A vehicle overlapping group detection method based on graph attention autoencoder

Abstract

The application provides a vehicle overlapping group detection method based on a graph attention automatic encoder, and belongs to the technical fields of Internet of Vehicles and social networks. The technical scheme comprises the following steps: step 1, inputting network topology structure and node attribute information into an encoder part of a graph attention automatic encoder; step 2, performing encoding on the input network topology structure and node attribute information by a graph attention mechanism of the encoder; step 3, combining new node embedding representation with original embedding representation by a decoder to restore the network topology structure and the node attribute information; step 4, further adjusting node embedding by using a modularity optimization enhancement module to optimize constraints; and step 5, performing semi-supervised clustering by using prior information to obtain a social group detection result based on node embedding representation. The application has the beneficial effect of effectively improving the accuracy of overlapping social group detection of vehicle nodes.

Description

Technical Field

[0001] This invention relates to the fields of vehicle networking technology and social networking technology, and in particular to a method for detecting overlapping vehicle groups based on a graph attention autoencoder. Background Technology

[0002] With the development of intelligent transportation systems, vehicles can interconnect via wireless communication technology to form a self-organizing network. This network can support applications such as information sharing, navigation, and traffic management. Utilizing vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication, this network aims to improve road safety, optimize traffic flow, and enhance environmental monitoring capabilities. Community detection, which identifies densely connected groups of nodes within the network—typically densely connected areas with strong inter-node connections and relatively sparse connections to other communities—is crucial for managing vehicles and maintaining security within the vehicle social network. In a vehicle social network, communication efficiency can be improved, security and privacy protection enhanced, resource allocation optimized, and local management achieved.

[0003] Most existing community detection methods focus on identifying non-overlapping communities. For example, the paper "Research on Hybrid Trust Evaluation Mechanism in In-Vehicle Social Networks" proposes a community detection algorithm based on user interest similarity, but this algorithm can only detect non-overlapping vehicle communities. However, overlapping communities are very common in in-vehicle social networks. For a vehicle, it may participate in different communities at different times or in different contexts. Summary of the Invention

[0004] Community detection is particularly important in vehicle social networks, helping to identify densely connected groups of vehicles. However, most existing community detection methods focus primarily on non-overlapping communities and typically only consider the network topology while ignoring node attribute information. To address these limitations, this invention provides a vehicle overlapping group detection method based on a graph attention autoencoder. This method includes a graph attention autoencoder module, a modularity optimization enhancement module, and a semi-supervised clustering module. First, the graph attention autoencoder module fuses topology information and node attribute information and encodes nodes using a graph attention mechanism. Second, the modularity optimization enhancement module accurately captures the structure of overlapping communities. Finally, the semi-supervised clustering module fuses prior information to improve the accuracy of community detection. Through these three modules, the accuracy of vehicle node overlapping community detection can be effectively improved.

[0005] To achieve the aforementioned objectives, the present invention employs the following technical solution: a vehicle overlap group detection method based on a graph attention autoencoder, comprising three modules: a graph attention autoencoder module, a module degree optimization and enhancement module, and a semi-supervised clustering module, specifically including the following steps:

[0006] Step 1: Input the network topology and node attribute information into the encoder part of the graph attention autoencoder. Through the autoencoder structure, not only can the latent representation of the graph be learned, but the model can also be optimized by reconstructing the loss, so that it can better capture the key structural features of the graph.

[0007] Step 2: The encoder's graph attention mechanism encodes the input network topology and node attribute information. After processing by the attention mechanism, it is represented as multiple different hidden layer embeddings. These different embeddings capture the complex relationships and attribute differences between nodes. Then, a nonlinear transformation is performed through an activation function to generate new node embeddings.

[0008] Step 3: The decoder combines the new node embedding representation with the original graph embedding representation, restores the network topology and node attribute information through the decoding process, and compares it with the original topology and attribute information to evaluate the quality of the reconstruction.

