A method and system for evaluating the impact of intelligent connected vehicle on road test traffic flow

Abstract

The present invention discloses a method for evaluating the impact of traffic flow on road tests of intelligent networked vehicles, including collecting vehicle driving status through a machine vision system and collecting vehicle position status through a GNSS positioning system; calculating traffic flow impact indicators according to vehicle form status and vehicle position status, the traffic flow impact indicators include vehicle speed, vehicle acceleration, lane change frequency, following distance, parking time, vehicle spatial distribution and traffic efficiency, determining traffic flow impact evaluation indicators, and constructing a sample data set from the traffic flow impact indicators and the traffic flow impact evaluation indicators; passing the sample data set to a SOM neural network as training data to train the SOM neural network, and using a hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network; and obtaining a traffic flow impact evaluation indicator from the trained SOM neural network based on the traffic flow impact indicators of the road test vehicle. The present invention also discloses a corresponding evaluation device, which can judge the impact of the road test vehicle on traffic.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic control, and in particular to a method and system for evaluating the impact of road testing of traffic flow by an intelligent networked vehicle. Background Art

[0002] With the rapid development and widespread application of intelligent connected vehicle technology, road testing of autonomous driving vehicles has become a core link in promoting technological innovation and market implementation. However, the social impact that may be caused during the road testing of autonomous driving cannot be ignored. These impacts include traffic accidents caused by technical defects such as perception errors and improper decision-making, which in turn cause public doubts about the safety of autonomous driving technology, as well as the actual troubles caused by the concentrated deployment of autonomous driving test vehicles in specific areas to local traffic flow and community life, such as traffic congestion, tight parking resources, and inconvenience in residents' travel, which in turn cause disgust and resistance from community residents. Therefore, it is necessary to properly judge the impact of vehicles on traffic during road testing. Summary of the invention

[0003] In view of the above-mentioned defects of the prior art, the present invention provides a method for evaluating the impact of road test traffic flow of intelligent connected vehicles and a device for evaluating the impact of road test traffic flow of intelligent connected vehicles to judge the impact of road test vehicles on traffic.

[0004] The technical solution of the present invention is as follows: A method for evaluating the impact of road testing of intelligent connected vehicles on traffic flow, comprising the following steps:

[0005] Step S1, collecting the vehicle driving status through the machine vision system and collecting the vehicle position status through the GNSS positioning system;

[0006] Step S2, calculating traffic flow impact indicators according to the vehicle form state and vehicle position state, the traffic flow impact indicators including vehicle speed, vehicle acceleration, lane change frequency, following distance, parking time, vehicle spatial distribution and traffic efficiency, determining traffic flow impact evaluation indicators, and constructing a data set based on the traffic flow impact indicators and traffic flow impact evaluation indicators;

[0007] Step S3, passing the sample data set to the SOM neural network as training data to train the SOM neural network, and using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network;

[0008] Step S4: obtaining a traffic flow impact evaluation index based on the traffic flow impact index of the road test vehicle using the trained SOM neural network.

[0009] Further, the traffic flow impact index is calculated according to the vehicle form state and the vehicle position state, including: obtaining the coordinates of the vehicle on the earth by the GNSS positioning system and determining the displacement s of the vehicle within the time interval t, the vehicle speed v=s / t, the vehicle acceleration a=Δv / Δt, Δv is the vehicle speed change, Δt is the time interval corresponding to the vehicle speed change; collecting the number of lane changes c of the vehicle within the time interval t by the machine vision system, and the lane change frequency p=c / t; obtaining the following distance AA within the time interval t by the machine vision system; according to the vehicle speed measured by the GNSS positioning system is 0, the duration within the time interval t is recorded to obtain the parking time tc; according to the GNSS positioning system, determining whether the vehicle is on an elevated road, on the ground or underground, and determining the vehicle spatial distribution kj; calculating the traffic efficiency E 总 =(E'*E)*100%, E=(T0-c*tt) / T0*100%, E'=α*v+β, E' is the influence of vehicle speed on efficiency, α and β are empirical constants, T0 is the theoretical travel time, and tt is the actual travel time.

