Lane traffic state estimation method and device based on network-connected vehicle, terminal and medium

Abstract

The invention provides a lane traffic state estimation method and device based on networked vehicles, a terminal and a medium, and belongs to the technical field of intelligent traffic, and the method comprises the steps: constructing a space-time grid of a target lane, and setting a sensing range for each networked vehicle; obtaining driving data of each network-connected vehicle report on the target lane; based on the driving data and the space-time grid, second driving data with a data state mark are obtained, and the data state comprises validity and missing; inputting the second driving data and a preset noise vector into a trained generator of a generative adversarial model, and outputting complemented third driving data after processing; and estimating a lane traffic state based on the third driving data. According to the invention, the driving data is acquired through the network connection vehicle, the data state mark is introduced into the driving data, and then the trained generator is used for complementing the data with the data state mark being missing to estimate the road condition, so that the accuracy of traffic state estimation under the low network connection vehicle permeability can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation, and particularly to a method, device, terminal, and medium for estimating lane traffic status based on connected vehicles. Background Art

[0002] Accurately estimating traffic status is the key to optimizing traffic management strategies, improving lane passing efficiency, and ensuring driving safety. Among the existing lane traffic status estimation methods, the most commonly used is to rely on fixed detectors, such as installing geomagnetic sensors, cameras, and other devices at specific positions beside the lane to collect traffic information and estimate the lane traffic status based on this information. However, due to the limited number of fixed detectors, it is difficult to collect data comprehensively and in detail.

[0003] With the development of V2X (Vehicle-to-Everything) communication technology, connected vehicles can be used as floating vehicles or mobile sensors to report their positions, speeds, and other relevant information at a preset frequency, providing a new idea for lane traffic status estimation. However, at present, most of the vehicles on the lane are non-connected vehicles, and the penetration rate of connected vehicles is still relatively low. This situation will cause data sparsity and incompleteness problems in traffic status estimation, resulting in low accuracy of the traffic status estimation results obtained based on this method.

[0004] Therefore, there are defects in the prior art and it needs to be improved and developed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, device, terminal, and medium for estimating lane traffic status based on connected vehicles, aiming to solve the problem of low accuracy of traffic status estimation results in the prior art.

[0006] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0007] In a first aspect, an embodiment of the present invention provides a method for estimating lane traffic status based on connected vehicles, the method comprising:

[0008] Constructing a spatio-temporal grid of a target lane and setting a sensing range of each connected vehicle on the target lane in the spatio-temporal grid;

[0009] Obtaining driving data reported by each connected vehicle on the target lane, the driving data including the own driving data of the connected vehicle and / or the driving data of non-connected vehicles collected by the connected vehicle within its sensing range;

[0010] Based on the driving data and the spatio-temporal grid, obtaining second driving data with data status marks, the data status including valid and missing;

[0011] Input the second driving data and a preset noise vector into the generator of the trained generative adversarial model, and output the completed third driving data after processing;

[0012] Estimate the lane traffic state based on the third driving data.

[0013] In one implementation, constructing a spatio-temporal grid of a target lane includes:

[0014] Obtain the lane geometry data of the target lane;

[0015] Generate a spatio-temporal grid of the target lane based on the lane geometry data and preset time resolution and space resolution.

[0016] In one implementation, obtaining the second driving data with data status tags based on the driving data and the spatio-temporal grid includes:

[0017] After performing data preprocessing on the driving data, obtain the preprocessed driving data;

[0018] According to the coverage status of each preprocessed driving data in the spatio-temporal grid, generate a mask matrix marking the data status of each grid;

[0019] Based on the mask matrix and all preprocessed driving data, obtain the second driving data with data status tags.

[0020] In one implementation, generating a mask matrix marking the data status of each grid according to the coverage status of each preprocessed driving data in the spatio-temporal grid includes:

[0021] Map each preprocessed driving data to the corresponding grid in the spatio-temporal grid;

[0022] For each grid, determine whether it is covered by the preprocessed driving data, and determine the data status of the grid according to the judgment result;

[0023] Generate a mask matrix based on the data status of all grids.

[0024] In one implementation, for each grid, determining whether it is covered by the preprocessed driving data and determining the data status of the grid according to the judgment result includes:

[0025] For each grid, determine whether there is the own driving data of a connected vehicle. If so, determine the data status of the grid as valid;

[0026] If there is no own driving data of a connected vehicle in the grid, then determine whether the grid is within the perception range of any connected vehicle;

[0027] If the grid is not within the perception range of any connected vehicle, determine that the data status of the grid is missing;

[0028] If the grid is within the perception range of any connected vehicle, determine its status as valid or missing according to the existence of non-connected vehicle driving data.

