A wireless relay communication method and device based on graph neural network path optimization

Abstract

The present invention discloses a wireless relay communication method and device based on graph neural network path optimization, which relates to the field of wireless relay communication technology. The method comprises: generating an embedding matrix of a node in a wireless relay communication system; training an encoder-decoder architecture based on the embedding matrix of the node to obtain a trained encoder-decoder architecture; inputting the embedding matrix of the node into the trained encoder-decoder architecture to obtain a path optimization result of the relay node in the wireless relay communication system. The present invention solves the problem of poor information transmission performance in multi-node wireless relay communication methods due to the mobility of unmanned systems and the instability of wireless communications. The method utilizes a graph neural network combined with the power of fixed nodes and the relative distance between fixed nodes and relay nodes to autonomously generate the mobile path of the unmanned system. This is a more generalized wireless relay communication method that is not restricted by distance and the number of fixed nodes.

Description

Technical Field

[0001] The present invention relates to the field of wireless relay communication technology, and in particular to a wireless relay communication method and device based on graph neural network path optimization. Background Art

[0002] With the advancement of science and technology, communications and network technologies have developed rapidly. Unmanned systems, thanks to their low cost and high flexibility, have been widely applied and developed in the communications field. Compared to traditional terrestrial communications, unmanned system-assisted communication systems have demonstrated excellent results due to their flexibility, ease of deployment, and high link capacity. When communication between base stations and devices, or between devices, is impossible due to distance or obstacles, unmanned systems can be used as relay nodes to achieve effective data transmission. This type of communication is called relay communication and is one of the main applications of unmanned systems in the communications field.

[0003] Compared to stationary relays like satellite relays, wireless relays assisted by unmanned systems can move quickly and adjust their positions, better adapting to channel conditions and achieving superior communication performance. However, the mobility of unmanned systems and the instability of wireless communications are also major contributors to the instability of wireless relay communication. To ensure a smooth information chain in wireless relay communication, the communication network structure can be modified by optimizing the movement paths of unmanned systems, thereby improving communication performance.

[0004] Path planning is a means of achieving high-performance wireless relay communication for unmanned systems. Appropriate path planning can significantly improve the performance of wireless relay communication systems, reduce energy consumption, and leverage external environmental advantages to increase communication throughput, reduce device outages, and avoid issues like information transmission distortion and lag. Developing an appropriate path plan is a key challenge in implementing wireless relay communication for unmanned systems. However, existing research models for wireless relay communication in unmanned systems are overly simplistic and limited in their applicability to a limited number of nodes. Therefore, it is necessary to consider more generalized wireless relay communication methods based on path planning.

[0005] With the development of deep learning, graph neural networks (GNNs) have been widely used to process unstructured data, such as sequences or networks. Graph neural networks are a deep learning method that aggregates information about surrounding nodes and, by increasing the number of neural network layers, aggregates higher-order information. Path planning is essentially a sequence-solving problem. Applying GNNs to path optimization for unmanned systems during wireless relay communications effectively aggregates each node's own information reception and its distance from surrounding nodes into its own embedding matrix. Therefore, designing a GNN-based path optimization method that can provide effective technical support for solving wireless relay communication transmission issues is a pressing issue facing existing technologies. Summary of the Invention

[0006] The present invention aims to solve the problem of how to design a path optimization method based on graph neural network, which can provide effective technical support for solving wireless relay communication transmission problems.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In one aspect, the present invention provides a wireless relay communication method based on graph neural network path optimization, which is implemented by an electronic device and includes:

[0009] S1. Generate an embedding matrix of nodes in a wireless relay communication system.

[0010] S2. Based on the embedding matrix of the node, the encoder-decoder architecture is trained to obtain a trained encoder-decoder architecture.

[0011] S3. Input the node embedding matrix into the trained encoder-decoder architecture to obtain the path optimization result of the relay node in the wireless relay communication system.

