A dynamic identification method for risk level of special road sections in intelligent network environment

Through the acquisition and analysis of multi-source and multi-scale traffic data in an intelligent connected environment, the risk level of special sections of highways is dynamically identified, which solves the problem that existing technology cannot provide risk information in real time, and improves drivers' vigilance and driving safety sex.

CN115938117BActive Publication Date: 2024-11-15CHINA MERCHANTS CHONGQING COMM RES & DESIGN INST
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211555524.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-06
Publication Date
2024-11-15
Estimated Expiration
2042-12-06

AI Technical Summary

Technical Problem

The driving safety model of special sections of the existing expressway is a passive statistical analysis and cannot provide risk level information in real time, making it difficult for drivers to know the risks in advance when facing complex traffic environments, increasing the probability of accidents.

Method used

Using multi-source and multi-scale traffic data acquisition and analysis methods in an intelligent networked environment, video and point cloud data are acquired through roadside cameras and radars, feature extraction network models are constructed, abnormal operating status features are extracted, and road congestion and vehicle collision risks are analyzed. , dynamically identify the risk level of special road sections.

Benefits of technology

Real-time identification of risk levels on special road sections is achieved, helping drivers to know risks in advance, reducing the probability of traffic accidents, improving the safety level of special sections of highways, and reducing the cost of smart highway facilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115938117B_ABST
    Figure CN115938117B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of intelligent transportation technology, and specifically to a method for dynamically identifying the risk level of a special road section in an intelligent network environment. By acquiring multi-source and multi-scale traffic data of a special road section, and then extracting abnormal operation state characteristics based on the multi-source and multi-scale traffic data, a macroscopic risk analysis of lane congestion and a microscopic risk analysis of collision between vehicles are performed on the special road section based on the abnormal operation state characteristics, and then the risk level of the special road section is acquired, thereby realizing dynamic identification of the risk level of the special road section, helping vehicle drivers to acquire the risk level of the special road section in advance, thereby increasing their vigilance against the special road section, and adopting more cautious and reasonable driving operations to pass through the special road section, effectively reducing the probability of traffic accidents on the special road section, and reducing the driving risk of the special road section.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a method for dynamically identifying the risk level of a special road section in an intelligent network environment. Background Art

[0002] Special sections of highways, such as merge and diverge areas, tunnel entrances and exits, etc., account for a relatively low proportion of the total road mileage, but such sections have complex traffic environments and frequent lane changes, and are potential accident sites where major and serious accidents are concentrated. In addition, drivers have poor adaptability to special sections, which makes the number of accidents in special sections much higher than that in general sections. Special sections of highways are important nodes for safety risk management and control, and are also one of the goals of smart highway construction.

[0003] Most of the existing driving safety models for special sections of highways are post-evaluation statistical analysis and modeling prediction. The management system built on this is often a passive information release model. Vehicle drivers are unable to obtain risk level information of special sections at the beginning of abnormal behavior or even before reaching the event section, causing drivers to face special sections in a conventional driving manner, increasing the driving risk of special sections. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a method for dynamically identifying the risk level of special road sections in an intelligent network environment to reduce the driving risk on special road sections.

[0005] The technical solution adopted by the present invention is a method for dynamically identifying the risk level of special road sections in an intelligent network environment.

[0006] In a first achievable manner, a method for dynamically identifying the risk level of a special road section in an intelligent network environment includes:

[0007] Obtain multi-source and multi-scale traffic data for special road sections;

[0008] Extract abnormal operation status characteristics based on multi-source and multi-scale traffic data;

[0009] Analyze the road congestion situation according to the abnormal operation status characteristics and obtain the road congestion risk value;

[0010] Analyze the vehicle collision situation according to the abnormal operation status characteristics to obtain the vehicle collision risk value;

[0011] The risk level of special road sections is obtained based on the road congestion risk value and vehicle collision risk value.

[0012] In combination with the first feasible method, in the second feasible method, multi-source and multi-scale traffic data of a special road section is obtained, including:

[0013] Use roadside smart cameras to collect data on special road sections and obtain video information;

[0014] Acquire lane-level traffic operation status data based on video information;

[0015] Use roadside radar to collect data on special road sections and obtain point cloud data;

[0016] The operating parameters of vehicles in each lane are obtained based on the point cloud data, and the lane-level traffic operation status data and the operating parameters of vehicles in each lane are determined as multi-source and multi-scale traffic data for special sections.