[0029] In one implementation, the training steps of the generative adversarial model include:

[0030] Obtain a training data set, which includes training subsets corresponding to multiple preset connected vehicle penetration rates. Each training subset contains training driving data with training data status labels, and reference driving data corresponding to each connected vehicle penetration rate. The reference driving data comes from the driving data collected by the traffic detection platform and connected vehicles. The training data status includes valid and missing;

[0031] Initialize the generative adversarial model, which includes a generator and a discriminator;

[0032] For each connected vehicle penetration rate, train the generator and the discriminator according to a preset number of rounds;

[0033] When the training for all connected vehicle penetration rates is completed, evaluate the data completion effect of the generative adversarial model. When passing the evaluation, obtain the trained generative adversarial model.

[0034] In one implementation, for each connected vehicle penetration rate, training the generator and the discriminator according to a preset number of rounds includes:

[0035] Fix the discriminator parameters, train the generator for the first preset number of rounds. In each round, input the noise training vector and the training driving data with training data status labels under the current connected vehicle penetration rate into the generator, complete the training driving data with the training data status of missing, and output the completed driving data;

[0036] Input the completed driving data and the reference driving data with the same connected vehicle penetration rate into the discriminator, calculate the generator loss and backpropagate to update the generator parameters;

[0037] Fix the generator parameters, train the discriminator for the second preset number of rounds, input the completed driving data and the reference driving data of the current connected vehicle penetration rate into the discriminator, calculate the discriminator loss and backpropagate to update the discriminator parameters.

[0038] In a second aspect, an embodiment of the present invention further provides a device for estimating the lane traffic state based on connected vehicles, including:

[0039] A grid construction module, configured to construct a spatio-temporal grid of the target lane and set the perception range of each connected vehicle on the target lane in the spatio-temporal grid;

[0040] A data acquisition module, configured to acquire the driving data reported by each connected vehicle on the target lane, where the driving data includes the own driving data of the connected vehicle and / or the driving data of non-connected vehicles collected by the connected vehicle within its sensing range;

[0041] A second driving data generation module, configured to obtain second driving data with data status tags based on the driving data and the spatio-temporal grid, where the data status includes valid and missing;

[0042] A data completion module, configured to input the second driving data and a preset noise vector into the generator of a trained generative adversarial model, and output, after processing, the completed third driving data;

[0043] A state estimation module, configured to estimate the lane traffic state based on the third driving data.

[0044] In a third aspect, an embodiment of the present invention further provides a terminal, where the terminal includes: a memory, a processor, and a connected-vehicle-based lane traffic state estimation program stored on the memory and executable on the processor. When the connected-vehicle-based lane traffic state estimation program is executed by the processor, the steps of the above-mentioned connected-vehicle-based lane traffic state estimation method are implemented.

[0045] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores a connected-vehicle-based lane traffic state estimation program, and the connected-vehicle-based lane traffic state estimation program can be executed to implement the steps of the above-mentioned connected-vehicle-based lane traffic state estimation method.

[0046] Advantages of the present invention: By constructing a spatio-temporal grid of a target lane and setting a sensing range for each connected vehicle; acquiring the driving data reported by each connected vehicle on the target lane; obtaining second driving data with data status tags based on the driving data and the spatio-temporal grid, where the data status includes valid and missing; inputting the second driving data and a preset noise vector into the generator of a trained generative adversarial model, and outputting, after processing, the completed third driving data; and estimating the lane traffic state based on the third driving data. The present invention collects driving data through connected vehicles, introduces data status tags thereto, and then uses a trained generator to complete the data with the data status tag of missing to estimate the road conditions, which can effectively improve the accuracy of traffic state estimation under low connected-vehicle penetration rate. Description of the Drawings

[0047] Figure 1 is a flowchart of a preferred embodiment of the connected-vehicle-based lane traffic state estimation method in the present invention.

[0048] Figure 2 It is a schematic diagram of the generation process of the mask matrix in the present invention.

[0049] Figure 3 It is a schematic diagram of the data processing process of the generator in the present invention.

[0050] Figure 4 It is a schematic diagram of the generation process of the final mask training matrix in the present invention.

[0051] Figure 5 It is a schematic diagram of the data processing process of the discriminator in the present invention.

[0052] Figure 6 It is a schematic diagram of the process of jointly processing data by the discriminator and the generator in the present invention.

[0053] Figure 7 It is a schematic diagram of the structure of a preferred embodiment of the lane traffic state estimation method based on connected vehicles in the present invention.

