Hybrid traffic flow vehicle interaction relation identification method and system

Abstract

The invention provides a mixed traffic flow vehicle interaction relation identification method and system, and the method comprises the steps: constructing a dynamic graph according to the vehicle trajectory data of a road section in a first preset historical time period, and obtaining an edge feature matrix It, a node feature matrix Zt and an adjacent matrix At of the dynamic graph at each moment t; inputting the edge feature matrix It, the node feature matrix Zt and the adjacent matrix At into a vehicle interaction relationship feature extraction network to obtain a vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; and inputting the vehicle interaction relationship feature vector into a vehicle interaction relationship classification network to obtain a vehicle interaction relationship. According to the method, the complex interaction relationship between the vehicles in the mixed traffic flow can be identified more accurately, so that the understanding ability of the intelligent network connection vehicle to a complex traffic scene is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation, and particularly to a method and system for identifying the interaction relationship between vehicles in a mixed traffic flow. Background Art

[0002] With the rapid development of Connected and Autonomous Vehicles (CAVs) technology, the traffic system is gradually transforming from a traditional mode dominated by human-driven vehicles to a traffic mode that combines intelligent connected vehicles and human-driven vehicles. In this mixed traffic flow scenario, intelligent connected vehicles can achieve information sharing and collaborative decision-making between vehicles through vehicle networking technology, while human-driven vehicles rely on the experience and reaction ability of drivers. This mixed traffic flow scenario not only increases the complexity of the traffic system but also poses higher requirements for identifying the interaction relationship between vehicles.

[0003] In the mixed traffic flow scenario, the interaction relationship between vehicles exhibits high dynamics and uncertainty. Intelligent connected vehicles can perceive the surrounding environment in real time through sensors and communication devices and make decisions, while the behavior of human-driven vehicles is affected by various factors such as driver intention, emotion, and reaction speed. This difference makes it difficult for traditional vehicle interaction relationship identification methods to effectively cope. For example, rule-based methods cannot cover the complex interaction behaviors between intelligent connected vehicles and human-driven vehicles, and statistical-based methods often show low accuracy and robustness when dealing with mixed traffic flow data.

[0004] In recent years, the application of deep learning technology in the traffic field has provided new solutions for identifying vehicle interaction relationships. In particular, Graph Neural Networks (GNNs) and Long Short-Term Memory networks (LSTMs) make it possible to capture the temporal and spatial features between vehicles. However, most of the existing methods are targeted at a single type of traffic flow (such as all human-driven or all intelligent connected vehicles) and do not fully consider the heterogeneity and complexity of vehicle behaviors in the mixed traffic flow scenario. In addition, traditional graph neural networks are difficult to effectively capture the interaction features between intelligent connected vehicles and human-driven vehicles when dealing with dynamically changing traffic scenarios. Summary of the Invention

[0005] The present invention provides a method and system for identifying the interaction relationship between vehicles in a mixed traffic flow, which is used to solve the defect that the identification of the interaction relationship between vehicles for intelligent connected vehicles in the prior art is inaccurate, and to achieve more accurate identification of the complex interaction relationship between vehicles in the mixed traffic flow, thereby improving the understanding ability of intelligent connected vehicles for complex traffic scenarios.

[0006] The present invention provides a method for identifying the interaction relationship between vehicles in a mixed traffic flow, including:

[0007] Construct a dynamic graph based on the vehicle trajectory data of a road section within a first preset historical time period, and obtain the edge feature matrix I of the dynamic graph at each moment t t , the node feature matrix Z t and the adjacency matrix A t ;

[0008] Input the edge feature matrix I t , the node feature matrix Z t and the adjacency matrix A t into the vehicle interaction relationship feature extraction network to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network;

[0009] Input the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network to obtain the vehicle interaction relationship.

[0010] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, the first preset historical time period includes T time steps, and the vehicle trajectory information corresponding to each time step includes vehicle abscissa, ordinate, vehicle length, vehicle width, speed, longitudinal speed, lateral speed, acceleration, longitudinal acceleration, and lateral acceleration;

[0011] The mixed traffic flow includes connected and automated vehicles and human-driven vehicles;

[0012] The vehicle trajectory data of the connected and automated vehicles comes from the self-perception of the connected and automated vehicles, and the vehicle trajectory data of the human-driven vehicles comes from the perception of roadside equipment.