[0017] In combination with the second implementable method, in the third implementable method, abnormal operation state features are extracted based on multi-source and multi-scale traffic data, including:

[0018] Construct feature extraction network model;

[0019] Input multi-source and multi-scale traffic data into the feature extraction network model to obtain the traffic operation feature vector;

[0020] The traffic operation feature vectors are clustered and analyzed to obtain the abnormal operation status characteristics.

[0021] In combination with the third implementable method, in the fourth implementable method, a feature extraction network model is constructed, including: connecting the input layer, 5 convolutional layers, a fully connected layer and a classification layer in sequence; inputting data into the input layer and transmitting the input data to the 5 convolutional layers; the 5 convolutional layers perform convolution operations on the input data in sequence to obtain the output results of the convolutional layers; inputting the output results of the convolutional layers into the fully connected layers to obtain the output results of the fully connected layers; inputting the output results of the fully connected layers into the classification layer for classification to obtain the output results of the feature extraction network model.

[0022] In combination with the third implementable manner, in the fifth implementable manner, the traffic operation feature vector is clustered and analyzed to obtain abnormal operation state features, including:

[0023] Obtain the distance between each traffic operation feature vector;

[0024] Continuously select the two traffic operation feature vectors with the smallest distance and merge them into a new traffic operation feature vector until the last traffic operation feature vector is left;

[0025] The two selected traffic operation characteristic vectors with the smallest distance are respectively used as branches, and the new traffic operation characteristic vectors obtained after merging are used as trunks to draw a traffic characteristic tree diagram;

[0026] Obtain abnormal operation status characteristics based on the traffic characteristic tree diagram.

[0027] In combination with the first possible implementation, in the sixth possible implementation, the road congestion situation is analyzed according to the abnormal operation state characteristics to obtain the road congestion risk value, including:

[0028] Get the average speed of the road section;

[0029] Obtain the expected speed of the road network based on the average speed of the road section;

[0030] Obtain travel time ratio based on expected vehicle speed on the road network;

[0031] Obtain road congestion risk value based on travel time ratio.

[0032] In combination with the first possible implementation, in the seventh possible implementation, the vehicle collision situation is analyzed according to the abnormal operation state characteristics to obtain the vehicle collision risk value, including:

[0033] Obtain the predicted collision time, rear intrusion time and dynamic position of the vehicle according to the abnormal running state characteristics;

[0034] The vehicle collision risk value is obtained based on the vehicle's predicted collision time, rear intrusion time and dynamic position.

[0035] In combination with the first achievable method, in the eighth achievable method, the risk level of a special road section is obtained according to the road congestion risk value and the vehicle collision risk value, including:

[0036] The DBSCAN clustering algorithm is used to cluster the congestion risk value and collision risk value to obtain multiple risk clusters;

[0037] The risk level of special road sections is obtained based on each risk cluster.

[0038] In combination with the eighth achievable manner, in a ninth achievable manner, the risk level of a special road section is obtained according to each risk cluster, including:

[0039] Classify each risk cluster according to the cluster classification;

[0040] Obtain the risk level corresponding to each risk cluster according to the cluster classification to which each risk cluster belongs;

[0041] The risk level of each location on the special road section is obtained according to the risk level corresponding to each risk cluster, and each risk cluster corresponds to each location on the special road section one by one.

[0042] It can be seen from the above technical solution that the beneficial technical effects of the present invention are as follows:

[0043] By acquiring multi-source and multi-scale traffic data of special sections, and then extracting abnormal operation status characteristics based on the multi-source and multi-scale traffic data, the macro-risk analysis of lane congestion and micro-risk analysis of vehicle collisions are performed on special sections based on the abnormal operation status characteristics, and then the risk level of special sections is obtained, thereby realizing dynamic identification of the risk level of special sections, helping vehicle drivers to obtain the risk level of special sections in advance, thereby increasing their vigilance on special sections and adopting more cautious and reasonable driving operations to pass through special sections, effectively reducing the probability of traffic accidents on special sections and reducing driving risks on special sections. At the same time, the real-time assessment of traffic operation risk levels provides technical support for the real-time assessment and control of risks in special sections of highways and the release of control information. At the same time, it is of great significance to improve the traffic safety level of special sections of highways and reduce the cost of smart highway facilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the prior art description. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0045] Figure 1 A schematic diagram of a method for dynamically identifying the risk level of a special road section in an intelligent network environment provided by the present invention;