[0054] Figure 8 It is a schematic block diagram of the terminal of the present invention. Detailed implementation manners

[0055] To make the objectives, technical solutions and advantages of the present invention clearer and more definite, the following further describes the present invention in detail with reference to the accompanying drawings and by way of examples. It should be understood that the specific examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0056] Accurately estimating the traffic state is the key to optimizing traffic management strategies, improving lane passing efficiency, and ensuring driving safety. Among the existing lane traffic state estimation methods, the most commonly used is to use fixed detectors, install geomagnetic sensors, cameras and other devices at specific positions beside the lane to collect traffic information and estimate the lane traffic state based on this information. However, due to the limited number of fixed detectors, it is difficult to collect data comprehensively and in detail.

[0057] With the development of V2X (i.e., vehicle-to-everything) communication technology, connected vehicles can be used as floating vehicles or mobile sensors to report their positions, speeds and other relevant information at a preset frequency, providing a new idea for lane traffic state estimation. However, at present, most of the lane vehicles are non-connected vehicles, and the penetration rate of connected vehicles is still relatively low. This situation will cause data sparsity and incompleteness problems in traffic state estimation, resulting in low accuracy of traffic state estimation results obtained based on this method.

[0058] In view of the above defects of the prior art, the present invention provides a method, device, terminal and medium for estimating lane traffic status based on connected vehicles, belonging to the field of intelligent transportation technology. The method includes: constructing a spatio-temporal grid for a target lane and setting a sensing range for each connected vehicle; obtaining the driving data reported by each connected vehicle on the target lane; based on the driving data and the spatio-temporal grid, obtaining second driving data with data status markers, where the data status includes valid and missing; inputting the second driving data and a preset noise vector into the generator of a trained generative adversarial model, and outputting, after processing, third driving data with missing data filled; estimating the lane traffic status based on the third driving data. By collecting driving data through connected vehicles, introducing data status markers, and then using the trained generator to fill in the data marked as missing to estimate the road conditions, the present invention can effectively improve the accuracy of traffic status estimation under low connected vehicle penetration rates.

[0059] Please refer to Figure 1 , the method for estimating lane traffic status based on connected vehicles according to the embodiment of the present invention includes the following steps:

[0060] Step S100, construct a spatio-temporal grid for the target lane and set the sensing range of each connected vehicle on the target lane in the spatio-temporal grid.

[0061] Specifically, the spatio-temporal grid is a form of grid data that combines space and time. After obtaining the lane geometric data of the target lane, based on the lane geometric data and preset time resolution and spatial resolution, the spatio-temporal grid of the target lane can be generated. The lane geometric data is the spatial position data of the lane, including parameters such as the shape, width, length, and curvature of the lane. These data come from third-party map data. The time resolution refers to the accuracy of the spatio-temporal grid in the time dimension, that is, the frequency of grid data update represented by each time unit (such as seconds, minutes, hours, etc.). The spatial resolution refers to the accuracy of the spatio-temporal grid in the spatial dimension, that is, the actual spatial range represented by each grid cell (such as meters, kilometers, etc.).

[0062] Connected vehicles, also known as intelligent connected vehicles or vehicle-to-everything (V2X) vehicles, include sensors such as speed sensors, GPS positioning sensors, acceleration sensors, lidar, cameras, etc., and communication modules. These sensors and modules enable connected vehicles to obtain information such as their own speed, position, and acceleration in real time, and exchange information with other vehicles or traffic infrastructure through the communication module. Connected vehicles can use these sensors to sense the presence of other non-connected vehicles and collect the driving data of other connected vehicles. In the spatio-temporal grid, with the connected vehicle as the center, a sensing distance of the connected vehicle is set at a preset distance. The preset distance can be set by factors such as the sensor performance of the connected vehicle and the traffic complexity of the target lane.

[0063] Please refer to Figure 1 , the lane traffic state estimation method based on connected vehicles according to the embodiments of the present invention includes the following steps:

[0064] Step S200: Obtain the driving data reported by each connected vehicle on the target lane, where the driving data includes the own driving data of the connected vehicle and / or the driving data of non-connected vehicles collected by the connected vehicle within its sensing range.

[0065] Specifically, the connected vehicle can report its own driving data by using its own sensors and can collect the driving data of non-connected vehicles (i.e., non-connected vehicle driving data) within its sensing range. The driving data includes vehicle speed, vehicle position, acceleration, driving direction, and timestamp, etc.

[0066] Please refer to Figure 1 , the lane traffic state estimation method based on connected vehicles according to the embodiments of the present invention further includes the following steps:

[0067] Step S300: Based on the driving data and the spatio-temporal grid, obtain the second driving data with data status marks, where the data status includes valid and missing.

[0068] Specifically, according to the driving data and the spatio-temporal grid, the driving data indicating data validity or missing can be obtained. Subsequently, this second driving data is input into the generator of the trained generative adversarial model for completion processing, and the completed driving data can be obtained. In this way, the problem of sparse traffic data collected by connected vehicles under low penetration rates can be solved.