[0013] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, the dynamic graph is a directed graph, including nodes and edges. The vehicles in the road section are used as the nodes, and the interaction relationships existing between the vehicles are used as the edges;

[0014] If the distance between any two vehicles at each moment is less than a preset threshold, then there is an interaction relationship between the two vehicles.

[0015] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, the calculation formula of the adjacency matrix A t is:

[0016] C t ={(i, j)|dis(i, j)<D}

[0017]

[0018] where C t is the pair of vehicles with an interaction relationship at time t, dis(i, j) is the distance between the i-th vehicle and the j-th vehicle, and D is the preset threshold;

[0019] The edge feature matrix wherein, for each edge, e t is the edge feature vector, which is obtained by mapping and encoding the position of vehicle j relative to vehicle i at time t represented by the one - hot vector ; M is the number of edges in the dynamic graph; T is the transpose operation;

[0020] The node feature matrix wherein, for each vehicle, X t = [x 1 , x 2 , which is obtained by concatenating the vehicle trajectory feature vector x 1 and the velocity field feature vector x 2 ; x 1 is obtained by generating spatio - temporal encoding for each time step using the Transformer position encoding method, and inputting the sum of the spatio - temporal encoding corresponding to each time step and the vehicle trajectory information of each vehicle into the LSTM for extraction; x 2 is obtained by calculating the average velocity of all vehicles having an interaction relationship with each vehicle and performing mapping encoding on the average velocity; N is the number of vehicles in the road section.

[0021] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, inputting the edge feature matrix I t , the node feature matrix Z t and the adjacency matrix A t into the vehicle interaction relationship feature extraction network, to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network, including:

[0022] Inputting the edge feature matrix I t , the node feature matrix Z t and the adjacency matrix A t into the GAT layer to process the node features and generate new node features;

[0023] Combining the edge feature matrix I t and the new node features to generate the vehicle interaction relationship feature vector.

[0024] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, before inputting the edge feature matrix I t , the node feature matrix Z t and the adjacency matrix A t into the vehicle interaction relationship feature extraction network, it further includes:

[0025] Construct a dynamic graph based on the vehicle trajectory data of the road section within the second preset historical time period, and obtain the edge feature matrix I, the node feature matrix Z, and the adjacency matrix A at each moment t of the dynamic graph in the second preset historical time period; t and the node feature matrix Z t and the adjacency matrix A t ;

[0026] Input the edge feature matrix I, the node feature matrix Z, and the adjacency matrix A corresponding to the second preset historical time period into the vehicle interaction relationship feature extraction network to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; t and the node feature matrix Z t and the adjacency matrix A t Input the edge feature matrix I, the node feature matrix Z, and the adjacency matrix A corresponding to the second preset historical time period into the vehicle interaction relationship feature extraction network to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network;

[0027] After reducing the dimension of the vehicle interaction relationship feature vector corresponding to the second preset historical time period, use the K-means clustering method for clustering, and calculate the silhouette coefficient of the clustering according to the clustering result;

[0028] In the case that the silhouette coefficient is less than the preset maximum value, adjust the parameters of the vehicle interaction relationship feature extraction network to pre-train the vehicle interaction relationship feature extraction network until the silhouette coefficient is greater than or equal to the preset maximum value.

[0029] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, before inputting the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network, it further includes:

[0030] Use different clustering methods to cluster the vehicle interaction relationship feature vectors corresponding to the second preset historical time period;

[0031] Use the clustering fusion method to integrate the clustering results of multiple clustering methods to generate groups;

[0032] Calculate the confidence of each group, and manually label the interaction relationship labels for the groups with a confidence greater than the preset minimum value. The interaction relationship labels include following, overtaking, conflict, and others;

[0033] Continue to group and label the remaining unlabeled vehicle interaction relationship feature vectors until all the vehicle interaction relationship feature vectors corresponding to the second preset historical time period are labeled;

[0034] Input the labeled vehicle interaction relationship feature vectors into the feedforward neural network for training to obtain a pre-trained vehicle interaction relationship classification network.

[0035] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, before inputting the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network, it further includes:

[0036] The pre-trained vehicle interaction relationship feature extraction network and the pre-trained vehicle interaction relationship classification network are combined into an overall model for end-to-end training.

