and the importance of road nodes. The law enforcement management weight, flow weight coefficient and flow conduction weight coefficient are comprehensively calculated to obtain the road topology weight value, which is compared with the preset weight threshold to screen out road nodes with a value higher than the preset weight threshold as key hub nodes.

[0030] In this embodiment, based on the four-dimensional semantic network, combined with historical data and real-time parameters, a time series mining algorithm is used to establish a knowledge base of abnormal event propagation rules, so that the accuracy of abnormal event propagation pattern recognition is effectively improved. Through the event evolution tree, multi-path sampling and pruning algorithm, the dynamic generation and accurate prediction of the propagation path are realized, which greatly improves the prediction efficiency, can intuitively display the spatiotemporal distribution of risks, provide visual decision support for traffic management, evaluate key hub nodes based on multi-dimensional weight coefficients, quantify the risk contribution of road nodes and rank them, provide a basis for accurate early warning and priority disposal, improve the efficiency of emergency resource allocation, improve the traffic capacity guarantee rate in key areas, effectively reduce the overall impact of abnormal events on the transportation system, and significantly enhance the ability to prevent and control traffic risks.

[0031] In this embodiment, before fusing the collected vehicle status, road environment and infrastructure data, the method further includes: performing noise reduction on the image data included in the road environment data; The denoising of the image data in the road environment data includes: Arbitrarily obtain an image from the road environment data as the image to be denoised; Performing grayscale processing on the image to be denoised to obtain a grayscale image; Randomly select a pixel point in the grayscale image as the first pixel point; determine the target area with the first pixel point as the center and a preset distance as the radius; Calculating the difference between the first pixel and the other pixels in the target area except the first pixel to obtain a plurality of difference values; comparing the plurality of difference values ​​with a preset difference threshold, and marking the pixels whose difference values ​​are greater than or equal to the preset difference threshold as abnormal; Traverse all pixels in the grayscale image and count the number of abnormal marks for each pixel; Comparing the number of abnormal marks with a preset abnormal mark threshold, taking pixels whose number of abnormal marks is greater than or equal to the preset abnormal mark threshold as first abnormal pixel points, to obtain a plurality of first abnormal pixel points; Perform edge recognition on the grayscale image to determine the edge pixels in the grayscale image; Subtracting a plurality of first abnormal pixel points from the edge pixel points to obtain a plurality of second abnormal pixel points; Calculating the discreteness of each second abnormal pixel point respectively, and comparing the discreteness with a preset discreteness threshold, taking the pixel point when the discreteness is greater than or equal to the preset discreteness threshold as a third abnormal pixel point, to obtain a plurality of third abnormal pixel points; Deleting several third abnormal pixels to obtain a denoised grayscale image; Traverse all images in the road environment data to obtain the denoised road environment data.

[0032] In this embodiment, the grayscale difference between the central pixel and other pixels within a certain range around it is calculated, and pixels whose difference exceeds a threshold are marked as abnormal. The essence of this is to capture pixels with significantly sudden changes in grayscale values ​​in a local area. This is different from the prior art in which directly determining abnormal pixels means making a final judgment on the pixel properties (normal / abnormal). If it is subsequently discovered that the threshold setting is unreasonable or the area range is inappropriate, the pixels that have been "determined" as abnormal will permanently lose their original properties (such as grayscale value, spatial position, etc.) and will be difficult to adjust retroactively. "Marking" only adds a temporary label to the pixel (such as "unverified abnormality"), retaining all the information of the original pixel. Subsequently, re-evaluation can be performed by adjusting the threshold, expanding / reducing the target area, etc., to avoid irreversible errors caused by initial parameter errors.

[0033] In this embodiment, edge recognition is performed on the grayscale image, specifically by using a pre-trained edge recognition model.

[0034] In this embodiment, a subtraction operation is performed on a plurality of first abnormal pixel points and the edge pixel points to obtain a plurality of second abnormal pixel points. Specifically, the pixel points that are repeated with the edge pixel points in the plurality of first abnormal pixel points are deleted, and the remaining pixel points are used as the second abnormal pixel points.

[0035] In this embodiment, the specific implementation method for calculating the discreteness of each second abnormal pixel point is as follows: the minimum Euclidean distance between each second abnormal pixel point and the edge pixel point is used as the discreteness of each second abnormal pixel point; and the average of the discreteness of all second abnormal pixels is used as the preset discreteness threshold.

