CN-115423303-B - V2X dynamic electronic lane planning method and device based on dynamic traffic flow

CN115423303BCN 115423303 BCN115423303 BCN 115423303BCN-115423303-B

Abstract

The application discloses a V2X dynamic electronic lane planning method and device based on dynamic traffic flow, and relates to the technical field of intelligent management and control system application. The method includes constructing an input feature vector based on X 0 and an average number of vehicles and an average speed of vehicles per lane per time slice Based on the width proportion of the manual dividing lane, an output feature vector is constructed The method comprises the steps of establishing a training set, a verification set and a test set based on historical traffic data in a preset period of manually dividing lanes, establishing a shallow neural network model comprising an activation function Softmax and a loss function D (P, Y), carrying out iterative training on parameters of the shallow neural network model, testing the neural network model with the parameters trained, and deploying the neural network model to a dynamic electronic lane planning environment if the test achieves a preset accuracy. The application dynamically adjusts the electronic lane planning of the current traffic environment in real time based on the high-precision GNSS positioning and the road side equipment information, and realizes the flexible and easy-maintenance electronic lane effect with various available scenes.

Inventors

HUAN HUAN
YAN XUELIANG
SUN YAFENG

Assignees

云控智行(上海)汽车科技有限公司

Dates

Publication Date: 20260512
Application Date: 20220830

Claims (9)