[0037] According to a method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention, the loss function for combining the pre-trained vehicle interaction relationship feature extraction network and the pre-trained vehicle interaction relationship classification network into an overall model for end-to-end training is:

[0038] L total = αL class +(1 - α)L contrast

[0039]

[0040] where L contrast is the loss of the vehicle interaction relationship feature extraction network, L class is the loss of the vehicle interaction relationship classification network, α is a preset adjustment coefficient, M is the number of samples of the vehicle interaction relationship feature vector, sim(z i , z j ) is the cosine similarity between the i-th sample z i and the j-th sample z j , P i is the set of positive samples of the same class as the i-th sample, N i is the set of negative samples of different classes from the i-th sample, τ is the temperature parameter used to control the sharpness of the similarity distribution; y i is the predicted value of the vehicle interaction relationship of the i-th sample, is the label of the vehicle interaction relationship of the i-th sample.

[0041] The present invention also provides a system for identifying vehicle interaction relationships in a mixed traffic flow, including:

[0042] A construction module for constructing a dynamic graph according to the vehicle trajectory data of a road section within a first preset historical time period, and obtaining the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t at each moment t of the dynamic graph;

[0043] An extraction module for inputting the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t into the vehicle interaction relationship feature extraction network to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network;

[0044] A classification module, configured to input the vehicle interaction relationship feature vector into a vehicle interaction relationship classification network to obtain a vehicle interaction relationship.

[0045] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method for identifying vehicle interaction relationships in a mixed traffic flow as described in any one of the above is implemented.

[0046] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for identifying vehicle interaction relationships in a mixed traffic flow as described in any one of the above is implemented.

[0047] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the method for identifying vehicle interaction relationships in a mixed traffic flow as described in any one of the above is implemented.

[0048] The method and system for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention model the interaction relationships between vehicles in a mixed traffic flow by constructing a dynamic graph, extract features from the edge feature matrix, node feature matrix, and adjacency matrix of the dynamic graph, obtain a more accurate vehicle interaction relationship feature vector, and thus can more accurately identify the complex interaction relationships between vehicles in a mixed traffic flow, improving the intelligent connected vehicle's understanding ability of complex traffic scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0050] Figure 1 is a flowchart of the method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention;

[0051] Figure 2 is a framework diagram of the method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention;

[0052] Figure 3 is a schematic diagram of the interaction scenario between vehicles in the method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention;

[0053] Figure 4 is a schematic diagram of the vehicle dynamic graph construction process in the method for identifying vehicle interaction relationships in a mixed traffic flow provided by the present invention;

[0054] Figure 5It is a schematic diagram of the velocity field around a certain vehicle in the method for identifying vehicle interaction relationships in mixed traffic flow provided by the present invention;

[0055] Figure 6 It is a schematic structural diagram of the system for identifying vehicle interaction relationships in mixed traffic flow provided by the present invention;

[0056] Figure 7 It is a schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0057] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0058] The following combines Figures 1 to 5 to describe a method for identifying vehicle interaction relationships in mixed traffic flow of the present invention, including:

[0059] Step 101: Construct a dynamic graph according to the vehicle trajectory data of the road section in the first preset historical time period, and obtain the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t ;

[0060] Step 102: Input the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t into the vehicle interaction relationship feature extraction network, and obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network;

[0061] Step 103: Input the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network to obtain the vehicle interaction relationship.

[0062] The present invention models the interaction relationships between vehicles in mixed traffic flow by constructing a dynamic graph, and through feature extraction of the edge feature matrix, node feature matrix and adjacency matrix of the dynamic graph, a more accurate vehicle interaction relationship feature vector is obtained, so that the complex interaction relationships between vehicles in mixed traffic flow can be identified more accurately, and the understanding ability of intelligent connected vehicles for complex traffic scenarios is improved.

[0063] Based on the above embodiments, in this embodiment, the first preset historical time period includes T time steps, and the vehicle trajectory information corresponding to each time step includes vehicle abscissa, vehicle ordinate, vehicle length, vehicle width, speed, longitudinal speed, lateral speed, acceleration, longitudinal acceleration, and lateral acceleration;

[0064] The mixed traffic flow includes connected and automated vehicles and human-driven vehicles;

[0065] The vehicle trajectory data of the connected and automated vehicles is sourced from the self-perception of the connected and automated vehicles, and the vehicle trajectory data of the human-driven vehicles is sourced from the perception of roadside devices.

[0066] The dimension of the vehicle trajectory data is [number of vehicles, time step T, data feature dimension 10]. The vehicle trajectory data is shown in Table 1 for details.