[0012] Optionally, generating an embedding matrix of a node in the wireless relay communication system in S1 includes:

[0013] S11. Acquire nodes in the wireless relay communication system; wherein the nodes include multiple fixed nodes and relay nodes.

[0014] S12. Build a graph neural network based on nodes.

[0015] S13. Acquire k+1-dimensional information of each of the multiple nodes; wherein the k+1-dimensional information includes the relative power of the node and the distance between the fixed node and the relay node.

[0016] S14. Aggregate the k+1-dimensional information of each node based on the graph neural network to generate an embedding matrix of the node in the wireless relay communication system.

[0017] Optionally, the fixed node in S11 is an energy-constrained node;

[0018] The energy constraint condition of any fixed node i among multiple fixed nodes is as follows:

[0019] t ri ×p i ≤E i (1)

[0020] Among them, t ri is the transmission time used in fixed node i; p i Refers to the power transmitted by the fixed node; E i Refers to the energy limit of a fixed node i.

[0021] Alternatively, the embedding matrix expression of the nodes in S1 is as follows (2):

[0022] H (l+1) =σ(D -1 / 2 AD 1 / 2 ·H (l) W (l) ) (2)

[0023] Among them, H (l+1) is the embedding matrix of k+1-dimensional information in the l+1th layer of the graph neural network; σ is the nonlinear activation function; is the diagonal matrix after the self-loop; is the adjacent order matrix after the self-loop; H (l) is the embedding matrix of k+1-dimensional information in the lth layer of the graph neural network; W (l) is the weight matrix of the lth layer.

[0024] Optionally, the node-based embedding matrix in S2 is used to train the encoder-decoder architecture, and the trained encoder-decoder architecture includes:

[0025] S21. Input the embedding matrix of the node into the encoder to obtain the state vector e of each unit hidden layer of the embedding matrix in the encoder. i .

[0026] S22, input the state vector of the last unit hidden layer of the encoder into the decoder, and obtain the state vector d of each unit hidden layer in the decoder j .

[0027] S23, the state vector d of each unit hidden layer in the decoder j Respectively with the state vector e of each unit hidden layer in the encoder iPerform similarity calculation to obtain the state vector with the highest similarity in the encoder, and use the node path corresponding to the state vector with the highest similarity as the output of the decoder to generate a complete path.

[0028] S24. Calculate the sum of the cross entropy loss functions of all units in the decoder, use the optimization algorithm to iteratively solve it, and update the weights until convergence to obtain the trained encoder-decoder architecture.

[0029] Optionally, the similarity calculation formula in S23 is as follows (3):

[0030]

[0031] in, is the similarity value; tanh is the nonlinear activation function; W1 and W2 are the weight matrices to be learned; m is the number of label vectors.

[0032] Optionally, the step of inputting the embedding matrix of the node into the trained encoder-decoder architecture in S3 to obtain a path optimization result of the relay node in the wireless relay communication system includes:

[0033] The node embedding matrix is ​​input into the trained encoder-decoder architecture, and the node with the highest output probability is selected in turn as the next node of the current path, thereby obtaining the path optimization result of the relay node in the wireless relay communication system.

[0034] Optionally, the performance evaluation index of the wireless relay communication system is minimization of information delay.

[0035] The information delay is calculated as follows:

[0036]

[0037] Among them, t r The time it takes for the relay node to receive data; t m is the inter-node movement time; k is the number of fixed nodes.

[0038] On the other hand, the present invention provides a wireless relay communication device based on graph neural network path optimization, which is used to implement a wireless relay communication method based on graph neural network path optimization, and the device includes:

[0039] The generation module is used to generate an embedding matrix of nodes in a wireless relay communication system.

[0040] The training module is used to train the encoder-decoder architecture based on the embedding matrix of the node to obtain the trained encoder-decoder architecture.

[0041] The output module is used to input the node embedding matrix into the trained encoder-decoder architecture to obtain the path optimization result of the relay node in the wireless relay communication system.