[0036] The working principle and beneficial effects of the above technical solution are as follows: first, an image is randomly selected from the road environment data as the image to be denoised, and then grayscale processing is performed on it to convert the color image into a grayscale image. The purpose of this is to simplify the image data and facilitate subsequent pixel analysis, because the grayscale image only contains brightness information, which reduces the complexity of the data; for each pixel in the grayscale image (first randomly select one as the first pixel), a target area is determined with the pixel as the center and a preset distance as the radius; the difference between the first pixel and other pixels in the target area is calculated, and abnormal pixels are marked by comparing with a preset difference threshold; this process preliminarily determines which pixels may be noise points based on the brightness difference of pixels in the local area, because under normal circumstances, the brightness difference between adjacent pixels will not be too large; all pixels in the entire grayscale image are traversed, and the number of abnormal marks for each pixel is counted. The number of anomaly labels is compared with a preset anomaly label threshold to identify the first anomaly pixel. This step further filters out pixels that may be noise, because if a pixel is marked as anomaly in multiple local areas, it is more likely to be a noise pixel. Edge recognition is performed on the grayscale image to determine edge pixels. The first anomaly pixel is subtracted from the edge pixel to obtain the second anomaly pixel. This is because edge pixels have large brightness variations and may be mistaken for noise pixels. This method can eliminate the interference of edge pixels. The discreteness of each second anomaly pixel is calculated, which reflects the distribution characteristics of pixels in the local area. The discreteness is compared with the preset discreteness threshold to identify and delete the third anomaly pixel, thus obtaining the denoised grayscale image. Finally, all images in the road environment data are traversed to complete the denoising of the entire road environment data image portion. Through detailed pixel analysis and screening, noise points in the image can be more accurately identified and removed. Compared with some simpler denoising methods, this method considers multiple factors such as the differences between pixels in local areas, the accumulation of anomaly labels, and the special conditions of edge pixels, thereby better preserving the effective information of the image while removing noise.

[0037] In this embodiment, the comprehensive risk index of the abnormal event is calculated based on the degree of harm and the risk of spread of the abnormal event, including steps 1-2: Step 1: Obtain the average risk value of vehicle a in historical driving data and the propagation risk value of abnormal events on the driving path of vehicle a. Based on the average risk value, the propagation risk value, and the expected risk value of the abnormal event, determine the comprehensive risk index of the abnormal event for vehicle a. ; in, represents the comprehensive risk index of the abnormal event to vehicle a; n represents the total number of risk factors in the abnormal event; Represents the average risk value of vehicle a in historical driving data; represents the risk value of the abnormal event spreading to the driving path of the vehicle a; Represents the risk level value of the i-th risk factor type among n risk factors; The probability density function of the risk of occurrence of the i-th risk factor type; It represents the cumulative probability of the occurrence of risk of the i-th risk factor in the next unit time interval starting from time t; It represents the total cumulative probability of the occurrence of risk of the i-th risk factor in all future time starting from time t; Step 2: Determine the comprehensive risk index of the abnormal event based on the comprehensive risk index of the abnormal event to vehicle a and the total number of vehicles in the driving path of vehicle a; ; in, Indicates the comprehensive risk index of abnormal events; Indicates the total number of vehicles in the driving path of vehicle a.

[0038] In this embodiment, the propagation probability distribution information of each propagation path in the propagation path set can determine the propagation risk value of the driving path of the vehicle a.

[0039] The working principle and beneficial effects of the above technical solution are: it comprehensively considers multiple factors, including the vehicle's own historical risk value, the risk of abnormal events on the driving path, the risk level of multiple risk factors, and the probability of each risk factor occurring. This multi-dimensional assessment can more comprehensively reflect the risk status of abnormal events and avoid the one-sidedness of single-factor assessment; the time factor is introduced through the time integral in the probability density function; the probability of risk factors occurring can be evaluated according to different time intervals, so as to better adapt to the situation where risks change over time; the comprehensive risk index of a single vehicle is first calculated, and then the comprehensive risk index (T) of the abnormal event is obtained through comprehensive calculation of all vehicles; it can not only evaluate the risk of a single vehicle in an abnormal event, but also grasp the risk impact of abnormal events on the entire vehicle group on the driving path as a whole, which is helpful to formulate targeted risk management strategies.