[0067] Table 1 Trajectory data table of a certain vehicle

[0068]

[0069] Based on the above embodiments, in this embodiment, the dynamic graph is a directed graph, including nodes and edges. The vehicles in the road section are taken as the nodes, and the interaction relationships existing between the vehicles are taken as the edges;

[0070] If the distance between any two vehicles at each moment is less than a preset threshold, then there is an interaction relationship between the two vehicles.

[0071] The construction of the dynamic graph is based on the vehicle trajectory data in the scenario of an urban expressway section without ramps. In this scenario, there are an indefinite number of vehicles in the road section, and different interaction relationships exist between the vehicles. Figure 3 is a schematic diagram of the interaction scenario between vehicles, Figure 4 is a schematic diagram of the dynamic graph construction process.

[0072] After the dynamic graph is constructed, the vehicles are regarded as points, and the Euclidean distance d between the vehicles at time t is calculated. The distance between the current vehicle and the vehicle in front is less than D 1 = 75 meters or the distance between the current vehicle and the vehicle behind is less than D 2 = 50 meters, then there is an interaction relationship between the current vehicle and the vehicle in front or the vehicle behind. An edge is established between the two vehicle nodes, and the adjacency matrix A t is used to represent the interaction connection between the vehicles at time t.

[0073] Based on the above embodiments, in this embodiment, the adjacency matrix A t has the following calculation formula:

[0074] C t = {(i, j)|dis(i, j) < D}

[0075]

[0076] Among them, C t is a pair of vehicles with an interaction relationship at time t, dis(i, j) is the distance between the i-th vehicle and the j-th vehicle, and D is a preset threshold;

[0077] The edge feature matrix Among them, the e corresponding to each edge t is an edge feature vector, which is obtained by mapping and encoding the position of vehicle j relative to vehicle i at time t represented by a one-hot vector ; M is the number of edges in the dynamic graph; T is the transpose operation;

[0078] The node feature matrix Among them, the X corresponding to each vehicle t = [x 1 , x 2 , which is obtained by concatenating the vehicle trajectory feature vector x 1 and the velocity field feature vector x 2 ; x 1 is obtained by generating spatio-temporal encoding for each time step using the Transformer position encoding method, and inputting the spatio-temporal encoding corresponding to each time step added to the vehicle trajectory information of each vehicle into the LSTM for extraction; x 2 is obtained by calculating the average velocity of all vehicles having an interaction relationship with each vehicle and performing mapping encoding on the average velocity; N is the number of vehicles in the section.

[0079] For example, at a certain moment, there are 4 vehicles, and the adjacency matrix may be:

[0080] As shown in Table 2, use an 8-dimensional one-hot vector to represent the direction encoding of vehicle i and vehicle j at time t, that is, to represent the position of vehicle j relative to vehicle i (front, rear, left, right, front left, front right, rear left, rear right), and map to the edge feature vector e t , and all edge feature vectors form the edge feature matrix at time t with dimensions [number of edges M, edge feature dimension number], and M is the number of non-zero values in A t , and in this embodiment, the edge feature dimension number can be taken as 256.

[0081] Table 2 One-hot direction encoding table

[0082]

[0083] The node feature matrix Z tIt is composed of the vehicle trajectory feature vectors and velocity field feature vectors of multiple vehicles, and the specific steps are as follows:

[0084] Step 1: Use the Transformer position encoding method to generate spatio-temporal encoding for each time step, and the calculation formula is:

[0085]

[0086] where pos is the time step position, i is the dimension index, and d is the feature dimension;

[0087] Step 2: Add the spatio-temporal encoding to the trajectory data to form new input features;

[0088] Step 3: Input the input features into an LSTM (Long Short Term Memory) to extract the vehicle trajectory feature vectors x of T time steps 1 ;

[0089] Step 4: Calculate the average speed of all vehicles around a certain vehicle at time t that have an interaction relationship with it. The formula is:

[0090]

[0091] where ν i is the speed of vehicle i, and n is the total number of all vehicles around a certain vehicle that have an interaction with it. Then map ν to encode it as the velocity field feature vector x 2 , Figure 5 is the schematic diagram of the velocity field around a certain vehicle;

[0092] There are differences in vehicle behaviors and interactions between vehicles at different speeds. Therefore, the concept of velocity field is introduced and incorporated into the node features.

[0093] Step 5: Concatenate x 1 and x 2 to form the node feature X t =[x 1 ,x 2 ;

[0094] Step 6: The expression of the node feature matrix Z t at time t is: The dimension is [number of vehicles N, number of node feature dimensions]. N is the number of vehicles in the research area at time t, and the size of the research area is set according to actual needs. In this embodiment, the number of node feature dimensions can be taken as 256, and the dimensions of x 1 and x 2 can both be taken as 128.