[0009] Step 4: Enhance the optimization constraints of the modules through modularity optimization, further adjust the node embedding, ensure the rationality of the community structure, use the M / M / 1 queuing model, and set q. ij ≥0 and The optimization is performed based on constraints, and the final output is a modularity optimization enhancement loss L. M To measure the rationality of community testing.

[0010] Step 5: Perform semi-supervised clustering using prior information, and reconstruct the loss L based on the graph. R Modularity optimization enhances loss L M Semi-supervised clustering loss L S Optimization is performed to obtain node-based embedded representations for generating community detection results.

[0011] This invention proposes a vehicle overlap group detection method based on a graph attention autoencoder, and the specific steps of step 2 are as follows:

[0012] Step 2.1: In the multi-head attention mechanism, l independent attention heads are introduced, each with its own linear transformation matrix W. (l) Used to extract the original node features h i The transformation is performed on the feature vector of each node individually, as shown in the following formula:

[0013]

[0014] in It is the new feature representation of node i in the l-th head.

[0015] Step 2.2, for each node v i and v j Calculate an attention score, representing node v j For node v i The importance of this score. This score is based on the feature representation after linear transformation and the calculation. The specific formula is:

[0016]

[0017] Where a (l) Let F' be the attention weight vector of the l-th head, with a length of 2F', used to map the concatenated vector to a scalar, where || represents the concatenation operation, and F' is the dimension of the new feature space for each head. LeakyReLU is the activation function.

[0018] Step 2.3: Use the softmax function to normalize the original attention scores of all neighboring nodes to obtain the final attention coefficients. This process ensures that the sum of the attention coefficients of all neighboring nodes is 1:

[0019]

[0020] Step 2.4: For each attention head l, use the attention coefficient. We perform a weighted summation of the features of neighboring nodes to obtain the feature output of each head l.

[0021]

[0022] Step 2.5: Average the output features of all heads to obtain the final node representation:

[0023]

[0024] This invention proposes a vehicle overlap group detection method based on a graph attention autoencoder, and the specific steps of step 3 are as follows:

[0025] Step 3.1: Reconstruct the adjacency matrix A of the graph using the latent representation Z of the nodes, where A is the predicted adjacency matrix. When using a multi-head attention mechanism, the latent representation z... i Typically, the output features z from multiple attention heads are used. i (l) It is aggregated. The latent representation of each node is obtained by averaging the output features of multiple heads:

[0026]

[0027] Step 3.2, the latent representation z of the node i It remains an F'-dimensional vector. The decoder formula remains unchanged, and the calculation formula is as follows:

[0028]

[0029] in This refers to the existence of the edge between node i and node j predicted by the decoder. i and z j These are the latent representations of nodes i and j, respectively, and σ is the activation function, typically the sigmoid function.

[0030]

[0031] Step 3.3: The final result of the decoder is used as the reconstructed node attribute value. The loss function for reconstructing the node attribute value is defined as Equation 9.

[0032]

[0033] Where X i It is the original binary attribute of the i-th node. This is the probability that the i-th node's attribute is 1, as predicted by the model. The loss function calculates the binary cross-entropy of the binary attribute reconstruction for each node, optimizing the model's reconstruction performance by minimizing the loss.

[0034] Step 3.4, the loss function for adjacency matrix prediction is defined as Equation 10:

[0035]

[0036] Where A ij This represents the elements in the original adjacency matrix, specifically whether an edge exists between node i and node j. If A ij =1, indicating that there is an edge between node i and node j; if A ij =0 indicates that there are no edges. It is an element in the adjacency matrix predicted by the model, representing the probability that there is an edge between node i and node j as predicted by the model.

[0037] Step 3.5: The final reconstruction loss function is formed by combining the reconstruction loss of topology and node attributes.

[0038] L R =αL attr +βL top (11)

[0039] α and β are hyperparameters used to balance the impact of node attribute reconstruction loss and adjacency matrix prediction loss.