[0069] In one implementation, based on the driving data and the spatio-temporal grid, obtaining the second driving data with data status marks includes:

[0070] After performing data preprocessing on the driving data, obtain the preprocessed driving data;

[0071] According to the coverage status of each preprocessed driving data in the spatio-temporal grid, generate a mask matrix for marking the data status of each grid;

[0072] Based on the mask matrix and all preprocessed driving data, obtain the second driving data with data status marks.

[0073] Specifically, after obtaining the driving data, preprocessing is first performed. The preprocessing includes data cleaning and format unification. The data processing uses an outlier detection algorithm based on statistical analysis and traffic rules to remove obviously incorrect or abnormal data, such as data with speed exceeding the reasonable range or position information not conforming to the lane topology structure. Format unification is to unify the data collected by different sensors into a standard format for subsequent processing. The mask matrix is a matrix with element values of 0 and 1, used to mark the data status of each grid. When the element value of the mask matrix is 0, it indicates that the data status of the grid is missing. When the element value of the mask matrix is 1, it indicates that the data status of the grid is valid. The mask matrix is used to simulate the data missing situation under different connected vehicle penetration rates to reflect the data sparsity caused by the limited number of connected vehicles in the actual lane.

[0074] In one implementation, according to the coverage status of each preprocessed driving data in the spatio-temporal grid, a mask matrix for marking the data status of each grid is generated, including:

[0075] Map each preprocessed driving data to the corresponding grid in the spatio-temporal grid;

[0076] For each grid, determine whether it is covered by the preprocessed driving data, and determine the data status of the grid according to the judgment result;

[0077] Generate a mask matrix based on the data status of all grids.

[0078] Specifically, the present invention converts discrete driving data into a continuous data representation in the spatio-temporal grid coordinate system, and uses a mask matrix to mark which grids have valid driving data and which grids have missing data. Subsequently, a second driving data with data status marks is generated using the mask matrix and the preprocessed driving data, which is convenient for the generator of the adversarial network model to complete.

[0079] In one implementation, for each grid, determine whether it is covered by the preprocessed driving data, and determine the data status of the grid according to the judgment result, including:

[0080] For each grid, determine whether there is its own driving data of a connected vehicle. If it exists, determine that the data status of the grid is valid;

[0081] If there is no its own driving data of a connected vehicle in the grid, then determine whether the grid is within the perception range of any connected vehicle;

[0082] If the grid is not within the perception range of any connected vehicle, determine that the data status of the grid is missing;

[0083] If the grid is within the sensing range of any connected vehicle, its status is determined to be valid or missing based on the existence of non-connected vehicle driving data.

[0084] Specifically, if the grid is within the sensing range of any connected vehicle, it is then further determined whether there is non-connected vehicle driving data. If there is non-connected vehicle driving data, its status is determined to be valid; if there is no non-connected vehicle driving data, its status is determined to be missing. A schematic diagram of the generation process of the mask matrix is as Figure 2 shown.

[0085] Please refer to Figure 1 , the method for estimating lane traffic status based on connected vehicles according to the embodiments of the present invention includes the following steps:

[0086] Step S400: Input the second driving data and a preset noise vector into the generator of the trained generative adversarial model, and output the completed third driving data after processing.

[0087] Specifically, the generative adversarial model includes a generator and a discriminator. Figure 3 It is a schematic diagram of the data processing flow of the generator. The generator includes an input layer, a flattening layer, a fully connected layer, an output layer, and a reshaping layer connected in sequence. When the second driving data and a preset noise vector are input into the input layer, data splicing is performed in the last dimension (i.e., the number of features). The noise vector can be set according to the complexity of the lane-level traffic speed data and the requirements for generation diversity. Then the flattening layer processes the spliced data into a one-dimensional vector for subsequent fully connected layer operations. Next, the data passes through multiple fully connected layers to extract global features, with the number of neurons decreasing, and the relu activation function is used to enhance the non-linear expression ability of the model. Finally, after passing through the output layer, it is activated by the linear activation function, and then restored to the original shape by the reshaping layer (i.e., the reshape layer).

[0088] In one implementation, the training steps of the generative adversarial model include:

[0089] Obtain a training data set, where the training data set includes training subsets corresponding to multiple preset connected vehicle penetration rates. Each training subset contains training driving data with training data status labels, and reference driving data corresponding to each connected vehicle penetration rate. The reference driving data comes from the driving data collected by the traffic detection platform and connected vehicles. The training data status includes valid and missing;

[0090] Initialize the generative adversarial model, which includes a generator and a discriminator;

[0091] For each connected vehicle penetration rate, train the generator and the discriminator according to a preset number of rounds;

[0092] When the training of the penetration rate of all connected vehicles is completed, evaluate the data completion effect of the generative adversarial model. After passing the evaluation, obtain the trained generative adversarial model.