1. A V2X dynamic electronic lane planning method based on dynamic traffic flow, comprising: s1, acquiring the total width X 0 of the road and the number m of lanes, setting a time slice interval n, and constructing an input feature vector based on X 0 and the average number and average speed of each lane in each time slice Based on the width proportion of the manually divided lanes, constructing an output feature vector Comprising m attribute values; S2, extracting information of each row according to the S1 field based on historical traffic data of a preset period of manually dividing lanes, and constructing a training set, a verification set and a test set; S3, constructing a shallow neural network model, wherein an activation function Softmax is used for determining that the sum of the lane width allocation proportion is 1, and a loss function D (P, Y) is based on lane width prediction Cross entropy with artificial division ratio; s4, performing iterative training on the shallow neural network model parameters based on the training set constructed in the S2 and the loss function of the S3 to determine optimal weights and offset parameters of Wr and br parameters, and performing fitting verification based on the verification set constructed in the S2; s5, testing the neural network model with the parameter training based on the test set constructed in the S2, and returning to the S4 training if the test does not reach the preset accuracy; s6, if the test reaches the preset accuracy, deploying the test to a dynamic electronic lane planning environment; the method further comprises the steps of: the road traffic data is collected in real time according to the road side sensing equipment, and an input feature vector is constructed Inputting the trained neural network model to obtain the prediction of the real-time dividing ratio of the electronic lane ; Distributing the total width of the current road to each lane according to the prediction proportion, and calculating the center line point trace coordinate of each lane based on the GNSS (Global navigation satellite System) road center line point trace coordinate; and updating the MAP message of the V2X communication message according to the width of each lane and the coordinate information of the central line trace, and broadcasting the MAP message as dynamic electronic lane information to the networked vehicles within a preset distance after the MAP message is coded.
2. The V2X dynamic electronic lane planning method based on dynamic traffic flow according to claim 1, characterized in that, when the number of lanes m=3, it specifically comprises: Obtaining the average passing vehicle number X 1 of the lane 1 in each n-second time slice, the average vehicle speed X 2 of the lane 1 in each n-second time slice, the average passing vehicle number X 3 of the lane 2 in each n-second time slice, the average vehicle speed X 4 of the lane 2 in each n-second time slice, the average passing vehicle number X 5 of the lane 3 in each n-second time slice, the average vehicle speed X 6 of the lane 3 in each n-second time slice, and constructing an input feature vector X= [ X 0 ,X 1 ,X 2 ,X 3 ,X 4 ,X 5 ,X 6 ]; based on the manually divided lane 1, lane 2, lane 3 width ratios Y 0 、Y 1 、Y 2 , an output feature vector y= [ Y 0 ,Y 1 ,Y 2 ] is constructed, wherein =1。
3. The V2X dynamic electronic lane planning method based on dynamic traffic flow according to claim 2, wherein said constructing a shallow neural network model specifically comprises: , Wherein, X 0 to X 6 are input feature vectors with strong correlation with the lane width dividing ratio defined in S1, P 0 、P 1 、P 2 is the target lane width distribution, w ij is the weight coefficient required for calculating the i-th lane width P i by using the real-time traffic feature X j , b i is the offset required for adding when calculating the i-th lane width, and w ij ,b i is determined by training.
4. A V2X dynamic electronic lane planning method based on dynamic traffic flow according to claim 3, characterized in that said activation function Softmax comprises in particular: , The activation function Softmax is used for determining that the sum of the lane width distribution ratios is 1, and Z i is the output value of the ith node of the last layer of the neural network.
5. The V2X dynamic electronic lane planning method based on dynamic traffic flow according to claim 4, wherein the output of the model has a probability distribution form with characteristics P 0 +P 1 +P 2 =1, and the loss function D (P, Y) specifically includes: , Wherein the loss function D (P, Y) is based on lane width prediction And the cross entropy of the artificial division ratio is Y i which is the actual width ratio of the road i, and P i which is the width ratio of the road i obtained by model prediction.
6. The V2X dynamic electronic lane planning method based on dynamic traffic flow according to claim 5, wherein said iteratively training said shallow neural network model parameters specifically comprises: And performing iterative training on the neural network model parameters on a high-performance server based on the training set and the loss function by using a BP algorithm.
7. The V2X dynamic electronic lane planning method based on dynamic traffic flow according to claim 6, wherein said testing the neural network model with the training of parameters to reach a preset accuracy comprises: randomly extracting data in a preset period from a test set and dividing the data into 4 groups according to traffic flow conditions, wherein the groups comprise no congestion, slight congestion, moderate congestion and severe congestion, and each group is provided with 1000 pieces of fixed data; predicting each piece of data in the 4 groups of data by using a neural network model trained by the complete parameters, and if the maximum probability item corresponding to the predicted guiding lane use probability distribution is the same as the lane use defined by the manual reality, determining that the prediction is correct; and respectively counting the accuracy of the prediction results of the 4 groups of data, wherein the accuracy is the ratio of the number of prediction accuracy of each group to the total number of test data of each group, if the accuracy is respectively larger than a preset threshold value corresponding to each group, determining that the actual use requirement is met, and if the accuracy is not met, determining that the actual use requirement is not met.
8. The V2X dynamic electronic lane planning method based on dynamic traffic flow according to claim 7, wherein said accuracy is greater than a preset threshold corresponding to each group, specifically including no congestion 70%, slight congestion 75%, moderate congestion 80%, heavy congestion 85%.
9. A V2X dynamic electronic lane planning device based on dynamic traffic flow is characterized by comprising the following specific components: a memory configured to store data and instructions; A processor in communication with a memory, wherein, when executing instructions in the memory, the processor is configured to: s1, acquiring the total width X 0 of the road and the number m of lanes, setting a time slice interval n, and constructing an input feature vector based on X 0 and the average number and average speed of each lane in each time slice Based on the width proportion of the manually divided lanes, constructing an output feature vector Comprising m attribute values; S2, extracting information of each row according to the S1 field based on historical traffic data of a preset period of manually dividing lanes, and constructing a training set, a verification set and a test set; S3, constructing a shallow neural network model, wherein an activation function Softmax is used for determining that the sum of the lane width allocation proportion is 1, and a loss function D (P, Y) is based on lane width prediction Cross entropy with artificial division ratio; s4, performing iterative training on the shallow neural network model parameters based on the training set constructed in the S2 and the loss function of the S3 to determine optimal weights and offset parameters of Wr and br parameters, and performing fitting verification based on the verification set constructed in the S2; s5, testing the neural network model with the parameter training based on the test set constructed in the S2, and returning to the S4 training if the test does not reach the preset accuracy; s6, if the test reaches the preset accuracy, deploying the test to a dynamic electronic lane planning environment; is also configured to construct an input feature vector based on real-time collection of road traffic data by the road side awareness apparatus Inputting the trained neural network model to obtain the prediction of the real-time dividing ratio of the electronic lane ; Distributing the total width of the current road to each lane according to the prediction proportion, and calculating the center line point trace coordinate of each lane based on the GNSS (Global navigation satellite System) road center line point trace coordinate; and updating the MAP message of the V2X communication message according to the width of each lane and the coordinate information of the central line trace, and broadcasting the MAP message as dynamic electronic lane information to the networked vehicles within a preset distance after the MAP message is coded.