[0040] This invention proposes a vehicle overlap group detection method based on a graph attention autoencoder, and the specific steps of step 4 are as follows:

[0041] Step 4.1: The final node representation matrix H output by the encoder is further refined through modularity optimization. Modularity is an indicator of the strength of the network community structure, and can be defined as:

[0042]

[0043] Where d i and d j H represents the degree of node i and node j. i and H j The attribute vectors representing nodes i and j. This represents the attribute similarity between nodes i and j.

[0044] Step 4.2: Add a Markov queuing system to the modularity maximization module to create an enhanced modularity optimization module. In the Markov queuing system, the state transitions between nodes can be represented by the state transition matrix P. Define the enhanced modularity optimization objective as:

[0045]

[0046] Where P ij A represents the probability of node i transitioning to node j. ij P ij This means that we consider not only whether there are direct edges between nodes, but also the strength of dynamic interactions between nodes.

[0047] Step 4.3: To optimize this enhanced modularity, the corresponding loss function is defined as follows:

[0048] L M =-Q M (14)

[0049] This invention proposes a vehicle overlap group detection method based on a graph attention autoencoder, and the specific steps of step 5 are as follows:

[0050] Step 5.1: By combining the known node label information Y with the semi-supervised clustering module, the community detection result is finally obtained. The semi-supervised strategy maintains the consistency of known labels while simultaneously uncovering the potential community affiliations of unlabeled nodes, making the overall community partitioning more reasonable. Therefore, the cross-entropy loss is used as the semi-supervised loss L. S :

[0051]

[0052] in Y is a collection of labeled data. ij This indicates the true label of whether the labeled node i belongs to community j.

[0053] Step 5.2: To obtain a more accurate community member representation matrix H, using a graph attention autoencoder alone is insufficient. Integrating the graph attention autoencoder, modularity optimization enhancement, and semi-supervised clustering module into a single unit yields more accurate community detection results. The final objective function is defined in Equation 16.

[0054] L = L R +γL S -λL M (16)

[0055] Where γ and λ are hyperparameters representing different loss weights, used to adjust the contribution of the corresponding modules. R L M and L S denoted as the loss of the graph attention autoencoder, modularity optimization enhancement, and self-trained clustering module, respectively.

[0056] Step 5.3: Train the model based on the final loss function L to obtain the final community detection result.

[0057] The main symbols and their related explanations are shown in Table 1 below.

[0058] Table 1. Symbol Explanation

[0059]

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] 1. Integration of Topology and Attribute Information: The method proposed in this invention utilizes a Graph Attention Autoencoder (GAAE) to effectively integrate topological structure and node attribute information, enabling the capture of more complex network features. This integration provides a more comprehensive understanding of the relationships and characteristics of vehicles in a social network, thereby achieving more accurate community detection. In vehicular social networks, vehicles not only interact with each other through topological connections but may also be influenced by the social attributes of drivers (such as occupation and interests). Such integration makes community detection more accurate, better reflecting the true structure of vehicular communities, and contributing to more efficient traffic management and resource optimization.

[0062] 2. Modularity Optimization Enhancement Module: This module introduces a Markov queuing model based on maximizing modularity. This innovation allows for a more flexible representation of overlapping communities, enabling nodes to belong to multiple communities simultaneously. This is particularly important in vehicular networks, as vehicle social relationships may change over time. Traditional community detection methods often strictly classify nodes into a single community, failing to effectively identify overlapping communities. By introducing the Markov queuing model, this module can flexibly represent the probability of a node belonging to multiple communities. This method is suitable for vehicular social networks because vehicles may participate in different communities at different times and in different situations. The optimized modularity method enhances the ability to model dynamic interactions and relationships, thereby improving the accuracy and reliability of community detection.