[0093] Specifically, obtain the driving data set collected by the traffic monitoring platform and the target training connected vehicles for the training lane within a preset time range. Then, screen and simulate the driving data set to generate benchmark driving data corresponding to several connected vehicle penetration rates.

[0094] The generation process of the training driving data with training data status markers includes: dividing the corresponding training spatio-temporal grid for the target training lane, and setting the perception range of each training connected vehicle on the target lane in the training spatio-temporal grid; obtaining the training driving data reported by each training connected vehicle on the target training lane, where the training driving data includes the own training driving data of the training connected vehicle and / or the non-connected vehicle training driving data collected by the training connected vehicle within its perception range; based on the coverage status of the training driving data in the training spatio-temporal grid, obtain an initial mask matrix, where the initial mask matrix is a matrix composed of 0 and 1, where 0 represents that the data of the grid is missing and 1 represents that the data of the grid is valid; map the benchmark driving data and the training driving data corresponding to the connected vehicle penetration rate to the training spatio-temporal grid respectively to obtain a first traffic state matrix and a second traffic state matrix; subtract the second traffic state matrix from the first traffic state matrix to obtain a difference matrix. At this time, the difference matrix and the mask training matrix have the same dimension and both correspond to the same training spatio-temporal grid; if there are elements with a value of 0 in the difference matrix, it means that there is no difference between the benchmark driving data and the training driving data. At this time, maintain the element value at the corresponding position of the mask training matrix as 1, indicating that the data is valid; if there are elements with a non-zero value in the difference matrix, it means that there is a difference between the benchmark driving data and the training driving data. At this time, update the element value at the corresponding position of the mask training matrix to 0, indicating that the data is missing. After judging the element values of all difference matrices and updating or maintaining the element values of the mask training matrix, obtain the final mask training matrix; based on the final mask training matrix and the driving training data at the connected vehicle penetration rate, obtain the training driving data with training data status markers. The generated final mask training matrix accurately simulates the data sparsity situation at low penetration rates and provides real data samples for subsequent model training.

[0095] In one implementation, obtaining the initial mask matrix based on the coverage status of the training driving data in the training spatio-temporal grid includes:

[0096] Map the training driving data to the corresponding grid of the training spatio-temporal grid;

[0097] For each grid, determine whether it is covered by training driving data, and determine the data status of the grid according to the judgment result;

[0098] Generate an initial mask matrix based on the data status of all grids.

[0099] Specifically, for each grid, determine whether there is the vehicle's own training driving data of the connected vehicle. If it exists, determine the data status of the grid as valid; if there is no vehicle's own training driving data of the connected vehicle in the grid, then determine whether the grid is within the perception range of any connected vehicle; if the grid is not within the perception range of any connected vehicle, determine the data status of the grid as missing; if the grid is within the perception range of any connected vehicle, determine its status as valid or missing according to whether there is non-connected vehicle driving training data. When there is non-connected vehicle driving training data, determine its status as valid, and if there is no non-connected vehicle driving training data, determine its status as invalid. The schematic diagram of the generation process of the final mask training matrix is as Figure 4 shown.

[0100] In one implementation, divide the training spatio-temporal grids corresponding to the target training lane, including:

[0101] Obtain the lane geometric data of the target training lane;

[0102] Generate the training spatio-temporal grids of the target training lane based on the lane geometric data of the target training lane and the preset second time resolution and second spatial resolution.

[0103] For each connected vehicle penetration rate, train the generator and discriminator according to the preset number of rounds, including:

[0104] Fix the discriminator parameters and train the generator for the first preset number of rounds. In each round, input the noise training vector and the training driving data with the training data status label under the current connected vehicle penetration rate into the generator to complete the training driving data with the missing training data status, and output the completed driving data;

[0105] Input the completed driving data and the reference driving data with the same connected vehicle penetration rate into the discriminator, calculate the generator loss and backpropagate to update the generator parameters;

[0106] Fix the generator parameters and train the discriminator for the second preset number of rounds. Input the completed driving data and the reference driving data of the current connected vehicle penetration rate into the discriminator, calculate the discriminator loss and backpropagate to update the discriminator parameters.