[0042] Optionally, a generation module is further configured to:

[0043] S11. Acquire nodes in the wireless relay communication system; wherein the nodes include multiple fixed nodes and relay nodes.

[0044] S12. Build a graph neural network based on nodes.

[0045] S13. Acquire k+1-dimensional information of each of the multiple nodes; wherein the k+1-dimensional information includes the relative power of the node and the distance between the fixed node and the relay node.

[0046] S14. Aggregate the k+1-dimensional information of each node based on the graph neural network to generate an embedding matrix of the node in the wireless relay communication system.

[0047] Optionally, the fixed node is an energy-constrained node;

[0048] The energy constraint condition of any fixed node i among multiple fixed nodes is as follows:

[0049] t ri ×p i ≤E i (1)

[0050] Among them, t ri is the transmission time used in fixed node i; p i Refers to the power transmitted by the fixed node; E i Refers to the energy limit of a fixed node i.

[0051] Optionally, the embedding matrix of the node is expressed as follows (2):

[0052] H (l+1) =σ(D -1 / 2 AD 1 / 2 ·H (l) W (l) ) (2)

[0053] Among them, H (l+1) is the embedding matrix of k+1-dimensional information in the l+1th layer of the graph neural network; σ is the nonlinear activation function; is the diagonal matrix after the self-loop; is the adjacent order matrix after the self-loop; H (l) is the embedding matrix of k+1-dimensional information in the lth layer of the graph neural network; W (l) is the weight matrix of the lth layer.

[0054] Optionally, the training module is further configured to:

[0055] S21. Input the embedding matrix of the node into the encoder to obtain the state vector e of each unit hidden layer of the embedding matrix in the encoder. i .

[0056] S22, input the state vector of the last unit hidden layer of the encoder into the decoder, and obtain the state vector d of each unit hidden layer in the decoder j .

[0057] S23, the state vector d of each unit hidden layer in the decoder j Respectively with the state vector e of each unit hidden layer in the encoder i Perform similarity calculation to obtain the state vector with the highest similarity in the encoder, and use the node path corresponding to the state vector with the highest similarity as the output of the decoder to generate a complete path.

[0058] S24. Calculate the sum of the cross entropy loss functions of all units in the decoder, use the optimization algorithm to iteratively solve it, and update the weights until convergence to obtain the trained encoder-decoder architecture.

[0059] Optionally, the similarity calculation formula is as follows (3):

[0060]

[0061] in, is the similarity value; tanh is the nonlinear activation function; W1 and W2 are the weight matrices to be learned; m is the number of label vectors.

[0062] Optionally, the output module is further configured to:

[0063] The node embedding matrix is ​​input into the trained encoder-decoder architecture, and the node with the highest output probability is selected in turn as the next node of the current path, thereby obtaining the path optimization result of the relay node in the wireless relay communication system.

[0064] Optionally, the performance evaluation index of the wireless relay communication system is minimization of information delay.

[0065] The information delay is calculated as follows:

[0066]

[0067] Among them, t r The time it takes for the relay node to receive data; t m is the inter-node movement time; k is the number of fixed nodes.

[0068] On the one hand, an electronic device is provided, which includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the above-mentioned wireless relay communication method based on graph neural network path optimization.

[0069] On the one hand, a computer-readable storage medium is provided, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement the above-mentioned wireless relay communication method based on graph neural network path optimization.

[0070] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0071] In the above solution, the present invention's wireless relay communication method based on graph neural network path optimization solves the problem of poor information transmission performance in multi-node wireless relay communication methods due to the mobility of unmanned systems and the instability of wireless communication. When the network structure does not meet the information transmission conditions, the wireless relay node path can be adjusted in a timely manner, allowing information to be transmitted without distortion and reducing communication latency.

[0072] The present invention utilizes graph neural networks to autonomously generate the movement path of the unmanned system during wireless relay communication by combining the power of fixed nodes and the relative distance between fixed nodes and relay nodes.