[0046] Figure 2 A schematic diagram of a feature extraction network model provided by the present invention. DETAILED DESCRIPTION

[0047] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0048] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should be the common meanings understood by technicians in the field to which the present invention belongs. The terms "first", "second", etc. in the specification and claims of the embodiments of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present disclosure described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. Unless otherwise specified, the term "multiple" means two or more. In the embodiments of the present disclosure, the character " / " indicates that the objects before and after are in an "or" relationship. For example, A / B means: A or B. The term "and / or" is a description of the association relationship of objects, indicating that there can be three relationships. For example, A and / or B means: A or B, or, A and B. The term "corresponding" can refer to an association relationship or a binding relationship, and A and B corresponding to each other means that there is an association relationship or a binding relationship between A and B.

[0049] Combination Figure 1 As shown, this embodiment provides a method for dynamically identifying the risk level of a special road section in an intelligent network environment, including:

[0050] Step S01, obtaining multi-source and multi-scale traffic data of a special road section;

[0051] Step S02: extracting abnormal operation status features based on multi-source and multi-scale traffic data;

[0052] Step S03: Analyze the road congestion situation according to the abnormal operation state characteristics to obtain a road congestion risk value;

[0053] Step S04: analyzing the vehicle collision situation according to the abnormal running state characteristics to obtain a vehicle collision risk value;

[0054] Step S05: Obtain the risk level of the special road section according to the road congestion risk value and the vehicle collision risk value.

[0055] By acquiring multi-source and multi-scale traffic data of special sections, and then extracting abnormal operating status characteristics based on the multi-source and multi-scale traffic data, and then obtaining the risk level of special sections, dynamic identification of the risk level of special sections can be achieved, so as to help vehicle drivers obtain the risk level of special sections in advance, thereby increasing their vigilance on special sections and adopting more cautious and reasonable driving operations to pass through special sections, effectively reducing the probability of traffic accidents on special sections.

[0056] Optionally, multi-source and multi-scale traffic data of special road sections are obtained, including: using roadside intelligent cameras to collect data on special road sections to obtain video information; obtaining lane-level traffic operation status data based on the video information; using roadside radars to collect data on special road sections to obtain point cloud data; obtaining operating parameters of vehicles in each lane based on the point cloud data, and determining the lane-level traffic operation status data and the operating parameters of vehicles in each lane as multi-source and multi-scale traffic data for the special road section.

[0057] In some embodiments, a roadside smart camera is used to obtain lane-by-lane video information of special sections of highway merge areas and tunnel entrances and exits, and all-weather lane-level traffic operation status data is extracted based on video information analysis. Lane-level traffic operation status data includes lane occupancy, event location, event type, etc. Point cloud data of each lane of a special section is obtained through a roadside radar, and the operation parameters of the vehicle in each lane are extracted based on point cloud data analysis. The operation parameters include real-time position, trajectory, instantaneous speed, average speed, acceleration, etc. In some embodiments, the lane occupancy is obtained from the vehicle information in the video information, and the radar directly outputs information such as event location, event type, real-time position, trajectory, instantaneous speed, average speed, acceleration, etc. of the vehicle.

[0058] Through the lane-by-lane video information and point cloud data of each lane of special sections obtained in the intelligent network environment, we can dynamically perceive the vehicle, lane environment and other information in real time, accurately judge and predict the risk level of the sections and dynamically control them, thereby transforming passive information release into active safety control, effectively improving driving safety on special sections of highways.

[0059] Optionally, extracting abnormal operation status features based on multi-source and multi-scale traffic data includes: constructing a feature extraction network model; inputting the multi-source and multi-scale traffic data into the feature extraction network model to obtain a traffic operation feature vector; and performing cluster analysis on the traffic operation feature vector to obtain abnormal operation status features.

[0060] In some embodiments, multi-source sensing devices such as roadside cameras and roadside radars collect real-time traffic data, and feature extraction is performed on the real-time traffic data collected by the multi-source sensing devices through convolutional neural networks (CNN) and hierarchical clustering methods to obtain abnormal operating status features.