[0107] Specifically, the discriminator is constructed based on a convolutional neural network, and the schematic diagram of its data processing flow is as Figure 5As shown in the figure. First, the input training driving data with training data status tags and the benchmark driving data under the penetration rate of this connected vehicle are converted into a three-dimensional form through a reshaping layer (i.e., reshape layer) to adapt to subsequent convolutional layer operations. Then, the data passes through multiple convolutional layers to extract deep features of the data. The padding method is same, ensuring that the output size after the convolutional operation is the same as the input. At the same time, the relu activation function is used to introduce non-linear features. Through these multiple convolutional layers, the discriminator can automatically learn the local features of the data in space, so as to better judge the authenticity of the data. After being processed by the convolutional layers, the data is processed into a one-dimensional vector by a flattening layer, and then the discriminant result is output through a fully connected layer (1 neuron, with the sigmoid activation function). The output range of the sigmoid function is between 0 and 1. Close to 1 indicates that the discriminator believes that the input data is real data, and close to 0 indicates that it is considered to be generated data. The discriminator extracts features and makes judgments on the benchmark driving data and the completed driving data output by the generator, providing feedback to the generator, prompting the generator to continuously optimize the generation strategy, improve the quality of the generated data, make the generated data more difficult to be distinguished by the discriminator, and finally achieve that the data generated by the generator is similar to the real data distribution. The schematic diagram of the process of the discriminator and the generator jointly processing data is as shown in Figure 6 As shown in the figure. The original data is the set of driving data collected from the traffic monitoring platform and the target training connected vehicle for the training lane within a preset time range. The real samples are the benchmark driving data corresponding to each connected vehicle penetration rate. The false samples output by the generator are the completed driving data corresponding to each connected vehicle penetration rate. The completed driving data and the benchmark driving data corresponding to each connected vehicle penetration rate are jointly input into the discriminator for judgment. Through continuous training, the loss function converges.

[0108] During the training process, the Adam optimizer and the binary cross-entropy loss function are used to iteratively optimize the GAN model.

[0109] The noise training vector is obtained by sampling from a normal distribution. The calculation formula of the binary cross-entropy loss function is as follows: where L represents the result of the binary cross-entropy loss function; N represents the number of samples; y i represents the true label of sample i, and its value is 0 or 1. In the discriminator training scenario, for real data samples (i.e., benchmark driving data), y i =1, and for generated data samples (i.e., completed driving data), y i =0; p i represents the probability that the model predicts sample i as a real data sample, and its value range is between [0,1].

[0110] After being trained, the generative adversarial model can accurately learn the distribution of traffic data, complete the input traffic data with missing values, and output the completed traffic data for subsequent evaluation of the completion effect and practical applications, providing reliable data support for traffic state estimation.

[0111] The steps for evaluating the generative adversarial model are as follows: From the driving data set collected by the traffic monitoring platform and the target training connected vehicles for the training lane within a preset time range, extract the missing reference data corresponding to the missing part in the final mask training matrix as the reference for evaluating the completion effect; in the completed driving data output by the generative adversarial model, screen out the missing completion data corresponding to the missing part according to the final mask matrix; calculate the root mean square error and mean absolute error between the missing reference data and the missing completion data. The calculation formula for the root mean square error is: The calculation formula for the mean absolute error is: where n is the number of missing data points, y i is the value of the missing reference data, is the value of the missing completion data.

[0112] The root mean square error (RMSE) is an indicator reflecting the overall deviation of the completed data from the original true data. The smaller the root mean square error, the more stable the model; the mean absolute error (MAE) reflects the absolute deviation between the model's completed data and the original missing data. The smaller the mean absolute error, the higher the prediction accuracy of the model. By calculating the root mean square error and mean absolute error under different penetration rates, comprehensively evaluate the completion performance of the model under different data sparsity levels, providing strong data support for the optimization of the model and traffic state estimation.

[0113] When both of these two indicators reach the corresponding preset thresholds, it is determined that the evaluation is passed, and the trained generative adversarial model is obtained.

[0114] Please refer to Figure 1 , the lane traffic state estimation method based on connected vehicles described in the embodiments of the present invention includes the following steps:

[0115] Step S500, estimate the lane traffic state based on the third driving data.

[0116] Specifically, the lane traffic state can be estimated based on the supplemented third driving data. The third driving data is the driving data including connected vehicles and non-connected vehicles. Since the data has been completed, the data is more complete. The third driving data can be combined with other relevant traffic information, such as road grade, traffic signal status, historical traffic flow data, and current weather conditions, etc., to comprehensively judge the lane-level traffic state.

[0117] In one implementation, based on the vehicle speed in the third driving data, combined with the road grade, traffic signal status, historical traffic flow data, and current weather conditions, the traffic state at the lane level is comprehensively judged.

[0118] Specifically, when the complemented vehicle speed has been lower than the preset speed threshold within a preset time period, and combined with the red light duration of the traffic signal and historical congestion data, it is determined that the lane is in a congested state; if the complemented vehicle speed is within the preset vehicle speed range and has little fluctuation, and at the same time the historical data flow of the lane is unobstructed during the same time period, and the current weather condition is good, then it is determined that the traffic condition is relatively smooth. In this way, the accurate estimation of the lane-level traffic state of connected vehicles under low penetration rate is realized, providing a scientific traffic guidance strategy for traffic management departments, planning reasonable travel routes for travelers, and providing a reliable decision-making basis for intelligent driving systems.

