Unmanned vehicle cooperative control method based on gating-hypergraph neural network

Abstract

The invention discloses an unmanned vehicle cooperative control method based on a gating-hypergraph neural network, and belongs to the technical field of Internet of Vehicles and automatic driving, and the method comprises the steps: obtaining the real-time state information of a plurality of vehicle clusters in a traffic scene; constructing a hypergraph representation model based on the real-time state information, obtaining a node feature matrix, a hyperedge feature matrix and an incidence matrix based on the hypergraph representation model, and inputting the node feature matrix, the hyperedge feature matrix and the incidence matrix into a pre-constructed gating-hypergraph neural network to obtain an action value and a weighted global feature; and obtaining a joint action of the unmanned vehicle cluster based on the action value and the weighted global feature. According to the method, the high-order interaction relation between the vehicles is modeled through the hypergraph structure, the binary limitation of a traditional graph model is broken through, and the multi-element interaction modeling capability of a complex mixed traffic scene is remarkably improved. In combination with a gating mechanism, multi-head attention features are dynamically screened dimension by dimension, noise is effectively filtered, key decision variable recognition is enhanced, and more robust and stable cooperative control is achieved under observation limitation or environmental disturbance.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of Internet of Vehicles and automatic driving, and particularly relates to a cooperative control method for unmanned vehicles based on a gated-hypergraph neural network. BACKGROUND

[0002] In an intelligent transportation system, the autonomous decision-making capability of unmanned vehicles directly affects the safety and operational efficiency of the overall traffic. With the popularization of mixed traffic environments, the differences in intelligent level, communication capability and behavior pattern of different types of vehicles continue to intensify, making it difficult for traditional methods that rely on rules and static strategies to cope with changing and uncertain scenarios. Therefore, cooperative decision-making methods for unmanned vehicle systems have gradually become the focus of invention, which model the interaction between vehicles to guide the global behavior optimization of unmanned vehicles under local observation conditions, thereby effectively improving the adaptability and cooperation efficiency of unmanned vehicles in complex environments.

[0003] In a mixed autonomous traffic system, unmanned vehicles face problems such as dense interaction of heterogeneous participants and information asymmetry, and the decision-making process highly depends on the understanding and modeling of the relationship between multiple vehicles in the traffic environment. Current unmanned vehicle decision-making methods generally have insufficient interaction modeling capabilities when dealing with complex traffic scenarios, making it difficult to capture high-order cooperative relationships between vehicle clusters. At the same time, in the face of information asymmetry and target differences caused by heterogeneous vehicle clusters, existing methods lack effective structural perception mechanisms and are difficult to achieve fine-grained strategy adjustment. SUMMARY

[0004] To solve the above technical problems, the application provides a cooperative control method for unmanned vehicles based on a gated-hypergraph neural network to solve the problems existing in the prior art.

[0005] To achieve the above purpose, the application provides a cooperative control method for unmanned vehicles based on a gated-hypergraph neural network, comprising:

[0006] obtaining real-time state information of a plurality of vehicle clusters in a traffic scenario;

[0007] constructing a hypergraph representation model based on the real-time state information, obtaining a node feature matrix, a hyperedge feature matrix and a correlation matrix based on the hypergraph representation model;

[0008] constructing an initial neural network based on a multi-head attention mechanism, introducing a gating mechanism in the initial neural network to obtain a gated-hypergraph neural network; the gated-hypergraph neural network comprises a local network, a global network and a hybrid network;

[0009] inputting the node feature matrix, the hyperedge feature matrix and the correlation matrix into the gated-hypergraph neural network to obtain local features and weighted global features;

[0010] obtaining a joint action of the unmanned vehicle cluster based on the local feature and the weighted global feature.

[0011] Optionally, the set of hyperedge in the hypergraph representation model includes three types of interactions, same-class interaction edges, different-class interaction edges, and mixed interaction edges.

[0012] Optionally, the local network includes two layers of hypergraph convolution structures and a gating mechanism; the two layers of hypergraph convolution structures include an attention-free hypergraph convolution layer and a multi-head attention hypergraph convolution layer.