[0061] In some embodiments, the multi-source multi-scale traffic data is a dataset {x i}, where x i =[a,b,…,n] 30x30 , where a, b, n are variables a, b, n in multi-source and multi-scale traffic data; function ψ represents the feature extraction process, {x i}After feature extraction, it becomes {ψ(x i )}, where ψ(x i )[p,q,…,z] 1x64 , where p, q, and z are abnormal operating status features. The data feature extraction process is to transform the data format, value, etc. to highlight the representative features of the data. The feature extraction process based on convolutional neural network is to build a feature extraction model ψ.

[0062] Optionally, constructing a feature extraction network model includes: connecting the input layer, 5 convolutional layers, a fully connected layer and a classification layer in sequence; inputting data into the input layer and transmitting the input data to the 5 convolutional layers; the 5 convolutional layers sequentially perform convolution operations on the input data to obtain the output results of the convolutional layers; inputting the output results of the convolutional layers into the fully connected layers to obtain the output results of the fully connected layers; inputting the output results of the fully connected layers into the classification layer for classification to obtain the output results of the feature extraction network model.

[0063] Optionally, combined Figure 2 As shown, the multi-source and multi-scale traffic data is a dataset {x i}, where x i =[a,b,…,n] 30x30 , the input layer will {x i} Input the convolution layer, after the first convolution output result L1, L1 inputs the second convolution, obtains the second convolution result L2, L2 inputs the third convolution, obtains the third convolution result L3, L3 inputs the fourth convolution, obtains the fourth convolution result L4, L4 inputs the fifth convolution, obtains the fifth convolution result L5, L5 is a 4x4x64 matrix. After L5 is input into the fully connected layer, the fully connected layer integrates the data to obtain the intermediate value L6, and classifies L6 to obtain the fully connected output result L7. After L7 passes through the softmax classification layer, the abnormal operation status characteristics are obtained.

[0064] Optionally, the convolutional neural network (CNN) includes an input layer, a convolution layer, a pooling layer, an activation function layer, a fully connected layer and an output sofmax layer connected in sequence.

[0065] In some embodiments, a feature extraction network structure is designed based on the matrix dimension of the original data, the dimensionality requirements of the output features, and the model structure characteristics of a typical convolutional neural network. The model input data includes three floating-point types of characteristic data: vehicle coordinates, speed, and acceleration. The basic unit is a (1×3)-dimensional [coordinate, speed, acceleration] vector. 300 groups of basic units constitute a (30×30)-dimensional input matrix, which represents a target characteristic data set within a time series. After the input matrix is ​​calculated through multiple convolutions, pooling, and activation functions, a (1×64)-dimensional feature vector is output at the 7th layer. Since the convolutional neural network model has excellent generalization ability, the output feature vector is not only greatly reduced in dimension, but also has better feature attributes, which is easy to analyze and study related algorithms. Convolutional neural networks do not require a backpropagation calculation process when extracting data features. The feature extraction performance of forward calculations is high, which can effectively cope with the challenge of increasing data volume.

[0066] Optionally, cluster analysis is performed on the traffic operation characteristic vectors to obtain abnormal operation status characteristics, including: obtaining the distance between each traffic operation characteristic vector; continuously selecting two traffic operation characteristic vectors with the smallest distance to merge into a new traffic operation characteristic vector until the last traffic operation characteristic vector remains; using the two selected traffic operation characteristic vectors with the smallest distance as branches and each new traffic operation characteristic vector obtained after merging as a trunk to draw a traffic characteristic tree diagram; and obtaining abnormal operation status characteristics according to the traffic characteristic tree diagram.

[0067] In some embodiments, cluster analysis of traffic operation feature vectors includes the following steps: Step S11, treating each traffic operation feature vector as a single cluster, and obtaining the distance between each cluster;

[0068] Step S12, merging the clusters with the smallest distance into a new cluster;

[0069] Step S13, determine whether the number of clusters is 1, if not 1, re-obtain the distance between each cluster and return to step S12, if it is 1, execute step S14;

[0070] Step S14, using each cluster before merging as a branch and each cluster after merging as a trunk to draw a traffic feature tree diagram;

[0071] Step S15: Acquire abnormal operation status characteristics according to the traffic characteristic tree diagram.

[0072] Optionally, acquiring abnormal operation status characteristics according to the traffic characteristic tree diagram includes: determining a demand quantity of the abnormal operation status characteristics, and extracting the abnormal operation status characteristics corresponding to the demand quantity in the traffic characteristic tree diagram.

[0073] In some embodiments, clustering analysis of traffic operation feature vectors is an unsupervised learning algorithm that groups unlabeled data points with similar features together, further clusters the extracted features, explores the true laws of the data, and obtains more representative abnormal operation status feature data.