Description

V2X dynamic electronic lane planning method and device based on dynamic traffic flow Technical Field The application relates to the technical field of intelligent management and control system application, in particular to a V2X dynamic electronic lane planning method and device based on dynamic traffic flow. Background For road traffic to be smooth, traffic management departments can divide lanes on roads, and commonly divide opposite lanes, fast and slow lanes and the like according to the driving direction and the driving speed. Conventionally, traffic management departments conduct lane planning on roads in advance, and special paint is used on physical road surfaces to divide lanes so as to distinguish functions and use authorities of different areas of the same road surface. The division of the lanes meets both the traffic standard and the traffic flow of the actual road section. When the vehicle runs on the road which is divided into fixed lanes according to the principle, the vehicle can indirectly realize the preset design targets (the running direction limit and the running speed limit) according to the lanes so as to achieve the purposes of simplifying traffic flow traffic blindness and avoiding traffic jams and traffic accidents. Although the traditional fixed lane scheme has obvious effect, the traditional fixed lane scheme still has the defects. The method mainly comprises the steps of 1) printing lane lines on the road surface, requiring special people to periodically check and maintain to ensure the availability of the lane lines, and having high maintenance cost, 2) realizing different optimal lane demarcation methods according to different factors such as weather conditions, traffic flow timeliness and the like, wherein the traditional lane demarcation method, namely the position, is inconvenient to dynamically adjust to meet the requirement of road condition change in a short period in a traditional lane demarcation method, and 3) in special occasions which are not suitable for dividing the fixed lanes by using paint, but do need to provide assistance for drivers in actual use, such as grasslands, village roads and the like, the traditional lane lines cannot effectively play roles. Therefore, it is desirable to provide a V2X dynamic electronic lane planning method and apparatus based on dynamic traffic flow, by fusing high-precision satellite positioning information GNSS, lane trace information described based on V2X communication method, and road side equipment to identify traffic states near the perceived lane in real time, such as running states of all traffic participants, average vehicle speed, average vehicle number, etc., that is, an automatic dynamic electronic lane dividing method and apparatus are provided. According to the method, the point-trace GNSS coordinate positions of the central lines of a plurality of lanes in a road in the MAP message in the V2X communication mechanism are dynamically adjusted automatically according to road conditions and traffic conditions and are broadcasted to vehicles with V2X communication capability on the periphery, so that the effect of displaying dynamic electronic lanes is achieved. Disclosure of Invention According to a first aspect of some embodiments of the present application, there is provided a V2X dynamic electronic lane planning method based on dynamic traffic flow, applied in a terminal (e.g., a networked vehicle, etc.), the method may include S1, acquiring a road total width X 0 and a lane number m, and setting a time slice interval n, constructing an input feature vector based on X 0 and an average vehicle number and an average vehicle speed of each lane in each time sliceComprises 2m+1 attribute values, and based on the width proportion of the manually divided lanes, constructing an output feature vectorThe method comprises m attribute values, S2, constructing a training set, a verification set and a test set by extracting information of each line according to S1 field based on historical traffic data of a preset period of manually dividing lanes, S3, constructing a shallow neural network model, wherein an activation function Softmax is used for determining that the sum of lane width distribution ratios is 1, and a loss function D (P, Y) is based on lane width predictionThe method comprises the steps of manually dividing cross entropy of the model parameters, carrying out iterative training on the shallow neural network model parameters based on a training set constructed in the step S2 and a loss function of the step S3 to determine optimal weights and offset parameters of Wr and br parameters, carrying out fitting verification based on a verification set constructed in the step S2, testing the neural network model trained by the parameters based on a test set constructed in the step S5, returning to the step S4 if the test does not reach a preset accuracy, and carrying out deployment to a dynamic electronic lane planning envir