[0095] In this embodiment, a dynamic graph is constructed to model the interaction relationships between vehicles in mixed traffic flow. The LSTM is used to capture the temporal characteristics of vehicle behaviors, the speed field information is combined to enhance the extraction of spatial characteristics, and the graph attention network is used to extract the interaction characteristics between vehicles. In this way, the complex interaction relationships between vehicles in mixed traffic flow can be more accurately identified, thereby improving the understanding ability of intelligent connected vehicles for complex traffic scenarios.

[0096] Based on the above embodiment, as Figure 2 shown, in this embodiment, the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t are input into the vehicle interaction relationship feature extraction network to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network, including:

[0097] Input the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t into the GAT (Graph Attention Layer) layer to process the node features and generate new node features;

[0098] Combine the edge feature matrix I t and the new node features to generate the vehicle interaction relationship feature vector. In the present invention, the dimension number of the vehicle interaction relationship feature vector can be 256.

[0099] Based on the above embodiments, in this embodiment, before inputting the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t into the vehicle interaction relationship feature extraction network, it further includes:

[0100] Construct a dynamic graph according to the vehicle trajectory data of the road section in the second preset historical time period, and obtain the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t of the dynamic graph at each moment t in the second preset historical time period;

[0101] Input the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t corresponding to the second preset historical time period into the vehicle interaction relationship feature extraction network to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network;

[0102] After dimensionality reduction of the vehicle interaction relationship feature vector corresponding to the second preset historical time period, use the K-means clustering method for clustering, and calculate the silhouette coefficient of the clustering according to the clustering result;

[0103] In the case where the silhouette coefficient is less than the preset maximum value, adjust the parameters of the vehicle interaction relationship feature extraction network to pre-train the vehicle interaction relationship feature extraction network until the silhouette coefficient is greater than or equal to the preset maximum value.

[0104] Reduce the dimension and visualize the samples of the vehicle interaction relationship feature vector, analyze the structure of the feature space, use the K-means clustering method for clustering, and calculate the silhouette coefficient.

[0105] The quality of the extracted vehicle interaction relationship feature vector directly affects the subsequent interaction relationship classification effect. Therefore, it is necessary to evaluate the vehicle interaction relationship vector. t-SNE (t-Distributed Stochastic Neighbor Embedding) is a non-linear dimensionality reduction method widely used for the visualization of high-dimensional data. It maps high-dimensional data to a low-dimensional space by preserving the local similarity between data points, thus facilitating visual analysis. Therefore, in this embodiment, the t-SNE dimensionality reduction method can be used to reduce the interaction relationship feature vector to a two-dimensional plane for evaluation and analysis.

[0106] After dimensionality reduction, use the K-means clustering method to cluster into 4 categories and then calculate the silhouette coefficient. The formula for the silhouette coefficient of each sample i is:

[0107]

[0108] Among them, a(i) is the average distance from sample i to other samples in the same cluster, representing the compactness of sample i with samples in the same cluster; b(i) is the average distance from sample i to other samples in other clusters, representing the separation of sample i from other samples.

[0109] The final silhouette coefficient is the average value of the silhouette coefficients of all samples, and the formula is:

[0110]

[0111] After adjusting the parameters of the vehicle interaction relationship feature extraction network, continue to reduce the dimension and cluster the extracted vehicle interaction relationship feature vector after adjustment, and calculate the silhouette coefficient until the silhouette coefficient after dimensionality reduction and clustering is greater than or equal to the preset maximum value, such as 0.75.

[0112] On the basis of the above embodiment, in this embodiment, before inputting the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network, it further includes:

[0113] Cluster the vehicle interaction relationship feature vectors corresponding to the second preset historical time period using different clustering methods;

[0114] Use a clustering fusion method to integrate the clustering results of multiple clustering methods to generate groups;

[0115] Calculate the confidence of each group, and manually label the interaction relationship labels for the groups with a confidence greater than the preset minimum value. The interaction relationship labels include following, overtaking, conflict, and others;

[0116] Continue to group and label the remaining unlabeled vehicle interaction relationship feature vectors until all the vehicle interaction relationship feature vectors corresponding to the second preset historical time period are labeled;

[0117] Input the labeled vehicle interaction relationship feature vectors into a feedforward neural network for training to obtain a pre-trained vehicle interaction relationship classification network.