[0074] In some embodiments, a macro-level analysis is performed on special road sections to obtain a road congestion risk value, and a micro-level analysis is performed on special road sections to obtain a vehicle collision risk value, thereby obtaining a traffic operation risk level that covers congestion risk (flow, speed, density of sections and sections) and collision risk. In this way, the operation risks of highway merge and diverge areas and tunnel entrances and exits are analyzed and judged from the micro and macro levels, which improves accuracy.

[0075] Optionally, the road congestion situation is analyzed according to the abnormal operating state characteristics to obtain a road congestion risk value, including: obtaining the average vehicle speed of the road section; obtaining the expected vehicle speed of the road network according to the average vehicle speed of the road section; obtaining the travel time ratio according to the expected vehicle speed of the road network; and obtaining the road congestion risk value according to the travel time ratio.

[0076] Optionally, obtaining the average vehicle speed of a road section includes:

[0077]

[0078] In the above formula, is the average speed of the road section, that is, the average speed of all sample vehicles passing through road section i within a specified time, l ij is the travel distance of the jth vehicle on road section i, t ij is the travel time of the jth vehicle on road section i, and m is the total number of vehicles running in abnormal states on road section i.

[0079] Optionally, obtaining the average speed of the road network includes:

[0080]

[0081] In the above formula, is the average speed of the road network, that is, the average speed of all sample vehicles in the road network within a specified time. is the average speed of road section i, l i The length of section i, q i The flow rate of road section i, n is the total number of all sample vehicles on road section i.

[0082] Optionally, obtaining the expected vehicle speed of the road network includes:

[0083]

[0084] In the above formula, is the expected speed of the road network, VKTi is the weight of road section i, v 85%i The 85th percentile accumulated average speed of segment i.

[0085] In some embodiments, under the free flow state, the 85th percentile vehicle speed of each type of vehicle on the cumulative speed distribution curve is used as the basis for limiting the maximum speed of the vehicle, VKT is defined as the weight of the road section, and the weight of each level of road is used to perform a weighted average of the 85th percentile cumulative speed of each level of road to obtain the expected speed of the road network.

[0086] Optionally, obtaining the travel time ratio includes:

[0087]

[0088] In the above formula, R T is the travel time ratio, is the actual travel time of the road network, is the expected travel time of the road network, is the expected vehicle speed of the road network.

[0089] Optionally, obtaining the road congestion risk value includes: inputting the travel time ratio into TPI=F(R T ), and obtain the congestion risk value TPI, where F(*) is the conversion relationship determined by the expert scoring.

[0090] Optionally, after obtaining the road congestion risk value, the TPI value is converted into a corresponding road network operation level according to a road network operation level classification standard.

[0091] In some embodiments, the travel time ratio is the ratio of the actual travel time of a road section or road network to the expected travel time, indicating the severity of the extra time spent on the current road condition compared to the expected speed condition. The traffic operation index of the travel time ratio is used to characterize the congestion risk, and the travel time ratio is converted to the TPI through expert scoring to obtain a specific vehicle congestion risk value.

[0092] Optionally, the vehicle collision situation is analyzed according to the abnormal operating state characteristics to obtain a vehicle collision risk value, including: obtaining the vehicle's predicted collision time, rear intrusion time and dynamic position according to the abnormal operating state characteristics; obtaining the vehicle's collision risk value according to the vehicle's predicted collision time, rear intrusion time and dynamic position.

[0093] Optionally, the predicted collision time, rear intrusion time and dynamic position of the vehicle are compared with preset thresholds to obtain a vehicle collision risk value.

[0094] Optionally, the abnormal operating state characteristics include operating parameters and road information of the vehicle in the abnormal operating state, and the distance and relative speed of any adjacent vehicles and the time it takes for each vehicle to arrive at a preset position are obtained based on the operating parameters and road information of the vehicle in the abnormal operating state.

[0095] Optionally, the distance between the two vehicles is divided by the relative speed of the two vehicles to obtain the predicted collision time of the vehicles.

[0096] Optionally, the rear invasion time is the time difference between two vehicles arriving at a specified section, which refers to the conflict time during the straight-ahead and right-turn merging processes. The rear invasion time is a traffic conflict identification indicator measured in time, and is also an evaluation indicator of the severity of the conflict.