[0040] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A transportation operation monitoring, early warning, and decision-making analysis method based on vehicle-road collaboration is characterized by: include: Multi-dimensional perception network construction: Build a multi-dimensional perception network to obtain real-time vehicle status, road environment and infrastructure data, and integrate the collected vehicle status, road environment and infrastructure data to generate multi-dimensional spatial data; Before fusing the collected vehicle status, road environment, and infrastructure data, the process also includes denoising the image data included in the road environment data, traversing all images in the road environment data, and obtaining the denoised road environment data. Dynamic traffic status analysis: Perform hierarchical feature extraction on the fused multi-dimensional spatial data to construct a dynamic traffic status model. Observe in real time the degree to which traffic flow parameters deviate from normal distribution. Combine historical observation data with real-time observations to perform multi-dimensional identification of abnormal events. Comprehensive risk prediction and assessment: Based on the identification results, the identified abnormal events are analyzed for correlation, the propagation path is predicted, and the comprehensive risk index of the abnormal events is calculated according to the degree of harm and propagation risk of the abnormal events; Warning generation and decision-making: Based on the comprehensive risk index, a multi-level response mechanism is triggered, and corresponding warning information is generated. A decision plan is established and the safety of the decision plan is verified. The decision instructions are pushed to vehicles and various infrastructure through a two-way communication link.

2. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 1 is characterized in that: The construction of multi-dimensional perception network also includes: Building bidirectional communication links between vehicles and between vehicles and infrastructure based on the V2X communication components installed in the vehicles, and establishing a connection with the traffic cloud platform based on the V2X communication components; The vehicle sends vehicle status data packets to adjacent vehicles based on a two-way communication link at a preset period, and each roadside device sends event information in real time, forming a dynamic communication network between vehicles and between vehicles and various infrastructures; Determine time synchronization between vehicles and between vehicles and infrastructure based on clock synchronization mechanism; The vehicle obtains monitoring information of adjacent vehicles based on a two-way communication link, including the real-time position, movement status and intention information of adjacent vehicles, and at the same time collects traffic signal status and road environment data actively uploaded by roadside infrastructure.

3. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 2 is characterized in that: Dynamic traffic status analysis also includes constructing a four-dimensional semantic network topology: Taking the physical entities of roads as the topological basic nodes, the structural skeleton of the road network topology map is constructed based on topological attributes and historical traffic characteristics; The real-time trajectory of the vehicle is mapped into a dynamic edge, and the vehicle position transfer within adjacent time slices is used as the edge connection relationship. Based on the edge attribute behavior characteristics, an edge set is constructed; Environmental parameters are used as weight factors in the graph structure, mapped to corresponding road nodes and vehicle behavior edges according to the spatiotemporal grid, and the identified abnormal events are embedded in the four-dimensional semantic network topology structure; At the same time, the association relationship between nodes in the four-dimensional semantic network topology structure is learned to construct a semantic representation of the traffic status.

4. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 3 is characterized in that: The dynamic traffic state analysis and hierarchical feature extraction further includes: The road network topology is spatially discretized. Based on the physical coordinate range of the road and the lane division, the road network is divided into spatiotemporal grid units. The basic characteristics of traffic flow in each spatiotemporal grid unit are extracted to form a spatial grid traffic state vector. Extract edge attributes formed by vehicle position transfer within adjacent time slices to generate a feature sequence that represents the temporal changes in vehicle movement. At the same time, combine historical traffic data to identify the periodic patterns and sudden change characteristics of traffic flow and generate a temporal dynamic feature vector. The spatial grid traffic state vector and the temporal dynamic feature vector are weightedly fused to form a composite feature representation, and then embedded into the four-dimensional semantic network topology structure.

5. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 4 is characterized in that: Comprehensive risk prediction assessment and correlation analysis, including: Acquire historical abnormal event data, combine it with the multidimensional spatial data and environmental parameters in the four-dimensional semantic network topology structure, extract the propagation patterns of different types of abnormal events under different environmental parameters, and build a propagation rule knowledge base; Taking the identified abnormal events as key nodes, based on the association relationship and edge attributes between nodes in the four-dimensional semantic network topology structure, and based on the propagation rule knowledge base, the potentially affected road nodes and adjacent event nodes are screened out to construct an event evolution tree; The attributes of tree nodes include event type, spatiotemporal location, current impact level, and severity level; the attributes of tree node connection relationships include event propagation direction, propagation probability, and time delay threshold.

6. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 5 is characterized in that: Predicting the propagation path, including: Taking the key node of the abnormal event as the target node, multi-path sampling is performed on the connection relationship attributes in the event evolution tree to generate at least one propagation path. During each sampling, the corresponding branch path is selected according to the propagation probability distribution of the current target node; Combined with the propagation time distribution of similar historical events, the time series of each target node affected in each propagation path is obtained to generate a propagation path set; Map the propagation probability distribution information of each propagation path in the propagation path set to the spatiotemporal grid unit, and cumulatively calculate the probability of each grid unit being affected in the future period to generate a risk probability cloud map for the future period; Identify the key hub nodes in the propagation path set, calculate the risk contribution of the key hub nodes in the propagation process, and prioritize each key hub node based on the risk contribution.

7. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 6 is characterized in that: Identify key hub nodes in the propagation path set, including: Extract node attributes and topological structure information of each road node in the propagation path set in the four-dimensional semantic network topological structure; Based on the node attributes and topological structure information, the basic flow characteristics of each road node and the flow transmission sensitivity between road nodes are obtained to determine the flow weight coefficient and flow transmission weight coefficient; According to the preset law enforcement management method information corresponding to each road node, the law enforcement management weight of the road node is determined based on the management level corresponding to the law enforcement management method; The law enforcement management weight, flow weight coefficient and flow conduction weight coefficient are comprehensively calculated to obtain the road topology weight value, which is compared with the preset weight threshold to screen out road nodes with a value higher than the preset weight threshold as key hub nodes.

8. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 7 is characterized in that: Multi-level response mechanism, including: Establish mapping rules between warning levels and response strategies. For example, if the warning level is low risk, text warnings will be pushed to surrounding vehicles, and cautious driving prompts will be given through the in-vehicle navigation. If the warning level is medium risk, the timing optimization of roadside traffic lights will be triggered simultaneously, and route suggestions will be sent to the logistics fleet. If the warning level is high risk, cross-regional emergency linkage will be initiated, and event details and disposal plans will be pushed to the traffic management center. At the same time, speed limit instructions will be enforced through a two-way communication link.

9. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 1, characterized in that: The denoising of the image data in the road environment data includes: Arbitrarily obtain an image from the road environment data as the image to be denoised; Performing grayscale processing on the image to be denoised to obtain a grayscale image; Randomly select a pixel point in the grayscale image as the first pixel point; determine the target area with the first pixel point as the center and a preset distance as the radius; Calculating the difference between the first pixel and the other pixels in the target area except the first pixel to obtain a plurality of difference values; comparing the plurality of difference values ​​with a preset difference threshold, and marking the pixels whose difference values ​​are greater than or equal to the preset difference threshold as abnormal; Traverse all pixels in the grayscale image and count the number of abnormal marks for each pixel; Comparing the number of abnormal marks with a preset abnormal mark threshold, taking pixels whose number of abnormal marks is greater than or equal to the preset abnormal mark threshold as first abnormal pixel points, to obtain a plurality of first abnormal pixel points; Perform edge recognition on the grayscale image to determine the edge pixels in the grayscale image; Subtracting a plurality of first abnormal pixel points from the edge pixel points to obtain a plurality of second abnormal pixel points; Calculating the discreteness of each second abnormal pixel point respectively, and comparing the discreteness with a preset discreteness threshold, taking the pixel point when the discreteness is greater than or equal to the preset discreteness threshold as a third abnormal pixel point, to obtain a plurality of third abnormal pixel points; Deleting several third abnormal pixels to obtain a denoised grayscale image; Traverse all images in the road environment data to obtain the denoised road environment data.

10. The method for monitoring, early warning and decision-making analysis of transportation operations based on vehicle-road collaboration according to claim 6, characterized in that: According to the degree of harm and spread risk of the abnormal event, the comprehensive risk index of the abnormal event is calculated, including steps 1-2: Step 1: Obtain the average risk value of vehicle a in historical driving data and the propagation risk value of abnormal events on the driving path of vehicle a. Based on the average risk value, the propagation risk value, and the expected risk value of the abnormal event, determine the comprehensive risk index of the abnormal event for vehicle a. ； in, represents the comprehensive risk index of the abnormal event to vehicle a; n represents the total number of risk factors in the abnormal event; Represents the average risk value of vehicle a in historical driving data; represents the risk value of the abnormal event spreading to the driving path of the vehicle a; Represents the risk level value of the i-th risk factor type among n risk factors; The probability density function of the risk of occurrence of the i-th risk factor type; It represents the cumulative probability of the occurrence of risk of the i-th risk factor in the next unit time interval starting from time t; It represents the total cumulative probability of the occurrence of risk of the i-th risk factor in all future time starting from time t; Step 2: Determine the comprehensive risk index of the abnormal event based on the comprehensive risk index of the abnormal event to vehicle a and the total number of vehicles in the driving path of vehicle a; ； in, Indicates the comprehensive risk index of abnormal events; Indicates the total number of vehicles in the driving path of vehicle a.

Images