[0118] In this embodiment, a labeled interaction relationship dataset is constructed and used as a training set to input into a feedforward neural network for training to obtain an interaction relationship classification pre-training network. The number of neurons in each layer can be 256, 128, 64, and 4.

[0119] Based on the above embodiment, in this embodiment, before inputting the vehicle interaction relationship feature vectors into the vehicle interaction relationship classification network, it further includes:

[0120] Combine the pre-trained vehicle interaction relationship feature extraction network and the pre-trained vehicle interaction relationship classification network into an overall model for end-to-end training, optimize the overall performance, and realize the identification of vehicle interaction relationships.

[0121] Based on the above embodiment, the loss function for combining the pre-trained vehicle interaction relationship feature extraction network and the pre-trained vehicle interaction relationship classification network into an overall model for end-to-end training in this embodiment is:

[0122] L total = αL class +(1 - α)L contrast

[0123]

[0124] where L contrast is the loss of the vehicle interaction relationship feature extraction network, L class is the loss of the vehicle interaction relationship classification network, α is a preset adjustment coefficient, M is the number of samples of the vehicle interaction relationship feature vectors, sim(z i ,z j ) is the similarity of the i-th sample z iand the j-th sample z j 's cosine similarity, P i is the set of positive samples in the same class as the i-th sample, N i is the set of negative samples in different classes from the i-th sample, and τ is the temperature parameter used to control the sharpness of the similarity distribution; y i is the predicted value of the vehicle interaction relationship of the i-th sample, is the label of the vehicle interaction relationship of the i-th sample.

[0125] In this embodiment, contrastive learning is used to optimize and train the interaction relationship feature extraction network. There are M vehicle interaction relationship feature vector samples, and each sample belongs to one of 4 categories. The loss function can be expressed as L contrast . The cross-entropy loss function is used as the loss function for the classification task.

[0126] Combining the pre-trained network for vehicle interaction relationship feature extraction and the pre-trained network for vehicle interaction relationship classification into an overall model for end-to-end training includes three steps: determining the loss function, determining the parameter update strategy, and training the overall model. Update the parameters of the feature extraction network and the classification network. The learning rate of the feature extraction network can be set to 0.0001, and the learning rate of the classification network can be set to 0.001.

[0127] In this embodiment, LSTM, the velocity field, and the graph attention network are combined to specifically model the dynamics of vehicle behavior in the mixed traffic flow scenario. In addition, by combining pre-training and end-to-end training, the performance of the model is further optimized, making it more accurate and robust in the mixed traffic flow scenario. This method can not only effectively identify the interaction relationships between vehicles in the mixed traffic flow but also provide important technical support for traffic management and autonomous driving decision-making in the mixed traffic flow scenario.

[0128] Next, the mixed traffic flow vehicle interaction relationship identification system provided by the present invention will be described. The mixed traffic flow vehicle interaction relationship identification system described below can be correspondingly referred to the mixed traffic flow vehicle interaction relationship identification method described above.

[0129] As Figure 6 shown, the system includes a construction module 601, an extraction module 602, and a classification module 603, where:

[0130] The construction module 601 is used to construct a dynamic graph based on the vehicle trajectory data of the road section within the first preset historical time period, and obtain the edge feature matrix I t , node feature matrix Z t , and adjacency matrix A t ;

[0131] The extraction module 602 is used to input the edge feature matrix I t , the node feature matrix Z t and the adjacency matrix A t into the vehicle interaction relationship feature extraction network, and obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network;

[0132] The classification module 603 is used to input the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network to obtain the vehicle interaction relationship.

[0133] In this embodiment, by constructing a dynamic graph to model the interaction relationship between vehicles in the mixed traffic flow, and extracting features from the edge feature matrix, node feature matrix and adjacency matrix of the dynamic graph, a more accurate vehicle interaction relationship feature vector can be obtained, so that the complex interaction relationship between vehicles in the mixed traffic flow can be more accurately identified, and the understanding ability of intelligent connected vehicles for complex traffic scenarios can be improved.