[0063] 3. Semi-supervised Clustering Module: This module utilizes prior information (known community labels) to guide community detection. By introducing known community labels as prior information, this module can significantly improve the accuracy of community detection, ensuring that the detected communities more closely match the actual community structure and overcoming the limitations of purely unsupervised methods. The semi-supervised method combines potential community affiliations with a small amount of labeled information and unlabeled data, enabling more reasonable community segmentation. This approach is particularly important in vehicle social networks because, in practical applications, vehicle social behavior and interaction patterns may be influenced by various factors; known label information can effectively guide model learning and improve the final detection performance. Attached Figure Description

[0064] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0065] Figure 1 This is a model framework diagram of an embodiment of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0067] Example 1

[0068] See Figure 1 The present invention provides the following technical solution: This invention is a vehicle overlap group detection method based on graph attention autoencoder (CDGAAE), and the overall process of the method is shown in the attached figure. Figure 1 As shown in the figure. The technical solution of the present invention will be further described below with reference to specific embodiments:

[0069] Step 1: Input the network topology and node attribute information into the encoder part of the graph attention autoencoder. Through the autoencoder structure, not only can the latent representation of the graph be learned, but the model can also be optimized through reconstruction loss to better capture the key structural features of the graph.

[0070] Step 2: The encoder's graph attention mechanism encodes the input network topology and node attribute information. After processing by the attention mechanism, it is represented as multiple different hidden layer embeddings. These different embeddings capture the complex relationships and attribute differences between nodes. A nonlinear transformation is then performed through an activation function to generate new node embeddings.

[0071] Step 3: The decoder combines the new node embedding representation with the original graph embedding representation to reconstruct the network topology and node attribute information through the decoding process. It then compares this reconstruction with the original topology and attribute information to evaluate the quality of the reconstruction.

[0072] Step 4: Enhance the optimization constraints of the modules through modularity optimization, further adjust node embedding, and ensure the rationality of the community structure. Use the M / M / 1 queuing model and set q. ij ≥0 and Optimization is performed based on constraints. The final output is a modularity-enhanced loss L. M To measure the rationality of community testing.

[0073] Step 5: Perform semi-supervised clustering using prior information, and reconstruct the loss L based on the graph. R Modularity optimization enhances loss L M Semi-supervised clustering loss L S Optimization is performed to obtain node-based embedded representations for generating community detection results.

[0074] The specific steps of step 2 are as follows:

[0075] Step 2.1: In the multi-head attention mechanism, l independent attention heads are introduced, each with its own linear transformation matrix W. (l) Used to extract the original node features h i The transformation is performed on the feature vector of each node individually, as shown in the following formula:

[0076]

[0077] in It is the new feature representation of node i in the l-th head.

[0078] Step 2.2, for each node v i and v j Calculate an attention score, representing node vj For node v i The importance of this score. This score is based on the feature representation after linear transformation and the calculation. The specific formula is:

[0079]

[0080] Where a (l) Let F' be the attention weight vector of the l-th head, with a length of 2F', used to map the concatenated vector to a scalar, where || represents the concatenation operation, and F' is the dimension of the new feature space for each head. LeakyReLU is the activation function.

[0081] Step 2.3: Use the softmax function to normalize the original attention scores of all neighboring nodes to obtain the final attention coefficients. This process ensures that the sum of the attention coefficients of all neighboring nodes is 1:

[0082]

[0083] Step 2.4: For each attention head l, use the attention coefficient. We perform a weighted summation of the features of neighboring nodes to obtain the feature output of each head l.

[0084]

[0085] Step 2.5: Average the output features of all heads to obtain the final node representation:

[0086]

[0087] The specific steps of step 3 are as follows:

[0088] Step 3.1: Reconstruct the adjacency matrix A of the graph using the latent representation Z of the nodes, where A is the predicted adjacency matrix. When using a multi-head attention mechanism, the latent representation z... i Typically, the output features are derived from multiple attention heads. It is aggregated. The latent representation of each node is obtained by averaging the output features of multiple heads:

[0089]

[0090] Step 3.2, the latent representation z of the node i It remains an F'-dimensional vector. The decoder formula remains unchanged, and the calculation formula is as follows:

[0091]

[0092] in This refers to the existence of the edge between node i and node j predicted by the decoder. i and z j These are the latent representations of nodes i and j, respectively, and σ is the activation function, typically the sigmoid function.