[0010] Furthermore, the use of the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network includes calculating the Euclidean distance of each frog individual in the frog population, updating the weights of the accessory neurons with the neuron with the smallest Euclidean distance as the center; calculating the absolute error between the output of the frog individual and the training sample in the sample data set; sorting the individuals based on the absolute error and dividing the sorted population into several sub-populations, iterating the individual position adjustment for each sub-population to obtain the individual with the smallest absolute error in the sub-population; mixing the individuals of each sub-population, and iterating with the individual with the smallest absolute error as the global optimum to obtain the final optimal individual.

[0011] Furthermore, the Euclidean distance calculation formula is:

[0012]

[0013] Among them, g i (t) represents the traffic flow impact index at time t, w il is the weight between the input layer and the mapping layer, and h is the number of neurons in the competition layer.

[0014] Furthermore, the weights are updated using the following formula:

[0015]

[0016] Among them, σ i,l* is the distance between i and l*, and μ is the neighborhood radius.

[0017] Furthermore, the calculation formula of the absolute error is:

[0018]

[0019] Among them, M j is the expected output of the training sample, output j is the output of the jth frog individual, and o is the number of samples in the sample data set.

[0020] Furthermore, the individual position adjustment is performed according to the following formula:

[0021] FF(p+1)=FF(p)+rand*(pbest-pworst)

[0022] Among them, FF is an individual, rand is a random number, pbest is the individual with the smallest absolute error in the current sub-population, and pworst is the individual with the largest absolute error in the current sub-population.

[0023] The present invention also provides a device for evaluating the impact of intelligent networked vehicle road test traffic flow, comprising:

[0024] Collection module: collects vehicle driving status through machine vision system and vehicle position status through GNSS positioning system;

[0025] Sample construction module: calculate traffic flow impact index according to the vehicle form state and vehicle position state, the traffic flow impact index includes vehicle speed, vehicle acceleration, lane change frequency, following distance, parking time, vehicle spatial distribution and traffic efficiency, determine traffic flow impact evaluation index, and construct a sample data set from the traffic flow impact index and traffic flow impact evaluation index;

[0026] Training module: passing the sample data set to the SOM neural network as training data to train the SOM neural network, and using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network; and,

[0027] Evaluation module: The trained SOM neural network obtains the traffic flow impact evaluation index based on the traffic flow impact index of the road test vehicle.

[0028] The present invention also provides a computer storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned intelligent connected vehicle road test traffic flow impact assessment method is implemented.

[0029] Compared with the prior art, the advantages of the technical solution provided by the present invention are:

[0030] The system uses machine vision technology and GNSS positioning to monitor the driving status and position status of autonomous vehicles in real time and accurately. The system uses the leapfrog algorithm to iteratively correct the weight parameters of the SOM neural network, which can effectively obtain excellent neurons suitable for traffic flow data, and then accurately evaluate complex traffic conditions based on machine vision technology and GNSS positioning data. The evaluation results not only provide accurate test feedback for autonomous driving technology research and development companies, helping them to continuously optimize algorithms and improve safety, but also provide scientific basis for regulatory agencies to guide the formulation of more reasonable road test policies and minimize threats to public safety and interference with community life. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The flowchart of the traffic flow impact assessment method for road testing of intelligent connected vehicles is shown in FIG. DETAILED DESCRIPTION

[0032] The present invention is further described below in conjunction with examples. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading this description, various equivalent modifications to this description by those skilled in the art fall within the scope defined by the claims attached to this application.

[0033] The intelligent connected vehicle road test traffic flow impact assessment device involved in this embodiment includes:

[0034] Collection module: collects vehicle driving status through machine vision system and vehicle position status through GNSS positioning system;

[0035] Sample construction module: calculate traffic flow impact index according to the vehicle form state and vehicle position state, the traffic flow impact index includes vehicle speed, vehicle acceleration, lane change frequency, following distance, parking time, vehicle spatial distribution and traffic efficiency, determine traffic flow impact evaluation index, and construct a sample data set from the traffic flow impact index and traffic flow impact evaluation index;

[0036] Training module: passing the sample data set to the SOM neural network as training data to train the SOM neural network, and using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network; and,

[0037] Evaluation module: The trained SOM neural network obtains the traffic flow impact evaluation index based on the traffic flow impact index of the road test vehicle.