[0013] Optionally, after the node feature matrix, the hyperedge feature matrix, and the association matrix are input into the local network, initial adjacency features are extracted by the attention-free hypergraph convolution layer, the initial adjacency features are processed by the multi-head attention hypergraph convolution layer to obtain multi-head splicing features, the multi-head splicing features are weighted processed by the gating mechanism, and then the local features are output by mapping the two layers of fully connected layers to the dimension of the action space; the gating mechanism includes a linear mapping layer and a Sigmoid activation function.

[0014] Optionally, the global network includes two layers of hypergraph convolution structures, a gating mechanism, a global average pooling layer, and a global attention layer; the two layers of hypergraph convolution structures include an attention-free hypergraph convolution layer and a multi-head attention hypergraph convolution layer.

[0015] Optionally, after the node feature matrix, the hyperedge feature matrix, and the association matrix are input into the global network, initial adjacency features are extracted by the attention-free hypergraph convolution layer, the initial adjacency features are processed by the multi-head attention hypergraph convolution layer to obtain multi-head splicing features, the multi-head splicing features are weighted processed by the gating mechanism, and then the weighted global features are obtained by processing the global average pooling layer and inputting the output of the global average pooling layer into the global attention layer.

[0016] Optionally, after the input local features and weighted global features are mapped to a high-dimensional hidden space by a feature mapping layer, the local features and the weighted global features are processed by a hypergraph convolution layer, a multi-head hypergraph attention layer, and a gated multi-head attention mechanism in sequence; after the processed local features and weighted global features are aggregated and reduced in dimension by a global average pooling layer, the global feature representation is obtained by processing the global attention layer, the global feature representation is mapped to the output space of the joint action value function by two layers of fully connected layers to obtain a decision signal, and the joint action of the unmanned vehicle cluster is obtained based on the decision signal.

[0017] Optionally, the attention-free hypergraph convolution layer, the multi-head attention hypergraph convolution layer, and the gating mechanism are connected by residual connections.

[0018] The application provides a computer device, comprising a memory, a processor to store a computer program on the memory and run the computer program on the processor, and the processor executes the computer program to realize the steps of the above method.

[0019] The application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the above method.

[0020] Compared with the prior art, the application has the following advantages and technical effects:

[0021] 1) The application proposes a hypergraph structure to model the relationship of traffic participants, which can depict the high-order interaction relationship among multiple vehicles, breaks through the limitation of traditional graph model that can only express pairwise relationship, and significantly improves the modeling ability of multi-element interaction mode in complex traffic scenes, and is especially suitable for mixed traffic environment with cluster coordination behavior and heterogeneous communication structure.

[0022] 2) The application combines a gating mechanism to dynamically filter the features output by the multi-head attention in each dimension, effectively filters redundant and noisy information, improves the identification ability of the model to key decision variables, and shows stronger robustness and stability under the condition that vehicle observation is limited or environmental disturbance is large. BRIEF DESCRIPTION OF DRAWINGS

[0023] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application and their explanations are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0024] Figure 1 The figure is a schematic diagram of the gating-hypergraph neural network of the embodiment of the application;

[0025] Figure 2 The figure is a schematic diagram of the hypergraph and the associated matrix of the embodiment of the application;

[0026] Figure 3 The figure is a schematic diagram of the inter-cluster association matrix construction method of the embodiment of the application;

[0027] Figure 4 The figure is a schematic diagram of the inter-cluster hyperedge feature matrix construction method of the embodiment of the application. DETAILED DESCRIPTION

[0028] It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict. The application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0029] It is noted that the steps shown in the flowcharts of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowcharts, in some cases the steps shown or described can be executed in an order different from that shown here.

[0030] Embodiment One

[0031] As a deep learning method capable of depicting structured relationships, Graph Neural Network (GNN) is suitable for expressing the dynamic connections between various vehicles in the traffic system. By introducing graph structure representation, the model can fully integrate the state information of neighboring vehicles, realize joint perception of local and global environment, and provide more accurate state representation for the decision-making process, thereby enhancing the reaction ability and coordination level of unmanned vehicles in complex traffic situations.

[0032] In actual traffic environment, the interaction relationship between vehicles has obvious structure and dynamics, and graph neural network can capture the coupling characteristics between individual behavior and group mode by modeling nodes and their connection relationships. In the decision-making system of unmanned vehicles, the introduction of graph structure not only improves the perception ability of the model to the local traffic state, but also provides structured prior information for strategy generation. Through dynamic updating of node representation and adaptive adjustment of edge weight, the system can more accurately reflect the influence degree between different vehicles, thereby guiding the unmanned vehicles to make more reasonable action selection and strategy execution in the scene with intensive interaction.