[0073] The present invention does not impose artificial path restrictions on relay nodes and is a more generalized wireless relay communication method that is not limited by distance and the number of fixed nodes, and achieves a wireless relay communication method with high end-to-end reliability and a wider range of applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0075] Figure 1 This is a flow chart of a wireless relay communication method based on graph neural network path optimization provided by an embodiment of the present invention;

[0076] Figure 2 1. It is a schematic diagram of the encoder-decoder training process of the wireless relay communication method based on path optimization of graph neural network provided by an embodiment of the present invention;

[0077] Figure 3This is a schematic diagram of path optimization during actual application of the wireless relay communication method based on graph neural network path optimization provided by an embodiment of the present invention;

[0078] Figure 4 This is a block diagram of a wireless relay communication device based on graph neural network path optimization provided by an embodiment of the present invention;

[0079] Figure 5 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0080] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0081] like Figure 1 As shown, an embodiment of the present invention provides a wireless relay communication method based on graph neural network path optimization, which can be implemented by an electronic device. Figure 1 The flowchart of the wireless relay communication method based on graph neural network path optimization is shown. The processing flow of the method may include the following steps:

[0082] S1. Generate an embedding matrix of nodes in a wireless relay communication system.

[0083] Optionally, generating an embedding matrix of a node in the wireless relay communication system in S1 includes:

[0084] S11. Acquire a node in the wireless relay communication system.

[0085] The nodes include multiple fixed nodes and relay nodes.

[0086] S12. Build a graph neural network based on nodes.

[0087] In a feasible implementation, the present invention implements path optimization through graph neural networks for application in specific scenarios to improve wireless relay communication performance.

[0088] Among them, the specific scenario refers to a multi-node wireless relay communication system, which deploys an unmanned system using dual-hop single-node or multi-hop multi-node wireless relay as the wireless communication method, forming a communication network between each node in the system.

[0089] A multi-node wireless relay communication system refers to a wireless network with multiple fixed nodes to transmit information. All fixed nodes are equipped with energy-constrained sensors and need to transmit data to a data center. However, due to damage to ground communication links, transmission on the ground is impossible, necessitating the deployment of unmanned systems as wireless relay nodes to assist in communication.

[0090] Dual-hop single-node wireless relay refers to the situation where the communication link from the source to the destination is damaged or the distance between the source and the destination is too far. Information is transmitted through a single wireless relay node to expand the coverage area, where the source to the wireless relay node is one hop and the wireless relay node to the destination is another hop.

[0091] Multi-hop, multi-node wireless relaying involves selecting a wireless relay node to meet the information transmission requirements of all communication nodes when there are multiple sources or destinations requiring communication. The number of hops increases with the number of sources, destinations, and wireless relay nodes.

[0092] Performance evaluation metrics for wireless relay communication systems refer to ensuring system communication performance in a multi-node wireless relay communication system. This requires considering not only the performance of a single node but all nodes in the entire communication system. Therefore, throughput cannot be the sole metric for evaluating a system. This approach aims to minimize information latency—the time it takes for an unmanned system to complete a wireless relay communication task—by minimizing the time it takes for information from each node to reach the unmanned system. This approach improves the performance of wireless relay communication systems through path optimization.

[0093] Path optimization refers to the auxiliary task of formulating the movement path of the unmanned system by adjusting the position, movement direction, speed and other parameters of the unmanned system as a wireless relay communication node to achieve wireless relay communication.

[0094] Specifically, in a multi-node wireless relay communication network, the total data capacity of the entire network is B. The data center is represented by parameter N0, and the wireless relay node unmanned system is represented by parameter U, and its position coordinates can be expressed as (x(t), y(t), h(t)). There are k fixed nodes to be transmitted in the network, through N = {N1, N2, ..., N k} to represent a fixed node set, the energy of the nodes are E={E1,E2,…,E k}, the coordinates of the data center and other nodes are (x i ,y i ,h i ),i=0,1,…,k。

[0095] Optionally, the performance evaluation index of the wireless relay communication system is minimization of information delay.