[0038] For details, please combine Figure 1 As shown, the intelligent connected vehicle road test traffic flow impact assessment method implemented by the device includes the following steps:

[0039] Step S1, collecting the vehicle driving status through the machine vision system, and collecting the vehicle position status through the GNSS positioning system; wherein the machine vision system collects the video stream of the front of the vehicle and the surrounding environment through the vehicle-mounted camera, and can identify the vehicle driving status (such as speed, acceleration, lane position), driver behavior (such as line of sight direction, operating action), traffic participant information (such as pedestrians, vehicle position, motion status), etc. The present invention mainly takes the vehicle driving status, and adopts a high-precision GNSS receiver to obtain the vehicle real-time position, driving trajectory, speed and other vehicle position status data to ensure accurate positioning and tracking of the vehicle in the road network.

[0040] Step S2: Calculate traffic flow impact indicators according to vehicle form status and vehicle position status. Traffic flow impact indicators include vehicle speed, vehicle acceleration, lane change frequency, following distance, parking time, vehicle spatial distribution and traffic efficiency. Determine traffic flow impact evaluation indicators. Construct a sample data set based on traffic flow impact indicators and traffic flow impact evaluation indicators.

[0041] The specific steps include:

[0042] S201. Obtain the three-dimensional coordinates of the vehicle on the earth through GNSS positioning. The spatial coordinates are the displacement s of the vehicle when the time interval is t seconds. The speed is v=s / t, and the acceleration is a=Δv / Δt. Let g1=v; g2=a.

[0043] S202. Use the machine vision camera Nvidia Orin NX board to identify the number of lane changes c of the vehicle within t seconds, then the lane change frequency p = c / t, and let g3 = p.

[0044] S203, the visual odometer using the monocular depth measurement technology of the machine vision camera Nvidia Orin NX board uses feature matching between consecutive frames and the camera motion model to estimate the depth, and obtains the following vehicle distance g with a time interval of t seconds, and sets g4=g.

[0045] S204, using GNSS positioning to check the vehicle speed, start recording the duration when the vehicle speed is 0, and then obtain the parking time tc when the time interval is t seconds, and set g5=tc.

[0046] S205, using GNSS positioning to check whether the vehicle is on an elevated road, on the ground or underground, and determine the parameter kj, where kj=11 indicates that the vehicle is on an elevated road, kj=22 indicates that the vehicle is on the ground, and kj=33 indicates that the vehicle is underground, and g6=kj.

[0047] S206, calculating the traffic flow efficiency E according to the above vehicle speed, number of lane changes, and travel time tt on a certain section of the road 总 , let g7=E 总 .

[0048] E 总 =(E'*E)*100%, E=(T0-c*tt) / T0*100%, E'=α*v+β

[0049] E' is the effect of vehicle speed on efficiency, α and β are empirical constants, T0 is the theoretical travel time, and c is the number of lane changes.

[0050] S207. Determine the traffic flow impact evaluation index M: There are four traffic flow impact evaluation indexes, among which M=0 means that the road traffic is very smooth, M=1 means that there are a small number of vehicles on the road and the traffic is relatively smooth, M=2 means that the road traffic is heavy and congested, M=3 means that the road traffic is heavy and very congested, and M=4 means that vehicles hardly move forward and special circumstances may occur.

[0051] S208. Through multiple calculations under multiple working conditions and at different times, a sample data set G is obtained. Each sample in the sample data set G includes 7 traffic flow impact indicators g1-g7 and a traffic flow impact evaluation indicator M.

[0052] Step S3: passing the sample data set G to the SOM neural network as training data to train the SOM neural network, and using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network. The specific process of this step includes:

[0053] S301, using a self-organizing neural SOM network to initialize network parameters, and t=1.

[0054] S302, passing the sample data set G as input to the SOM neural network as training data to train the SOM neural network, wherein the sample data set G contains o samples in total.

[0055] S303, initializing the population parameters of the hybrid frog leaping algorithm, including the number of subpopulation iterations P and the number of iterations of the entire population T. Randomly generate a frog population w.

[0056] S304. For each frog individual w l The Euclidean distance is calculated using formula (1), and the lth neuron with the smallest Euclidean distance dist is taken as the best neuron, denoted as l*.