[0119] In summary, the present invention constructs a spatio-temporal grid of the target lane and sets a sensing range for each connected vehicle; obtains the driving data reported by each connected vehicle on the target lane; based on the driving data and the spatio-temporal grid, obtains the second driving data with data status markers, and the data status includes valid and missing; inputs the second driving data and a preset noise vector into the generator of the trained generative adversarial model, and outputs the complemented third driving data after processing; estimates the lane traffic state based on the third driving data. The present invention collects driving data through connected vehicles, introduces data status markers, and then uses the trained generator to complement the data with data status marked as missing to estimate the road conditions, which can effectively improve the accuracy of traffic state estimation under low connected vehicle penetration rate.

[0120] In one embodiment, as Figure 7 shown, based on the above-mentioned lane traffic state estimation method based on connected vehicles, the present invention also correspondingly provides a lane traffic state estimation device based on connected vehicles, including:

[0121] A grid construction module 100, configured to construct a spatio-temporal grid of the target lane and set the sensing range of each connected vehicle on the target lane in the spatio-temporal grid;

[0122] A data acquisition module 200, configured to acquire the driving data reported by each connected vehicle on the target lane, where the driving data includes the own driving data of the connected vehicle and / or the driving data of non-connected vehicles collected by the connected vehicle within its sensing range;

[0123] A second driving data generation module 300, configured to obtain the second driving data with data status markers based on the driving data and the spatio-temporal grid, where the data status includes valid and missing;

[0124] A data completion module 400 is configured to input the second driving data and a preset noise vector into a generator of a trained generative adversarial model, and output, after processing, the completed third driving data.

[0125] A state estimation module 500 is configured to estimate the lane traffic state based on the third driving data.

[0126] It should be noted that the foregoing explanations of the embodiments of the method for estimating the lane traffic state based on a connected vehicle also apply to the device for estimating the lane traffic state based on a connected vehicle in this embodiment, and details are not described herein again.

[0127] Based on the above embodiments, the present invention further provides a terminal, and a schematic block diagram thereof may be as Figure 8 shown. The above terminal includes a processor, a memory, a network interface, and a display screen connected through a device bus. Among them, the processor of the terminal is configured to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating device and a program for estimating the lane traffic state based on a connected vehicle. The internal memory provides an environment for the operation of the operating device and the program for estimating the lane traffic state based on a connected vehicle in the non-volatile storage medium. The network interface of the terminal is configured to communicate with an external terminal through a network connection. When the program for estimating the lane traffic state based on a connected vehicle is executed by the processor, the steps of any one of the methods for estimating the lane traffic state based on a connected vehicle described above are implemented. The display screen of the terminal may be a liquid crystal display screen or an electronic ink display screen.

[0128] Those skilled in the art can understand that Figure 8 the schematic block diagram shown in

[0129] is only a block diagram of some structures related to the solution of the present invention, and does not constitute a limitation on the terminal to which the solution of the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0130] In one embodiment, a terminal is provided. The above terminal includes a memory, a processor, and a program for estimating the lane traffic state based on a connected vehicle stored on the above memory and executable on the above processor. When the program for estimating the lane traffic state based on a connected vehicle is executed by the above processor, the steps of any one of the methods for estimating the lane traffic state based on a connected vehicle provided by the embodiments of the present invention are implemented.

[0131] It should be understood that the sequence numbers of the steps in the above embodiments do not imply the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0132] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional unit and module is used as an example. In practical applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the above device can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present invention. The specific working process of the units and modules in the above device can refer to the corresponding process in the foregoing method embodiments and will not be elaborated herein.

[0133] In the above embodiments, each embodiment is described with emphasis. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0134] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0135] In the embodiments provided by the present invention, it should be understood that the disclosed device / terminal device and method can be implemented in other ways. For example, the device / terminal device embodiments described above are only illustrative. For example, the above division of modules or units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0136] The above-described 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 foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not essentially depart from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A lane traffic state estimation method based on a connected vehicle, characterized in that: The method comprises: Constructing a spatiotemporal grid of the target lane and setting a sensing range in the spatiotemporal grid for each connected vehicle on the target lane; Acquire driving data reported by each connected vehicle on the target lane, wherein the driving data includes the connected vehicle's own driving data and / or driving data of non-connected vehicles collected by the connected vehicle within its sensing range; Based on the driving data and the spatiotemporal grid, obtaining second driving data with a data status mark, wherein the data status includes valid and missing; Inputting the second driving data and the preset noise vector into a generator of a trained generative adversarial model, and outputting the completed third driving data after processing; The lane traffic state is estimated based on the third driving data.