[0096] Information delay refers to the time it takes for the unmanned system to complete the task of receiving data and move to the data center, which includes the time t used to receive data. r and inter-node movement time t m , the calculation expression of information delay (1):

[0097]

[0098] Among them, t r The time it takes for the relay node to receive data; t m is the inter-node movement time; k is the number of fixed nodes.

[0099] Furthermore, the time t taken to receive data r , which means that in order to ensure that the nodes transmit all data to the unmanned system within the limited energy range as much as possible, the unmanned system will move to the top of all nodes to receive data to ensure the best channel quality. The time the unmanned system receives information above the node is called the time it takes to receive data t r During this period, since the positions of the fixed node and the relay node are relatively static, the channel environment will not change. Both will transmit data at a fixed transmission power. However, due to certain energy limitations of the nodes, the energy limitation condition of the i-th node is as follows (2):

[0100] t ri ×p i ≤E i (2)

[0101] Among them, t ri is the transmission time used in fixed node i; p i Refers to the power transmitted by the fixed node; E i Refers to the energy limit of a fixed node i.

[0102] Combined with the total amount of data B that needs to be transmitted in the entire network, the following formula (3) needs to be satisfied:

[0103]

[0104] Furthermore, the inter-node movement time t m , refers to the time t taken by the unmanned system to move to the position above a node at the maximum speed m The maximum speed of the unmanned system is v max , the distance between the relay node and the fixed node is d iu , the calculation formula is as follows (4):

[0105]

[0106] S13. Obtain k+1-dimensional information of each node in the multiple nodes.

[0107] The k+1-dimensional information includes the relative power of the nodes and the distance between the fixed nodes and the relay nodes.

[0108] In a feasible implementation, each node contains k+1-dimensional information consisting of the relative power of each node and the distance between the fixed node and the relay node.

[0109] S14. Aggregate the k+1-dimensional information of each node based on the graph neural network to generate an embedding matrix of the node in the wireless relay communication system.

[0110] In a feasible implementation, a graph neural network is used to aggregate the information of each node to generate an embedding matrix of the node.

[0111] Alternatively, the embedding matrix expression of the nodes in S1 is as follows (5):

[0112] H (l+1) =σ(D -1 / 2 AD 1 / 2 ·H (l) W (l) ) (5)

[0113] Among them, H (l+1) is the embedding matrix of k+1-dimensional information in the l+1th layer of the graph neural network; σ is the nonlinear activation function; is the diagonal matrix after the self-loop; is the adjacent order matrix after the self-loop; H (l) is the embedding matrix of k+1-dimensional information in the lth layer of the graph neural network; W (l) is the weight matrix of the lth layer.

[0114] In a feasible implementation, the wireless relay communication network graph neural network refers to using a graph G to describe the entire communication network, where the graph G is defined as an ancestor G = (V, E) of a set of nodes V and a set of edges E, and the node V includes fixed nodes and relay nodes.

[0115] Among them, the angle matrix after the self-loop It is obtained by calculating the degree of each node in the graph G.

[0116] The adjacent matrix after the self-loop Obtained by calculating the edges of each node in the graph G.

[0117] Embedding matrix H (l) , which contains k+1-dimensional information consisting of the relative power of each node and the distance between the fixed node and the relay node.

[0118] Nonlinear activation function, the ReLU activation function can be used. The calculation formula of the ReLU activation function is as follows (6):

[0119] ReLU(x)=max(0,x) (6)

[0120] S2. Based on the embedding matrix of the node, the encoder-decoder architecture is trained to obtain a trained encoder-decoder architecture.

[0121] Optionally, the above step S2 may include the following steps S21-S24:

[0122] S21. Input the embedding matrix of the node into the encoder to obtain the state vector e of each unit hidden layer of the embedding matrix in the encoder. i .

[0123] In one feasible implementation, the forward propagation of the training process is first performed: the embedding matrix of the graph neural network is input into the encoder as a training sample, and the state vector e of the embedding matrix in each unit hidden layer of the encoder is obtained. i .