[0033] As shown in Figure 1 The embodiment provides an unmanned vehicle cooperative control method based on a gated-hypergraph neural network, which includes:

[0034] Obtaining real-time state information of a plurality of vehicle clusters in a traffic scene; constructing a hypergraph representation model based on the real-time state information, obtaining a node feature matrix, a hyperedge feature matrix, and a correlation matrix based on the hypergraph representation model; constructing an initial neural network based on a multi-head attention mechanism, introducing a gating mechanism in the initial neural network to obtain a gated-hypergraph neural network; the gated-hypergraph neural network includes a local network, a global network, and a hybrid network; inputting the node feature matrix, the hyperedge feature matrix, and the correlation matrix into the gated-hypergraph neural network to obtain local features and weighted global features; obtaining joint actions of the unmanned vehicle clusters based on the local features and the weighted global features.

[0035] Further, by constructing a hypergraph representation at the cluster level, the multi-element relationships between different vehicles are modeled in the form of hyperedges, and a gating mechanism is introduced to filter key feature information. The local network is used to capture the behavior dependence within the cluster, and the global network integrates cross-cluster interaction information and aggregates features under the guidance of multi-head attention mechanism. Finally, the hybrid network outputs a joint action value function to assign optimal control instructions to each type of unmanned vehicle, achieving efficient and stable multi-vehicle cooperative strategy generation.

[0036] ;

[0037] wherein, is a joint action value function, representing the cumulative discounted reward expected to be obtained by the system when performing joint action u under the overall state s, is a state set composed of all agent states, is a joint action set of all agent actions, is a parameter of the hybrid network, denotes a set of all parameters. The joint action value function takes as input the local action value (Q value) of each agent obtained by the local network and the global feature provided by the global network, and realizes nonlinear aggregation and mapping through the hybrid network.

[0038] The hybrid network (Mixing Network) structure is designed to fuse the feature information output by the local network (Local Network) and the global network (Global Network) to generate a joint action value function (Joint Action-Value Function) suitable for multi-CAV cluster cooperative control. Specifically, the hybrid network is implemented through a hypergraph neural network (Hypergraph Neural Network) structure, which includes feature mapping layers, hypergraph convolution layers, multi-head hypergraph attention layers, gated multi-head attention mechanism layers, global average pooling layers, and global attention layers, among other core modules. The specific design process is as follows: First, the local features and weighted global features are mapped to a high-dimensional hidden space through the feature mapping layer; then the hypergraph convolution layer captures the high-order interaction relationship between nodes and hyperedges; the multi-head hypergraph attention layer further learns the weight information of multi-scale interactions between different clusters; then the gated multi-head attention mechanism is used for dynamic filtering and enhancement of features; next, the global average pooling layer is used to aggregate and reduce the dimension of the above features to obtain the overall scene feature representation; finally, the global attention layer obtains a fine global feature representation, and further maps this global feature to the joint action value function output space through two fully connected layers to generate the final decision signal.

[0039] The local network is responsible for processing and outputting the feature information and local action value function (local Q-value) within each CAV cluster, capturing the local control strategy of a single cluster. The global network is used to capture the global state features of the entire traffic environment to provide global contextual information support. The hybrid network plays a crucial role in bridging the gap between the local and global networks, effectively fusing the local Q-values ​​of each cluster output by the local network with the global feature information provided by the global network, ultimately outputting a joint action value function applicable to the entire unmanned vehicle system. Therefore, the hybrid network plays a pivotal role in the network structure and information flow path, and is a key network structure for achieving efficient fusion of local and global decisions. The hybrid network structure is shown in Table 1.

[0040] Table 1

[0041] in, For the first The number of CAVs within a cluster, here is .

[0042] The detailed parameters of the gated attention layer network set in this embodiment are shown in Table 2.