[0134] Figure 7 FIG. illustrates a schematic physical structure diagram of an electronic device, as Figure 7 shown. The electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740. Among them, the processor 710, the communication interface 720, and the memory 730 complete mutual communication through the communication bus 740. The processor 710 can call the logical instructions in the memory 730 to execute the method for identifying the vehicle interaction relationship in the mixed traffic flow. The method includes: constructing a dynamic graph according to the vehicle trajectory data of the road section in the first preset historical time period, and obtaining the edge feature matrix I of the dynamic graph at each moment t t , the node feature matrix Z t and the adjacency matrix A t ; inputting the edge feature matrix I t , the node feature matrix Z t and the adjacency matrix A t into the vehicle interaction relationship feature extraction network, and obtaining the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; inputting the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network to obtain the vehicle interaction relationship.

[0135] In addition, when the logical instructions in the above-mentioned memory 730 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0136] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for identifying the vehicle interaction relationship in the mixed traffic flow provided by the above-mentioned various methods. The method includes: constructing a dynamic graph according to the vehicle trajectory data of a section within a first preset historical time period, and obtaining the edge feature matrix I of the dynamic graph at each moment t t , node feature matrix Z t and adjacency matrix A t ; inputting the edge feature matrix I t , node feature matrix Z t and adjacency matrix A t into a vehicle interaction relationship feature extraction network to obtain a vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; inputting the vehicle interaction relationship feature vector into a vehicle interaction relationship classification network to obtain a vehicle interaction relationship.

[0137] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the method for identifying the vehicle interaction relationship in the mixed traffic flow provided by the above-mentioned various methods. The method includes: constructing a dynamic graph according to the vehicle trajectory data of a section within a first preset historical time period, and obtaining the edge feature matrix I of the dynamic graph at each moment t t , node feature matrix Z t and adjacency matrix A t ; inputting the edge feature matrix I t , node feature matrix Z t and adjacency matrix A tInput the vehicle interaction relationship feature extraction network to obtain the vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; input the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network to obtain the vehicle interaction relationship.

[0138] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0139] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; 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 for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying vehicle interaction relationships in mixed traffic flow, characterized in that: include: Construct a dynamic graph based on the vehicle trajectory data of the road section within the first preset historical time period, and obtain the edge feature matrix I of the dynamic graph at each time t t , node feature matrix Z t and the adjacency matrix A t ; The edge feature matrix I t , node feature matrix Z t and the adjacency matrix A t Inputting a vehicle interaction relationship feature extraction network to obtain a vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; The vehicle interaction relationship feature vector is input into a vehicle interaction relationship classification network to obtain a vehicle interaction relationship.

2. The mixed traffic flow vehicle interaction relationship identification method according to claim 1, characterized in that: The first preset historical time period includes T time steps, and the vehicle trajectory information corresponding to each time step includes the vehicle horizontal coordinate, vertical coordinate, vehicle length, vehicle width, speed, longitudinal speed, lateral speed, acceleration, longitudinal acceleration and lateral acceleration; The mixed traffic flow includes intelligent connected vehicles and human-driven vehicles; The vehicle trajectory data of the intelligent connected vehicle comes from the intelligent connected vehicle's own perception, and the vehicle trajectory data of the human-driven vehicle comes from the roadside equipment perception.

3. The mixed traffic flow vehicle interaction relationship identification method according to claim 1, characterized in that: The dynamic graph is a directed graph including nodes and edges, wherein the vehicles in the road section are taken as the nodes, and the interaction relationships between the vehicles are taken as the edges; If the distance between any two vehicles at any moment is less than a preset threshold, then the two vehicles have an interactive relationship.

4. The mixed traffic flow vehicle interaction relationship identification method according to claim 3 is characterized in that: The adjacency matrix A t The calculation formula is: C t ={(i,j)|dis(i,j)<D} Among them, C t is the vehicle pair that has an interactive relationship at time t, dis(i,j) is the distance between the i-th vehicle and the j-th vehicle, and D is the preset threshold; The edge feature matrix Among them, the e corresponding to each edge t is the edge feature vector, by The position of vehicle j relative to vehicle i at time t is obtained by mapping and encoding as the direction encoding; M is the number of edges in the dynamic graph; T is the transposition operation; The node feature matrix Among them, each vehicle corresponds to X t =[x1,x2], which is obtained by concatenating the vehicle trajectory feature vector x1 and the velocity field feature vector x2; x1 is obtained by generating a spatiotemporal code for each time step using the Transformer position encoding method, and the spatiotemporal code corresponding to each time step is added to the vehicle trajectory information of each vehicle and then input into the LSTM for extraction; x2 is obtained by calculating the average speed of all vehicles that have an interactive relationship with each vehicle, and mapping and encoding the average speed; N is the number of vehicles in the road section.