[0124] S22, input the state vector of the last unit hidden layer of the encoder into the decoder, and obtain the state vector d of each unit hidden layer in the decoder j .

[0125] In one feasible implementation, the state vector of the last hidden layer of the encoder is used as the input of the decoder to obtain the state vector d of each unit hidden layer in the decoder. j .

[0126] S23, the state vector d of each unit hidden layer in the decoder j Respectively with the state vector e of each unit hidden layer in the encoder i Perform similarity calculation to obtain the state vector with the highest similarity in the encoder, and use the node path corresponding to the state vector with the highest similarity as the output of the decoder to generate a complete path.

[0127] In one feasible implementation, the state vector d of each hidden layer in the decoder is j Respectively with the state vector e of each unit hidden layer in the encoder i Perform similarity calculation, take the node path corresponding to the hidden layer state vector with the highest similarity in the encoder as the output of the decoder, sort them according to the similarity, and then generate a complete path.

[0128] Optionally, the similarity calculation formula in S23 is as follows (7):

[0129]

[0130] in, is the similarity value; tanh is the nonlinear activation function; W1 and W2 are the weight matrices to be learned; m is the number of label vectors.

[0131] The output of the decoder is obtained according to the similarity as shown in the following formula (8):

[0132]

[0133] S24. Calculate the sum of the cross entropy loss functions of all units in the decoder, use the optimization algorithm to iteratively solve it, and update the weights until convergence to obtain the trained encoder-decoder architecture.

[0134] In one feasible implementation, after obtaining the result of forward propagation, backpropagation is performed: the sum of the cross entropy loss functions of all units in the decoder is calculated, the optimization algorithm is used to iteratively solve the problem, and the relevant weights are updated until convergence to obtain the trained encoder-decoder architecture.

[0135] The calculation formula of the loss function is as follows (9):

[0136]

[0137] in, is the jth label vector of the i-th unit of the decoder.

[0138] Furthermore, both the encoder and the decoder can be a long short-term memory network.

[0139] For example, if Figure 2 As shown, the graph neural network has three layers in total, and the original input is H (0) , H is obtained through the graph neural network (1) , H (2) and H (3) are the state vector matrices of the first hidden layer, the second hidden layer, and the last hidden layer of the graph neural network respectively. The state vector matrix of the third hidden layer is selected as H (3) is the input of the encoder-decoder.

[0140] The encoder-decoder architecture is a long short-term memory network, and each unit includes the following formula (10):

[0141]

[0142] Where σ is the sigmoid activation function, expressed as σ(x)=1 / (1+e -x ), tanh is the tanh activation function, parameter W f ,W i ,W o ,W c The parameters b are the weight matrices of the encoder-decoder's forget gate, input gate, output gate, and connected hidden layer of the computational memory unit at time t-1 and time t. i ,b f ,b o ,b cThey are the forget gate, input gate, output gate of the encoder-decoder and the deviation matrix of the connected hidden layer of the computational memory unit at time t-1 state and time t state, x t Refers to the input of the t-th unit, h t-1 Refers to the hidden layer vector output by the t-1th unit.

[0143] S3. Input the node embedding matrix into the trained encoder-decoder architecture to obtain the path optimization result of the relay node in the wireless relay communication system.

[0144] Optionally, the step of inputting the embedding matrix of the node into the trained encoder-decoder architecture in S3 to obtain a path optimization result of the relay node in the wireless relay communication system includes:

[0145] The node embedding matrix is ​​input into the trained encoder-decoder architecture, and the node with the highest output probability is selected in turn as the next node of the current path, thereby obtaining the path optimization result of the relay node in the wireless relay communication system.

[0146] In one feasible implementation, the input of each unit in the encoder is the state vector matrix of the third hidden layer output by the graph neural network in the previous step. The embedding matrix containing 10 nodes, including fixed nodes and relay nodes, is sequentially input into the encoder, and each unit of the encoder outputs a hidden layer vector.