[0043] Table 2

[0044] For mixed traffic scenarios, the hyperedge feature matrix, node feature matrix, and correlation matrix are first obtained through cluster-based hypergraph representation, including both local and global (local / global hyperedge feature matrix, local / global node feature matrix, and local / global correlation matrix). These are then input into the local and global networks, respectively. For the local network, node features are first mapped to the hidden layer feature space; subsequently, the processed features are processed through hypergraph convolution; the processed data is then guided by a multi-head hypergraph attention mechanism, utilizing hyperedge features to guide information propagation; next, a gated multi-head attention mechanism is used to process the features; finally, two fully connected layers map the features to the action space dimension and output local features (joint action value function (Q-value)). The global network differs from the local network in that after weighting the features using a gated multi-head attention mechanism, a global average pooling layer is used for further processing, followed by a global attention layer to obtain weighted global features. Furthermore, residual connections are used before and after hypergraph convolution, hypergraph attention, and the gating mechanism throughout the entire process. The outputs of the two networks can eventually be fed into algorithms such as MAPPO (Multi-Agent Proximal Policy Optimization), QMIX, and COMA (Counterfactual Multi-Agent Policy Gradients) to generate joint actions for each CAV cluster.

[0045] The basic principle of hypergraphs:

[0046] Hypergraph can be used To indicate, among which, Indicates a hypergraph A set of nodes, and The diagram shows... A hypergraph is a set of hyperedges. Unlike traditional graph representations, edges in a hypergraph can connect any number of nodes. In other words, each hyperedge... Both are sets of nodes A subset, namely By using hyperedges to represent the multivariate relationships between multiple nodes, hypergraphs can more naturally represent complex relationships in applications such as mixed transportation, social groups, and recommendation systems.

[0047] The incidence matrix and the hyperedge feature matrix can be used to mathematically represent a hypergraph.

[0048] Correlation Matrix It can be used to represent the connection relationship between a node and a hyperedge, where, The number of nodes in the hypergraph. Let be the number of hyperedges in the hypergraph. The construction rules for the incidence matrix are as follows: Figure 2 As shown, for node and super edge ,have:

[0049] .

[0050] The hyperedge feature matrix can be used to represent the features of the hyperedge. For a given hyperedge... There are 10 superedges, and the feature dimension of each superedge is 0. The hypergraph whose hyperedge feature matrix It can be represented as:

[0051] ;

[0052] in, For super-edge eigenvectors.

[0053] Cluster-based hypergraph representation:

[0054] The cluster-based hypergraph representation method in this embodiment is an extension of the cluster-based graph representation method. Assuming that in... In the moment-to-moment scene A cluster of CAVs (Connected and Automated Vehicles), for the cluster It contains Vehicles of CAVs, and these CAVs observed a total of The remaining vehicles. Then the entire transportation system is... Hypergraph representation of time It can be jointly characterized by the observation hypergraph of all clusters at that moment, that is:

[0055] ;

[0056] in, Indicates at time No. The observation hypergraph structure of a CAV cluster The features are stored in the node feature matrix and hyperedge feature matrix In the cluster, the topological connection relationship between vehicle nodes and hyperedges is specifically manifested as the attribution and connection structure between nodes and hyperedges, which is determined by the correlation matrix. express, Let be the correlation matrix of the i-th cluster at time t.

[0057] These three concepts will be introduced one by one.

[0058] First, for the node feature matrix (In the hypergraph structure at time t, the first...) The system constructs a feature representation matrix for each node. Each row of this matrix typically corresponds to a node (e.g., an autonomous vehicle or a human-driven vehicle), and each column represents the specific characteristics of the node at that moment (e.g., vehicle speed, position, acceleration, safety factor, etc.). Secondly, for the association matrix, based on the vehicle types and interaction properties in the TMA (Traffic Mixed Autonomy System), this embodiment defines three hyperedge categories to accurately characterize the conflicts and potential cooperation relationships in different vehicle clusters.

[0059] Similar interaction edge Within the same CAV cluster, all vehicles possess the same perception capabilities and control strategies, thus exhibiting a close collaborative mode. Therefore, Connecting CAV nodes of the same type to the same superedge highlights information sharing and control interaction during collaborative driving within the cluster, thereby capturing the cooperative intent of resource allocation within the cluster.

[0060] Heterogeneous interaction edge Different CAV clusters often have different task objectives and perception accuracy, and can influence each other in the same environment. To address cross-cluster interactions between CAVs, this embodiment defines combinations containing different types of CAV nodes as heterogeneous interaction edges in the association matrix. It connects two or more CAV nodes from different clusters, thus representing the conflicts between them caused by path planning and communication protocols.