5. The mixed traffic flow vehicle interaction relationship identification method according to claim 1, characterized in that: The edge feature matrix I t , node feature matrix Z t and the adjacency matrix A t Inputting a vehicle interaction relationship feature extraction network to obtain a vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network includes: The edge feature matrix I t , node feature matrix Z t and the adjacency matrix A t Input the GAT layer to process node features and generate new node features; The edge feature matrix I t Combined with the new node feature, a vehicle interaction relationship feature vector is generated.

6. The mixed traffic flow vehicle interaction relationship identification method according to any one of claims 1 to 5, characterized in that: In the edge feature matrix I t , node feature matrix Z t and the adjacency matrix A t Before inputting the vehicle interaction relationship feature extraction network, it also includes: A dynamic graph is constructed based on the vehicle trajectory data of the road section in the second preset historical time period, and an edge feature matrix I of the dynamic graph of the second preset historical time period at each time t is obtained. t , node feature matrix Z t and the adjacency matrix A t ; The edge feature matrix I corresponding to the second preset historical time period t , node feature matrix Z t and the adjacency matrix A t Inputting a vehicle interaction relationship feature extraction network to obtain a vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; After reducing the dimension of the vehicle interaction relationship feature vector corresponding to the second preset historical time period, clustering is performed using a K-means clustering method, and a clustering silhouette coefficient is calculated according to the clustering result; When the silhouette coefficient is less than a preset maximum value, the parameters of the vehicle interaction relationship feature extraction network are adjusted to pre-train the vehicle interaction relationship feature extraction network until the silhouette coefficient is greater than or equal to the preset maximum value.

7. The mixed traffic flow vehicle interaction relationship identification method according to claim 6, characterized in that: Before the vehicle interaction relationship feature vector is input into the vehicle interaction relationship classification network, the method further includes: Clustering the vehicle interaction relationship feature vectors corresponding to the second preset historical time period using different clustering methods; Use clustering fusion method to integrate the clustering results of multiple clustering methods to generate groups; Calculate the confidence of each group, and manually label the groups whose confidence is greater than a preset minimum value with interaction relationship labels, wherein the interaction relationship labels include following, overtaking, conflict, and others; Continue to group and mark the remaining unlabeled vehicle interaction relationship feature vectors until all vehicle interaction relationship feature vectors corresponding to the second preset historical time period are labeled; The labeled vehicle interaction relationship feature vectors are input into a feedforward neural network for training to obtain a pre-trained vehicle interaction relationship classification network.

8. The mixed traffic flow vehicle interaction relationship identification method according to claim 7, characterized in that: Before the vehicle interaction relationship feature vector is input into the vehicle interaction relationship classification network, the method further includes: The pre-trained vehicle interaction relationship feature extraction network and the pre-trained vehicle interaction relationship classification network are combined into an overall model for end-to-end training.

9. The mixed traffic flow vehicle interaction relationship identification method according to claim 8, characterized in that: The loss function of combining the pre-trained vehicle interaction relationship feature extraction network and the pre-trained vehicle interaction relationship classification network into an overall model for end-to-end training is: L total =αL class +(1-α)L contrast Among them, L contrast is the loss of the vehicle interaction feature extraction network, L class is the loss of the vehicle interaction relationship classification network, α is the preset adjustment coefficient, M is the number of samples of the vehicle interaction relationship feature vector, sim(z i ,z j ) is the i-th sample z i and the jth sample z j The cosine similarity, P i is the set of positive samples of the same category as the i-th sample, N i is a set of negative samples of different categories from the i-th sample, τ is a temperature parameter used to control the sharpness of the similarity distribution; y i is the predicted value of the vehicle interaction relationship of the i-th sample, is the vehicle interaction relationship label of the i-th sample.

10. A mixed traffic flow vehicle interaction relationship identification system, characterized in that: include: A construction module is used to construct a dynamic graph based on the vehicle trajectory data of the road section within the first preset historical time period, and obtain the edge feature matrix I of the dynamic graph at each time t t , node feature matrix Z t and the adjacency matrix A t ; Extraction module, used to extract the edge feature matrix I t , node feature matrix Z t and the adjacency matrix A t Inputting a vehicle interaction relationship feature extraction network to obtain a vehicle interaction relationship feature vector output by the vehicle interaction relationship feature extraction network; The classification module is used to input the vehicle interaction relationship feature vector into the vehicle interaction relationship classification network to obtain the vehicle interaction relationship.

Images