[0097] In some embodiments, the post-intrusion time, i.e., the time difference between the i-th vehicle and the preceding vehicle (i.e., the i-1 vehicle) when they arrive at a certain specified section, is i The calculation formula for PET is i =t i-1 -t i , the time difference PET between the i-1th vehicle and the following vehicle (i.e., vehicle i) arriving at a specified section i-1 The calculation formula for PET is i-1 =t i -t i-1 .

[0098] Optionally, obtaining the dynamic position of the vehicle includes: initializing w i =1 / m i , x i is the position of the vehicle, v i is the speed of the vehicle, and the following steps are repeated once every Δt:

[0099] Step S31, processing external force: v i ←v i +Δtw i f ext (x i );

[0100] Step S32, velocity damping: dampVelocities(v1, v2..., v N );

[0101] Step S33, next position prediction: p i ←x i +Δtv i

[0102] Step S34: Collision detection, generating M col1 Collision constraints: generateCollisionConstraints(xi →p i );

[0103] Step S35: perform several iterations, processing all constraints once each time.

[0104] Step S36: Update the speed and position of the vehicle: i ←(p i -x1) / Δt,x i ←p i ;

[0105] Step S37: Update the velocity of the collision node of the adjacent vehicle: velocity Update(v i , ..., v N );

[0106] In the above steps, x i is the vehicle position in abnormal operation state, v i is the vehicle speed in abnormal operation state, w i is the angular velocity, m i is the mass, f ext (*) is the force that cannot be converted into position constraint, p i is the predicted next position of the vehicle, Given constraints and time-varying collision constraints. dampVelocities represents velocity damping, projectConstrainst represents system constraints, and velocity Update represents velocity update.

[0107] In some embodiments, the predicted collision time and post-intrusion time are used to analyze road risks in the time dimension, and the time dimension focuses on the traffic operation state changes before, during, and after abnormal events. The dynamic position is used to analyze road risks in the spatial dimension, and the spatial dimension focuses on the traffic operation state changes in the section 100-300 meters before and after the special section (such as the nose of the merge area, the tunnel entrance and exit section, etc.). In this way, the collision time, post-intrusion time, and dynamic position are used as vehicle collision risk parameters to analyze the special road section, and the traffic operation state characteristics and similarities and differences of the special road section in the time and space dimensions are obtained, which is conducive to clarifying the evolution law of vehicle collision risks in special road sections, and is conducive to identifying the turning points and mutation points of the traffic operation state in a certain time and space period, and further improving the accuracy of identifying the risk level of special road sections.

[0108] Optionally, the risk level of the special road section is obtained according to the road congestion risk value and the vehicle collision risk value, including: clustering the congestion risk value and the collision risk value using a DBSCAN clustering algorithm to obtain multiple risk clusters; and obtaining the risk level of the special road section according to each risk cluster.

[0109] In some embodiments, the DBSCAN clustering process is as follows:

[0110] Step S21: Input sample set D = {x1, x2, ..., x m}, the sample set D includes road congestion risk value and vehicle collision risk value, and the neighborhood parameter {∈, MinPts};

[0111] Step S22, initializing the core object set and initializing the category;

[0112] Step S23, traverse the elements of D to determine whether they are core objects, and if so, add them to the core object set;

[0113] Step S24, determine whether all elements in the core object set have been accessed, if yes, the algorithm ends, if not, execute step S25;

[0114] Step S25, randomly select an unvisited core object o in the core object set Ω, first mark o as visited, then mark o as category k, and finally store the unvisited data in o's ∈ neighborhood into the seed set Seeds;

[0115] Step S26: Determine whether the seed set Seeds is Φ. If so, the current cluster C k If the generation is completed and k=k+1, jump to step S24. If not, execute step S27;

[0116] Step S27, select a seed point Seed from the seed set Seeds, mark it as visited, and mark category k; then determine whether Seed is a core object. If so, add the unvisited seed points in Seed to the seed set and jump to step S26.

[0117] Optionally, the risk level of the special road section is obtained according to each risk cluster, including: classifying each risk cluster according to the cluster classification; obtaining the risk level corresponding to each risk cluster according to the cluster classification to which each risk cluster belongs; obtaining the risk level of each location of the special road section according to the risk level corresponding to each risk cluster, and each risk cluster corresponds one-to-one to each location of the special road section.