[0057]

[0058] Where: g i (t) represents the traffic flow impact index at time t, w ilis the weight between the input layer and the mapping layer of the SOM neural network. The specific value is provided by the population w according to the order of each frog individual, and h is the number of neurons in the competition layer.

[0059] S305, taking l* as the center, update the weights of nearby neurons. The weight update formula is expressed as:

[0060]

[0061] Where: i,l* is the distance between i and l*, and μ is the neighborhood radius.

[0062] S306, using formula (3) to calculate the j-th frog individual output output j And the expected output M of the training sample j The absolute error between j As an indicator of the accuracy of system traffic flow assessment.

[0063]

[0064] S307, according to Error j The individuals are arranged in order from largest to smallest to form a new population, and then the F population is randomly divided into four sub-populations FF1, FF2, FF3 and FF4.

[0065] S308, let p = 0, and use formula (4) to adjust the individual positions of the FF1 population:

[0066] FF1(p+1)=FF1(p)+rand*(pbest-pworst)(4)

[0067] Where rand is a random number, pbest is the individual with the smallest absolute error in the current sub-population, pworst is the individual with the largest absolute error in the current sub-population, and FF1(p) represents each frog individual in FF1 at p iterations.

[0068] S309, calculate the absolute error of the updated population. If the absolute error is smaller than the previous one, use the new individual obtained by changing formula (4) to replace the original pworst individual, otherwise use the best individual gbest in the entire population to replace pbest, and then set p=p+1 to determine whether p>P is true. If not, return to step S308. If so, execute step S310.

[0069] S310. The worst case is that after P consecutive iterations, no individual with an absolute error smaller than the previous one can be obtained. Then an individual is randomly generated to replace the pworst individual, and finally a new population V1 is obtained.

[0070] S311. Repeat steps S308 to S310 to iteratively calculate population FF2 to obtain a new population V2.

[0071] S312. Repeat steps S308 to S310 to iteratively calculate population FF3 to obtain a new population V3.

[0072] S313. Repeat steps S308 to S310 to perform iterative calculations on population FF4 to obtain a new population V4.

[0073] S314. Combine the populations V1, V2, V3, and V4 updated in steps S310 to S313 to form a new population VV.

[0074] S315, calculate the absolute error for each frog individual in the population VV. Update gbest with the individual with the smallest absolute error, and t=t+1, and determine whether t>T is established. If not, return to step S304. If so, the obtained gbest value is the weight w in the SOM neural network. opt .

[0075] Step S4: obtaining a traffic flow impact evaluation index based on the traffic flow impact index of the road test vehicle using the trained SOM neural network.

[0076] It should be noted that the specific methods of the above embodiments can form a computer program product. Therefore, the computer program product implemented in this application can be stored on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.).

Claims

1. A method for evaluating the impact of intelligent connected vehicle road testing on traffic flow, comprising the following steps: Step S1, collecting the vehicle driving status through the machine vision system and collecting the vehicle position status through the GNSS positioning system; Step S2, calculating traffic flow impact indicators according to the vehicle driving state and vehicle position state, the traffic flow impact indicators including vehicle speed, vehicle acceleration, lane change frequency, following distance, parking time, vehicle spatial distribution and traffic efficiency, determining traffic flow impact evaluation indicators, and constructing a sample data set based on the traffic flow impact indicators and traffic flow impact evaluation indicators; Step S3, passing the sample data set to the SOM neural network as training data to train the SOM neural network, and using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network; Step S4, obtaining a traffic flow impact evaluation index based on the traffic flow impact index of the road test vehicle by the trained SOM neural network; The method of using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network includes calculating the Euclidean distance of each frog individual in the frog population, updating the weights of nearby neurons with the neuron with the smallest Euclidean distance as the center; calculating the absolute error between the output of the frog individual and the training sample in the sample data set; sorting the individuals based on the absolute error and dividing the sorted population into a plurality of sub-populations, iterating the individual position adjustment for each sub-population to obtain the individual with the smallest absolute error in the sub-population; mixing the individuals of each sub-population, and iterating the individual with the smallest absolute error as the global optimum to obtain the final optimal individual; The Euclidean distance calculation formula is: Among them, g i (t) represents the traffic flow impact index at time t, w il is the weight between the input layer and the mapping layer, and h is the number of neurons in the competition layer; The weights are updated by the following formula: Among them, σ i,l* is the distance between i and l*, μ is the neighborhood radius; The calculation formula of the absolute error is: Among them, M j is the expected output of the training sample, output j is the output of the jth frog individual, and o is the number of samples in the sample data set; The individual position adjustment is performed according to the following formula FF(p+1)=FF(p)+rand*(pbest-pworst) Among them, FF is the individual, p represents the number of iterations, rand is a random number, pbest is the individual with the smallest absolute error in the current sub-population, and pworst is the individual with the largest absolute error in the current sub-population.