[0061] Hybrid Interactive Edge In the TMA system, there is a high degree of interaction between connected autonomous vehicles (CAVs) and human vehicles (HVs). Therefore, this invention will combine nodes containing at least one CAV and one or more HVs, and use a hybrid interaction edge. They are concatenated to map the interaction between CAV and HV in the association matrix.

[0062] Based on the above analysis, the construction method of the inter-cluster association matrix is ​​as follows: Figure 3 As shown.

[0063] The inter-cluster association matrix can be constructed by concatenating all intra-cluster association matrices, i.e.:

[0064] ;

[0065] No. The shape of the intra-cluster association matrix within each cluster is: , This represents the total number of vehicle clusters in the scene.

[0066] For the hyperedge eigenma matrix The local hyperedge feature matrix at time t is , can be represented as:

[0067] ;

[0068] in, for Always Beyond the Edge The eigenvectors of can be represented as:

[0069] .

[0070] like Figure 4 As shown, the hyperedge feature matrix between clusters The construction of the global correlation matrix is ​​similar to that of the global correlation matrix; it can be formed by concatenating all the local hyperedge feature matrices, i.e.:

[0071] ;

[0072] where, represents the dimension of the feature vector of the hyperedge in the hypergraph model.

[0073] The basic principle of the gating mechanism:

[0074] The gating mechanism is a technique used to dynamically control the flow of information in neural networks. Its core lies in controlling the flow of information through one or more gates (Gates), each of which is a weight between 0 and 1, representing the degree of information passing or being closed. The gating mechanism mainly consists of three parts: 1. Input gate, used to control the degree of influence of the current input; 2. Forget gate, used to control the degree of influence of the previous state, deciding how much past memory needs to be forgotten; 3. Output gate, deciding which current state information needs to be output.

[0075] Assume the input of the model is , the state of the previous time is , and the calculation of each gate is based on the linear transformation of the input and the state of the previous time plus the bias, and then the non-linear transformation is obtained through the activation function.

[0076] The calculation method of the input gate is:

[0077] ;

[0078] where, is the activation function, is the weight matrix of the input gate, is the weight matrix of the previous state to the input gate, is the bias term, represents the batch size, and are the feature dimension of the input and the dimension of the hidden state, respectively.

[0079] The calculation method of the forget gate is:

[0080] ;

[0081] where, is the weight matrix of the forget gate, is the weight matrix of the previous state to the forget gate, is the bias term.

[0082] The calculation method of the output gate is:

[0083] ;

[0084] where, Woutis the weight matrix of the output gate, Wprevis the weight matrix of the previous time step state to the output gate, bias is the bias term.

[0085] Next, the candidate state is computed by combining the current input with the previous time step state:

[0086] ;

[0087] Then, the current state is updated:

[0088] ;

[0089] where, denotes element-wise multiplication. Finally, the final output is obtained by a linear transformation, computed as:

[0090] .

[0091] where, Wfinalis the weight matrix of the final fully connected layer in the hybrid network, which specifically serves to linearly map the high-dimensional hidden state feature vector obtained after multiple layers of nonlinear processing in the network to the output space, generating action values or control signals available for decision-making. bias is the bias vector of the final fully connected layer, which serves to provide a proper offset in the above linear mapping process to improve the flexibility and expressiveness of the output, ensuring that the final output value can accurately fit the required action space.

[0092] By introducing a gating mechanism at each time step, the network can dynamically adjust the flow of information, thereby selectively remembering or forgetting past information.

[0093] Basic principles of hypergraph neural networks:

[0094] Hypergraph Neural Network (HGNN) is an extension of GNN based on hypergraph structure, used to handle complex high-order structures. Common models include Hypergraph Convolutional Neural Network (HGCN) and Hypergraph Attention (HGAT).

[0095] HGCN is an extension method of traditional GCN, which is specially used to process hypergraph structure data. Unlike traditional GCN, hypergraph convolution aggregates the information of multiple nodes through hyperedge. The process is as follows:

[0096] ;

[0097] wherein, is the vertex metric matrix, which describes the node connected to the number of hyperedges, that is:

[0098] ;

[0099] is the hyperedge metric matrix, which describes the hyperedge connected to the number of nodes, that is:

[0100] ;

[0101] is the hyperedge weight matrix, is the node feature matrix of the layer, is the and weight matrix between and is a nonlinear activation function.