[0118] In some embodiments, the preset clusters are classified into safe clusters, warning clusters and dangerous clusters. After clustering the risk values, multiple risk clusters are obtained. After classifying each risk cluster according to the cluster classification, the risk level corresponding to each risk cluster is obtained according to the cluster classification to which each risk cluster belongs. The risk levels of different locations on special sections of the highway are obtained according to the correspondence between each risk cluster and each location on the special section.

[0119] In some embodiments, a density-based clustering method with noise is used to perform parameter clustering on road congestion risk values ​​and vehicle collision risk values ​​in the risk level set, and the risk levels are divided into three levels: safe, alert, and dangerous according to the clustering situation. The thresholds of the multidimensional risk levels are then inferred, thereby achieving dynamic identification and division of risk levels of special sections of highways.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for dynamically identifying the risk level of special road sections in an intelligent network environment, characterized in that: include: Obtain multi-source and multi-scale traffic data for special road sections; Extracting abnormal operation status features according to the multi-source and multi-scale traffic data; Analyze the road congestion situation according to the abnormal operation state characteristics to obtain a road congestion risk value; Analyze the vehicle collision situation according to the abnormal running state characteristics to obtain a vehicle collision risk value; Obtaining a risk level of a special road section according to the road congestion risk value and the vehicle collision risk value; Obtain multi-source and multi-scale traffic data for special road sections, including: Use roadside smart cameras to collect data on special road sections and obtain video information; Acquire lane-level traffic operation status data based on video information; Use roadside radar to collect data on special road sections and obtain point cloud data; The operating parameters of the vehicles in each lane are obtained according to the point cloud data, and the lane-level traffic operating status data and the operating parameters of the vehicles in each lane are determined as multi-source and multi-scale traffic data of special sections; Extracting abnormal operation status features according to the multi-source and multi-scale traffic data includes: Construct feature extraction network model; Inputting the multi-source and multi-scale traffic data into the feature extraction network model to obtain a traffic operation feature vector; Performing cluster analysis on the traffic operation feature vector to obtain abnormal operation state features; Constructing the feature extraction network model includes: sequentially connecting an input layer, five convolutional layers, a fully connected layer, and a classification layer; inputting data into the input layer, and transmitting the input data to the five convolutional layers; performing convolution operations on the input data in turn to obtain output results of the convolutional layers; inputting the output results of the convolutional layers into the fully connected layers to obtain output results of the fully connected layers; inputting the output results of the fully connected layers into the classification layer for classification to obtain output results of the feature extraction network model; The traffic operation feature vector is clustered and analyzed to obtain abnormal operation state features, including: Obtain the distance between each traffic operation feature vector; Continuously select the two traffic operation feature vectors with the smallest distance and merge them into a new traffic operation feature vector until the last traffic operation feature vector is left; The two selected traffic operation characteristic vectors with the smallest distance are respectively used as branches, and the new traffic operation characteristic vectors obtained after merging are used as trunks to draw a traffic characteristic tree diagram; Obtain abnormal operation status characteristics based on the traffic characteristic tree diagram; Obtaining the risk level of a special road section according to the road congestion risk value and the vehicle collision risk value includes: Using a DBSCAN clustering algorithm to cluster the congestion risk value and the collision risk value to obtain multiple risk clusters; The risk level of special road sections is obtained based on each risk cluster.

2. The method according to claim 1, characterized in that Analyze the road congestion situation according to the abnormal operation status characteristics and obtain the road congestion risk value, including: Get the average speed of the road section; Obtain the expected speed of the road network based on the average speed of the road section; Obtaining a travel time ratio according to the expected vehicle speed of the road network; A road congestion risk value is obtained according to the travel time ratio.

3. The method according to claim 1, characterized in that Analyze the vehicle collision situation according to the abnormal operation status characteristics to obtain the vehicle collision risk value, including: Obtain the predicted collision time, rear intrusion time and dynamic position of the vehicle according to the abnormal running state characteristics; The vehicle collision risk value is obtained based on the vehicle's predicted collision time, rear intrusion time and dynamic position.

4. The method according to claim 3, characterized in that Obtain the risk level of a particular road section based on each risk cluster, including: Classify each risk cluster according to the cluster classification; Obtain the risk level corresponding to each risk cluster according to the cluster classification to which each risk cluster belongs; The risk level of each position of the special road section is obtained according to the risk level corresponding to each risk cluster, and each risk cluster corresponds to each position of the special road section one by one.