2. The method for evaluating the impact of intelligent connected vehicle road test on traffic flow according to claim 1, characterized in that: Calculating the traffic flow impact index according to the vehicle driving state and vehicle position state includes: obtaining the coordinates of the vehicle on the earth by the GNSS positioning system and determining the displacement s of the vehicle within the time interval t, the vehicle speed v=s / t, the vehicle acceleration a=Δv / Δt, Δv is the vehicle speed change, Δt is the time interval corresponding to the vehicle speed change; collecting the number of lane changes c of the vehicle within the time interval t by the machine vision system, and the lane change frequency p=c / t; obtaining the following distance within the time interval t by the machine vision system; according to the vehicle speed measured by the GNSS positioning system is 0, recording the duration within the time interval t to obtain the parking time tc; determining whether the vehicle is on an elevated road, on the ground or underground according to the GNSS positioning system, and determining the vehicle spatial distribution kj; calculating the traffic efficiency E 总 =(E'*E)*100%, E=(T0-c*tt) / T0*100%, E'=α*v+β, E' is the influence of vehicle speed on efficiency, α and β are empirical constants, T0 is the theoretical travel time, and tt is the actual travel time.

3. A device for evaluating the impact of road traffic flow on intelligent connected vehicles, characterized in that: include: Collection module: collects vehicle driving status through machine vision system and vehicle position status through GNSS positioning system; Sample construction module: calculate traffic flow impact index according to the vehicle driving state and vehicle position state, the traffic flow impact index includes vehicle speed, vehicle acceleration, lane change frequency, following distance, parking time, vehicle spatial distribution and traffic efficiency, determine traffic flow impact evaluation index, and construct a sample data set from the traffic flow impact index and traffic flow impact evaluation index; Training module: passing the sample data set to the SOM neural network as training data to train the SOM neural network, and using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network; and, Evaluation module: The trained SOM neural network obtains the traffic flow impact evaluation index based on the traffic flow impact index of the road test vehicle; The method of using the hybrid frog leaping algorithm to optimize the weights between the input layer and the mapping layer in the SOM neural network includes calculating the Euclidean distance of each frog individual in the frog population, updating the weights of nearby neurons with the neuron with the smallest Euclidean distance as the center; calculating the absolute error between the output of the frog individual and the training sample in the sample data set; sorting the individuals based on the absolute error and dividing the sorted population into a plurality of sub-populations, iterating the individual position adjustment for each sub-population to obtain the individual with the smallest absolute error in the sub-population; mixing the individuals of each sub-population, and iterating the individual with the smallest absolute error as the global optimum to obtain the final optimal individual; The Euclidean distance calculation formula is: Among them, g i (t) represents the traffic flow impact index at time t, w il is the weight between the input layer and the mapping layer, and h is the number of neurons in the competition layer; The weights are updated by the following formula: Among them, σ i,l* is the distance between i and l*, μ is the neighborhood radius; The calculation formula of the absolute error is: Among them, M j is the expected output of the training sample, output j is the output of the jth frog individual, and o is the number of samples in the sample data set; The individual position adjustment is performed according to the following formula FF(p+1)=FF(p)+rand*(pbest-pworst) Among them, FF is the individual, p represents the number of iterations, rand is a random number, pbest is the individual with the smallest absolute error in the current sub-population, and pworst is the individual with the largest absolute error in the current sub-population.

4. A computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for evaluating the impact of intelligent connected vehicle road testing on traffic flow as claimed in claim 1 or 2 is implemented.

Images