[0093]

[0094] Step 3.3: The final result of the decoder is used as the reconstructed node attribute value. The loss function for reconstructing the node attribute value is defined as Equation 9.

[0095]

[0096] Where X i It is the original binary attribute of the i-th node. This is the probability that the i-th node's attribute is 1, as predicted by the model. The loss function calculates the binary cross-entropy of the binary attribute reconstruction for each node, optimizing the model's reconstruction performance by minimizing the loss.

[0097] Step 3.4, the loss function for adjacency matrix prediction is defined as Equation 10:

[0098]

[0099] Where A ij This represents the elements in the original adjacency matrix, specifically whether an edge exists between node i and node j. If A ij =1, indicating that there is an edge between node i and node j; if A ij =0 indicates that there are no edges. It is an element in the adjacency matrix predicted by the model, representing the probability that there is an edge between node i and node j as predicted by the model.

[0100] Step 3.5: The final reconstruction loss function is formed by combining the reconstruction loss of topology and node attributes.

[0101] L R =αL attr +βL top (11)

[0102] α and β are hyperparameters used to balance the impact of node attribute reconstruction loss and adjacency matrix prediction loss.

[0103] The specific steps of step 4 are as follows:

[0104] Step 4.1: The final node representation matrix H output by the encoder is further refined through modularity optimization. Modularity is an indicator of the strength of the network community structure, and can be defined as:

[0105]

[0106] Where d i and d j H represents the degree of node i and node j. i and H j The attribute vectors representing nodes i and j. This represents the attribute similarity between nodes i and j.

[0107] Step 4.2: Add a Markov queuing system to the modularity maximization module to create an enhanced modularity optimization module. In the Markov queuing system, the state transitions between nodes can be represented by the state transition matrix P. Define the enhanced modularity optimization objective as:

[0108]

[0109] Where P ij A represents the probability of node i transitioning to node j. ij P ij This means that we consider not only whether there are direct edges between nodes, but also the strength of dynamic interactions between nodes.

[0110] Step 4.3: To optimize this enhanced modularity, the corresponding loss function is defined as follows:

[0111] L M =-Q M (14)

[0112] The specific steps of step 5 are as follows:

[0113] Step 5.1: By combining the known node community label information Y with the semi-supervised clustering module, the final community detection result is obtained. The semi-supervised strategy maintains the consistency of known labels while simultaneously uncovering the potential community affiliations of unlabeled nodes, making the overall community partitioning more reasonable. Therefore, the cross-entropy loss is used as the semi-supervised loss L. S :

[0114]

[0115] in Y is a collection of labeled data. ij This indicates the true label of whether the labeled node i belongs to community j.

[0116] Step 5.2: To obtain a more accurate community member representation matrix H, using a graph attention autoencoder alone is insufficient. Integrating the graph attention autoencoder, modularity optimization enhancement, and semi-supervised clustering module into a single unit yields more accurate community detection results. The final objective function is defined in Equation 16.

[0117] L = L R +γL S -λL M (16)

[0118] Where γ and λ are hyperparameters representing different loss weights, used to adjust the contribution of the corresponding modules. R L M and L S denoted as the loss of the graph attention autoencoder, modularity optimization enhancement, and self-trained clustering module, respectively.

[0119] Step 5.3: Train the model based on the final loss function L to obtain the final community detection result.

[0120] Example 2

[0121] Comparative experiments were conducted using the Facebook414 social network dataset. To assess the accuracy of overlapping community detection, three widely used evaluation metrics were introduced: Overlap Normalized Mutual Information (ONMI), Average F1 Score (F1), and Jaccard Index. The results were also compared with five competing methods, including both unsupervised and semi-supervised methods, based on whether prior information was integrated. The proportion of prior information was varied from 0% to 10% in 2% increments. The results for ONMI, F1, and Jaccard Index are shown in Tables 2, 3, and 4, respectively.