2. The method for estimating lane traffic status based on a connected vehicle according to claim 1, characterized in that: Construct the spatiotemporal grid of the target lane, including: Obtain lane geometry data of the target lane; A spatiotemporal grid of the target lane is generated based on the lane geometry data and preset temporal resolution and spatial resolution.

3. The method for estimating lane traffic status based on a connected vehicle according to claim 1, characterized in that: Based on the driving data and the space-time grid, second driving data with a data state mark is obtained, including: After performing data preprocessing on the driving data, preprocessed driving data is obtained; According to the coverage state of each pre-processed driving data in the space-time grid, a mask matrix marking the data state of each grid is generated; Based on the mask matrix and all pre-processed driving data, second driving data with data status marks are obtained.

4. The method for estimating lane traffic status based on a connected vehicle according to claim 3, characterized in that: According to the coverage state of each pre-processed driving data in the space-time grid, a mask matrix marking the data state of each grid is generated, including: Mapping each preprocessed driving data to a corresponding grid of the space-time grid; For each grid, judging whether it is covered by the pre-processed driving data, and determining the data state of the grid according to the judging result; Generates a mask matrix based on the data state of all rasters.

5. The method for estimating lane traffic status based on a connected vehicle according to claim 4, characterized in that: For each grid, it is determined whether it is covered by the pre-processed driving data, and the data state of the grid is determined according to the determination result, including: For each grid, it is determined whether there is self-driving data of the connected vehicle, and if so, the data status of the grid is determined to be valid; If the grid does not contain the connected vehicle's own driving data, then determine whether the grid is within the sensing range of any connected vehicle; If the grid is not within the sensing range of any connected vehicle, the data status of the grid is determined to be missing; If the grid is within the sensing range of any connected vehicle, its status is determined to be valid or missing based on whether there is driving data of a non-connected vehicle.

6. The method for estimating lane traffic status based on a connected vehicle according to claim 1, characterized in that: The training steps of generating the adversarial model include: Acquire a training data set, wherein the training data set includes a plurality of training subsets corresponding to preset connected vehicle penetration rates, each training subset includes training driving data marked with a training data status, and benchmark driving data corresponding to each connected vehicle penetration rate, wherein the benchmark driving data comes from driving data collected by a traffic detection platform and connected vehicles, and the training data status includes valid and missing; Initializing a generative adversarial model, wherein the generative adversarial model includes a generator and a discriminator; For each connected vehicle penetration rate, train the generator and discriminator for a preset number of rounds; When the training of the penetration rate of all connected vehicles is completed, the data completion effect of the generative adversarial model is evaluated. After passing the evaluation, the trained generative adversarial model is obtained.

7. The method for estimating lane traffic status based on a connected vehicle according to claim 6, characterized in that: For each connected vehicle penetration rate, the generator and discriminator are trained in a preset round, including: The discriminator parameters are fixed, and the generator is trained for the first preset round. In each round, the noise training vector and the training driving data marked with the training data status at the current connected vehicle penetration rate are input into the generator, and the training driving data with the missing training data status is completed, and the completed driving data is output; The completed driving data and the baseline driving data with the same connected vehicle penetration rate are input into the discriminator, the generator loss is calculated and back-propagated to update the generator parameters; Fix the generator parameters, train the discriminator for the second preset round, input the completed driving data and benchmark driving data of the current connected vehicle penetration rate into the discriminator, calculate the discriminator loss and back-propagate to update the discriminator parameters.

8. A lane traffic state estimation device based on a connected vehicle, characterized in that: include: A grid construction module, used to construct a spatiotemporal grid of the target lane and set a sensing range in the spatiotemporal grid for each connected vehicle on the target lane; A data acquisition module, used to acquire driving data reported by each connected vehicle on the target lane, wherein the driving data includes the connected vehicle's own driving data and / or the driving data of non-connected vehicles collected by the connected vehicle within its sensing range; A second driving data generating module, configured to obtain second driving data with a data status mark based on the driving data and the spatiotemporal grid, wherein the data status includes valid and missing; A data completion module, used for inputting the second driving data and a preset noise vector into a generator of a trained generative adversarial model, and outputting completed third driving data after processing; A state estimation module is used to estimate the lane traffic state based on the third driving data.

9. A terminal, characterized in that: The terminal includes a memory, a processor, and a lane traffic state estimation program based on a connected vehicle stored in the memory and executable on the processor. When the lane traffic state estimation program based on a connected vehicle is executed by the processor, the steps of the lane traffic state estimation method based on a connected vehicle are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a lane traffic state estimation program based on a connected vehicle. When the lane traffic state estimation program based on a connected vehicle is executed by a processor, the steps of the lane traffic state estimation method based on a connected vehicle as described in any one of claims 1-7 are implemented.

Images