[0102] HGAT enhances the expression of hypergraph convolution by introducing an attention mechanism. This method dynamically adjusts the weight of information propagation by learning the relationship between different nodes and hyperedges. For node and the hyperedge connected to it, the calculation method of its attention coefficient is as follows:

[0103] ;

[0104] wherein, is the neighbor node of node , is the feature vector of node , is the feature vector of hyperedge , is the learnable attention weight. Using the calculated attention coefficient , the information of the neighbor node can be weighted and summed, so as to update the information of node , that is:

[0105] ;

[0106] wherein, is the learnable weight matrix, According to The node is updated based on its attention coefficient and information about surrounding nodes. Information.

[0107] Multi-head HGAT can further enhance the model's expressive power. It allows the model to simultaneously decelerate the attention coefficient from multiple angles and integrate the results, thereby capturing richer relational information. Its calculation formula is as follows:

[0108] ;

[0109] in, For the number of attention heads, and The first Attention coefficients and learnable weight matrices for each attention head. For nodes The set of neighboring nodes in a hypergraph structure.

[0110] The following are embodiments of this application in highways and roundabouts. First, in traffic scenarios, hypergraphs can be used... To indicate, among which, Indicates a hypergraph A set of nodes, and The diagram shows... A set of edges. Adaptation matrix. It can be used to represent the connection relationship between a node and a hyperedge, where, The number of nodes in the hypergraph. Let be the number of hyperedges in the hypergraph. For a given hypergraph, _____. There are 10 superedges, and the feature dimension of each superedge is 0. The hypergraph whose hyperedge feature matrix It can be represented as:

[0111] ;

[0112] in, For super-edge eigenvectors.

[0113] The shape of the intra-cluster association matrix is:

[0114] ;

[0115] in, This represents the total number of vehicles (total number of nodes) in the scene. Represents cluster The total number of internal CAVs. In the above formula, 1 describes the cluster. Internal interaction, the first Cluster of How do each CAV interact with the rest of the cluster, and the second... Describes cluster of The interaction between CAVs and HVs.

[0116] The global association matrix can be constructed by concatenating all the association matrices within the cluster, that is:

[0117] ;

[0118] use Figure 3 The correlation matrix obtained by the construction method shown is:

[0119] .

[0120] For the edge feature matrix Local hyperedge feature matrix at time step It can be represented as:

[0121] ;

[0122] in, for Always Beyond the Edge The eigenvectors of can be represented as:

[0123] ;

[0124] in, Indicates the superedge The average velocity of all nodes. Indicates the superedge The average acceleration of all nodes. It can be represented as:

[0125] ;

[0126] Inter-cluster hyperedge feature matrix The construction of the global correlation matrix is ​​similar to that of the global correlation matrix; it can be formed by concatenating all the local hyperedge feature matrices, i.e.:

[0127] ;

[0128] The combination of the current input and the state at the previous time step:

[0129] ;

[0130] Then, update the current state:

[0131] ;

[0132] get the final output :

[0133] ;

[0134] The state information of the vehicle nodes is aggregated based on a gated-hypergraph neural network, which includes two layers of hypergraph convolution structures, the first layer being an attention-free hypergraph convolution used for extracting initial adjacency features, and the second layer being a multi-head attention hypergraph convolution used for capturing high-order interaction relationships between nodes to obtain a multi-head spliced feature vector output; after the second layer convolution output, a gating mechanism is introduced to dynamically weight the spliced features, the gating mechanism including a gating network including a linear mapping layer and a Sigmoid activation function, used for compressing and nonlinearly mapping the spliced feature vectors to output gating weights corresponding to each channel; the gating weights are copied to match the dimension of the original features, and are multiplied element-wise with the original multi-head features, so as to realize the importance weighting of each channel, filter out redundant information, and retain key features.

[0135] The state of each unmanned vehicle individual is modeled through a local policy network to output an action value function; a global fusion network is used to integrate the local Q value and the structural features to output a joint action value; a robust Q-Nexus engine (RQNE) algorithm is used to train the above network, and the system determines the optimal action according to the joint Q value at each time step, and the control commands are distributed to the vehicle controllers by a centralized control module to drive the actual behavior; the experience quadruplets generated in the interaction process are stored in an experience replay buffer for offline updating of the policy network and the fusion network, and the training continues until the number of rounds or the performance convergence condition is met.