[0122] Table 2 fb414_ONMI

[0123] method 0％ 2％ 4％ 6％ 8％ 10％ CESNA 0.2 0.2 0.2 0.2 0.2 0.2 SCI 0.28 0.28 0.28 0.28 0.28 0.28 Bespoke 0.15 0.15 0.15 0.15 0.15 0.15 NOCD 0.5 0.5 0.5 0.5 0.5 0.5 SSGCAE 0.26 0.31 0.35 0.36 0.37 0.4 CDGAAE 0.4 0.5 0.51 0.52 0.53 0.55

[0124] Table 3 fb414_F1

[0125] method 0％ 2％ 4％ 6％ 8％ 10％ CESNA 0.49 0.49 0.49 0.49 0.49 0.49 SCI 0.19 0.19 0.19 0.19 0.19 0.19 Bespoke 0.39 0.39 0.39 0.39 0.39 0.39 NOCD 0.41 0.41 0.41 0.41 0.41 0.41 SSGCAE 0.4 0.6 0.61 0.61 0.61 0.61 CDGAAE 0.48 0.7 0.73 0.77 0.8 0.81

[0126] Table 4 fb414_Jaccard Index

[0127] method 0％ 2％ 4％ 6％ 8％ 10％ CESNA 0.39 0.39 0.39 0.39 0.39 0.39 SCI 0.27 0.27 0.27 0.27 0.27 0.27 Bespoke 0.3 0.3 0.3 0.3 0.3 0.3 NOCD 0.22 0.22 0.22 0.22 0.22 0.22 SSGCAE 0.53 0.67 0.69 0.69 0.69 0.7 CDGAAE 0.58 0.75 0.8 0.82 0.83 0.85

[0128] As can be seen on the Facebook414 dataset, CDGAAE's performance becomes increasingly superior with the increase of the proportion of prior information. Based on the results in Tables 2, 3, and 4, we can draw the following conclusions.

[0129] (1) CDGAAE outperforms unsupervised methods such as CESNA and NOCD, indicating that our method effectively incorporates prior information. Compared to these unsupervised methods, the performance of CDGAAE becomes increasingly superior with the increase in the proportion of prior information.

[0130] (2) CDGAAE outperforms the semi-supervised method SSGCAE, which demonstrates the importance of extracting community structure and reconstructing relevant attribute features. CDGAAE introduces a graph attention mechanism that dynamically assigns different weights to each node, and uses an optimized modular metric to obtain node representations, which helps to obtain topologically compact community results.

[0131] Example 3

[0132] The Facebook1684 social network dataset was selected for the second comparative experiment. The evaluation metrics used were Overlap Normalized Mutual Information (ONMI), average F1 score (F1), and Jaccard Index. The results were also compared with five competing methods, including both unsupervised and semi-supervised methods, based on whether prior information was integrated. The proportion of prior information was varied from 0% to 10% in 2% increments. The results for ONMI, F1, and Jaccard Index are shown in Tables 5, 6, and 7, respectively.

[0133] Table 5 fb1684_ONMI

[0134] method 0％ 2％ 4％ 6％ 8％ 10％ CESNA 0.19 0.19 0.19 0.19 0.19 0.19 SCI 0.24 0.24 0.24 0.24 0.24 0.24 Bespoke 0.15 0.15 0.15 0.15 0.15 0.15 NOCD 0.26 0.26 0.26 0.26 0.26 0.26 SSGCAE 0.27 0.42 0.43 0.45 0.45 0.46 CDGAAE 0.27 0.45 0.46 0.47 0.48 0.5