[0136] The embodiment provides a computer device, comprising a memory, a processor to store a computer program on the memory and run the computer program on the processor, and the processor executes the computer program to realize the steps of the above method.

[0137] The embodiment provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the above method.

[0138] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any changes or replacements within the technical range disclosed by the present application can be easily thought of by those skilled in the art, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for collaborative control of unmanned vehicles based on a gated-hypergraph neural network, characterized in that, The method comprises the following steps: obtaining real-time state information of a plurality of vehicle clusters in a traffic scene; constructing a hypergraph representation model based on the real-time state information, obtaining a node feature matrix, a hyperedge feature matrix and an association matrix based on the hypergraph representation model; constructing an initial neural network based on a multi-head attention mechanism, introducing a gating mechanism into the initial neural network to obtain a gated-hypergraph neural network; the gated-hypergraph neural network comprises a local network, a global network and a hybrid network; inputting the node feature matrix, the hyperedge feature matrix and the association matrix into the gated-hypergraph neural network to obtain local features and weighted global features; obtaining joint actions of an unmanned vehicle cluster based on the local features and the weighted global features.

2. The unmanned vehicle cooperative control method based on the gated-hypergraph neural network according to claim 1, wherein the node set and the hyperedge set in the hypergraph representation model comprise three types of interactions, same-type interaction edges, different-type interaction edges and mixed interaction edges.

3. The unmanned vehicle cooperative control method based on the gated-hypergraph neural network according to claim 1, wherein the local network comprises two layers of hypergraph convolution structures and a gating mechanism; the two layers of hypergraph convolution structures comprise an attention-free hypergraph convolution layer and a multi-head attention hypergraph convolution layer.

4. The unmanned vehicle cooperative control method based on the gated-hypergraph neural network according to claim 3, wherein after the node feature matrix, the hyperedge feature matrix and the association matrix are input into the local network, initial adjacency features are extracted by the attention-free hypergraph convolution layer, the initial adjacency features are processed by the multi-head attention hypergraph convolution layer to obtain multi-head splicing features, the multi-head splicing features are weighted by the gating mechanism, and then the two layers of fully connected layers are used to map the weighted multi-head splicing features to the dimension of the action space and output the local features; wherein the gating mechanism comprises a linear mapping layer and a Sigmoid activation function.

5. The unmanned vehicle cooperative control method based on the gated-hypergraph neural network according to claim 4, wherein the global network comprises two layers of hypergraph convolution structures, a gating mechanism, a global average pooling layer and a global attention layer; the two layers of hypergraph convolution structures comprise an attention-free hypergraph convolution layer and a multi-head attention hypergraph convolution layer.

6. The unmanned vehicle cooperative control method based on the gated-hypergraph neural network according to claim 5, wherein after the node feature matrix, the hyperedge feature matrix and the association matrix are input into the global network, initial adjacency features are extracted by the attention-free hypergraph convolution layer, the initial adjacency features are processed by the multi-head attention hypergraph convolution layer to obtain multi-head splicing features, the multi-head splicing features are weighted by the gating mechanism, and then the global average pooling layer is used to process the weighted multi-head splicing features, the output of the global average pooling layer is input into the global attention layer to obtain weighted global features.

7. The unmanned vehicle cooperative control method based on the gated-hypergraph neural network according to claim 6, wherein The mixed network maps the input local features and weighted global features to a high-dimensional hidden space through a feature mapping layer, and then sequentially processes the local features and the weighted global features through a hypergraph convolution layer, a multi-head hypergraph attention layer and a gated multi-head attention mechanism; after the processed local features and the weighted global features are aggregated and reduced in dimension through a global average pooling layer, a global attention layer is used to obtain a global feature representation, which is mapped to a joint action value function output space through two fully connected layers to obtain a decision signal, and a joint action of the unmanned vehicle cluster is obtained based on the decision signal.

8. The unmanned vehicle cooperative control method based on the gated-hypergraph neural network according to claim 7, characterized in that, The attention-free hypergraph convolution layer, the multi-head attention hypergraph convolution layer and the gating mechanism are connected through residual connections.

9. A computer apparatus comprising: A memory and a processor for storing a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method according to any one of claims 1-8.

Images