[0135] Table 6 fb1684_F1

[0136] method 0％ 2％ 4％ 6％ 8％ 10％ CESNA 0.37 0.37 0.37 0.37 0.37 0.37 SCI 0.15 0.15 0.15 0.15 0.15 0.15 Bespoke 0.31 0.31 0.31 0.31 0.31 0.31 NOCD 0.15 0.15 0.15 0.15 0.15 0.15 SSGCAE 0.1 0.4 0.41 0.5 0.51 0.61 CDGAAE 0.33 0.6 0.62 0.64 0.68 0.7

[0137] Table 7 fb1684_Jaccard Index

[0138] method 0％ 2％ 4％ 6％ 8％ 10％ CESNA 0.37 0.37 0.37 0.37 0.37 0.37 SCI 0.22 0.22 0.22 0.22 0.22 0.22 Bespoke 0.26 0.26 0.26 0.26 0.26 0.26 NOCD 0.15 0.15 0.15 0.15 0.15 0.15 SSGCAE 0.49 0.5 0.51 0.52 0.6 0.61 CDGAAE 0.6 0.76 0.79 0.8 0.81 0.82

[0139] The results in Tables 5, 6, and 7 are consistent with those obtained in Example 2. On the Facebook414 dataset, the performance of CDGAAE becomes increasingly superior as the proportion of prior information increases. It outperforms unsupervised methods such as CESNA and NOCD, demonstrating the effectiveness of fusing prior information; it outperforms the semi-supervised method SSGAAE, proving that the graph attention mechanism can dynamically assign different weights to each node, and demonstrating the importance of the optimized modular metric for node representation.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting overlapping vehicle groups based on a graph attention autoencoder, characterized in that, Includes the following steps: Step 1: Input the network topology and node attribute information into the encoder part of the graph attention autoencoder. Through the autoencoder structure, not only is the potential representation of the vehicle nodes learned, but the model is also optimized by reconstructing the loss so that it can capture the key structural features. Step 2: The encoder's graph attention mechanism encodes the input network topology and node attribute information. After processing by the attention mechanism, it is represented as multiple different hidden layer embeddings. These different embeddings capture the complex relationships and attribute differences between nodes. Then, a nonlinear transformation is performed through an activation function to generate new node embeddings. Step 3: The decoder combines the new node embedding representation with the original embedding representation, restores the network topology and node attribute information through the decoding process, and compares it with the original topology and attribute information to evaluate the quality of reconstruction. Step 4: Enhance the optimization constraints of the modules through modularity optimization, further adjust the node embedding, ensure the rationality of the community structure, use the M / M / 1 queuing model, and set q. ij ≥0 and The optimization is performed based on constraints, and the final output is a modularity optimization enhancement loss L. M To measure the rationality of community testing; Step 5: Perform semi-supervised clustering using prior information, and reconstruct the loss L based on the graph. R Modularity optimization enhances loss L M Semi-supervised clustering loss L S Optimization is performed to obtain node-based embedded representations for generating community detection results.

2. The vehicle overlap group detection method based on graph attention autoencoder according to claim 1, characterized in that, The specific steps of step 2 are as follows: Step 2.1: In the multi-head attention mechanism, l independent attention heads are introduced, each with its own linear transformation matrix W. (l) Used to extract the original node features h i The transformation is performed on the feature vector of each node separately, and the formula is as follows: in It is the new feature representation of node i in the l-th head; Step 2.2, for each node v i and v j Calculate an attention score, representing node v j For node v i The importance of this score is calculated based on the feature representation after linear transformation, using the following formula: Where a (l) is the attention weight vector of the l-th head, with a length of 2F', used to map the concatenated vector to a scalar, where || represents the concatenation operation, F' is the dimension of the new feature space of each head, and LeakyReLU is the activation function; Step 2.3: Use the softmax function to normalize the original attention scores of all neighboring nodes to obtain the final attention coefficients. This process ensures that the sum of the attention coefficients of all neighboring nodes is 1: Step 2.4: For each attention head l, use the attention coefficient. We perform a weighted summation of the features of neighboring nodes to obtain the feature output of each head l. Step 2.5: Average the output features of all heads to obtain the final node representation:

3. The vehicle overlap group detection method based on graph attention autoencoder according to claim 1, characterized in that, Step 3 includes the following steps: Step 3.1: Reconstruct the adjacency matrix A using the latent representation Z of the nodes, where A is the predicted adjacency matrix. When using a multi-head attention mechanism, the latent representation Z... i Output features from multiple attention heads The aggregated features are averaged across multiple heads, and the latent representation of each node is: Step 3.2, the latent representation z of the node i Still using an F'-dimensional vector, the decoder formula remains unchanged, and the calculation formula is as follows: in z is the existence of the edge between node i and node j predicted by the decoder. i and z j These are the latent representations of nodes i and j, respectively, and σ is the activation function, typically the sigmoid function. Step 3.3: The final result of the decoder is used as the reconstructed node attribute value. The loss function for reconstructing the node attribute value is defined as Equation (9). Where X i It is the original binary attribute of the i-th node. It is the probability that the i-th node attribute predicted by the model is 1. This loss function is to calculate the binary cross-entropy of the binary attribute reconstruction of each node, and optimize the model’s reconstruction effect of node attributes by minimizing the loss. Step 3.4, the loss function for adjacency matrix prediction is defined as equation (10): Where A ij Indicates the elements in the original adjacency matrix, whether there is an edge between node i and node j, if A ij =1, indicating that there is an edge between node i and node j; if A ij =0 indicates that there are no edges. These are elements in the adjacency matrix predicted by the model, representing the probability that there is an edge between node i and node j as predicted by the model. Step 3.5: The final reconstruction loss function is formed by combining the reconstruction loss of topology and node attributes; L R =αL attr +βL top (11); where α and β are hyperparameters used to balance the effects of node attribute reconstruction loss and adjacency matrix prediction loss.

4. The vehicle overlap group detection method based on graph attention autoencoder according to claim 1, characterized in that, Step 4 includes the following steps: Step 4.1: The final node representation matrix H output by the encoder is further refined through modularity optimization. Modularity is an indicator of the strength of the network community structure, defined as: Where d i and d j H represents the degree of node i and node j. i and H j The attribute vectors representing nodes i and j. This represents the attribute similarity between nodes i and j; Step 4.2: Add a Markov queuing system to the modularity maximization module to create a modularity optimization enhancement module. In the Markov queuing system, the state transitions between nodes are represented by the state transition matrix P. Define the enhanced modularity optimization objective as: Where P ij A represents the probability of node i transitioning to node j. ij P ij This means that we consider not only whether there are direct edges between nodes, but also the strength of dynamic interactions between nodes; Step 4.3: To optimize this enhanced modularity, the corresponding loss function is defined as follows: L M =-Q M (14)。 5. The vehicle overlap group detection method based on graph attention autoencoder according to claim 1, characterized in that, Step 5 includes the following steps: Step 5.1: By combining the known node label information Y with the semi-supervised clustering module, the community detection results are finally obtained. The semi-supervised strategy maintains the consistency of known labels while simultaneously mining the potential community affiliations of unlabeled nodes, making the overall community partitioning more reasonable. Therefore, the cross-entropy loss is used as the semi-supervised loss L. S : in Y is a collection of labeled data. ij Indicates whether the labeled node i belongs to the true label of node j; Step 5.2: To obtain a more accurate community member representation matrix H, the graph attention autoencoder, modularity optimization enhancement, and semi-supervised clustering module are integrated into a whole to obtain more accurate community detection results. The final objective function is defined in equation (16). L=L R +γL S -λL M (16); Where γ and λ are hyperparameters representing different loss weights, used to adjust the contribution of the corresponding modules, L R L M and L S Let represent the losses of the graph attention autoencoder, modularity optimization enhancement, and self-trained clustering module, respectively; Step 5.3: Train the model based on the final loss function L to obtain the final community detection result.

Images