[0147] For example, in the decoder, the state vector of the last unit of the encoder will be used as the input of the decoder, and then the hidden layer vectors of the decoder will be obtained in sequence. The similarity calculation is performed on the state vector of each hidden layer in turn, and finally the 10 points (0, 8, 3, 4, 7, 1, 6, 5, 2, 9) are obtained, such as Figure 3 As shown, the path 0=>8=>3=>4=>7=>1=>6=>5=>2=>9 is output.

[0148] This invention is applied to a multi-node wireless relay communication system, deploying an unmanned system using dual-hop single-node or multi-hop multi-node wireless relay as the wireless communication method, forming a communication network between the nodes in the system. Minimizing information delay is used as a performance evaluation metric for wireless relay communication systems. By adjusting the power of each fixed node and the relative position of the fixed node and relay node, a graph neural network is used to characterize the unmanned system's multi-node wireless relay communication process and achieve path optimization for the unmanned system in the wireless relay communication system.

[0149] In an embodiment of the present invention, a wireless relay communication method based on graph neural network path optimization solves the problem of poor information transmission performance in multi-node wireless relay communication methods due to the mobility of unmanned systems and the instability of wireless communications. When the network structure does not meet the information transmission conditions, the wireless relay node path can be adjusted in a timely manner, allowing information to be transmitted without distortion and reducing communication latency.

[0150] The present invention utilizes graph neural networks to autonomously generate the movement path of the unmanned system during wireless relay communication by combining the power of fixed nodes and the relative distance between fixed nodes and relay nodes.

[0151] The present invention does not impose artificial path restrictions on relay nodes and is a more generalized wireless relay communication method that is not limited by distance and the number of fixed nodes, and achieves a wireless relay communication method with high end-to-end reliability and a wider range of applicability.

[0152] like Figure 4 As shown, an embodiment of the present invention provides a wireless relay communication device 400 based on graph neural network path optimization. The device 400 is applied to implement a wireless relay communication method based on graph neural network path optimization. The device 400 includes:

[0153] The generation module is used to generate an embedding matrix of nodes in a wireless relay communication system.

[0154] The training module is used to train the encoder-decoder architecture based on the embedding matrix of the node to obtain the trained encoder-decoder architecture.

[0155] The output module is used to input the node embedding matrix into the trained encoder-decoder architecture to obtain the path optimization result of the relay node in the wireless relay communication system.

[0156] Optionally, a generation module is further configured to:

[0157] S11. Acquire nodes in the wireless relay communication system; wherein the nodes include multiple fixed nodes and relay nodes.

[0158] S12. Build a graph neural network based on nodes.

[0159] S13. Acquire k+1-dimensional information of each of the multiple nodes; wherein the k+1-dimensional information includes the relative power of the node and the distance between the fixed node and the relay node.

[0160] S14. Aggregate the k+1-dimensional information of each node based on the graph neural network to generate an embedding matrix of the node in the wireless relay communication system.

[0161] Optionally, the fixed node is an energy-constrained node;

[0162] The energy constraint condition of any fixed node i among multiple fixed nodes is as follows:

[0163] t ri ×p i ≤E i (1)

[0164] Among them, t ri is the transmission time used in fixed node i; p i Refers to the power transmitted by the fixed node; E i Refers to the energy limit of a fixed node i.

[0165] Optionally, the embedding matrix of the node is expressed as follows (2):

[0166] H (l+1) =σ(D -1 / 2 AD 1 / 2 ·H (l) W (l) ) (2)

[0167] Among them, H (l+1) is the embedding matrix of k+1-dimensional information in the l+1th layer of the graph neural network; σ is the nonlinear activation function; is the diagonal matrix after the self-loop; is the adjacent order matrix after the self-loop; H (l) is the embedding matrix of k+1-dimensional information in the lth layer of the graph neural network; W (l) is the weight matrix of the lth layer.

[0168] Optionally, the training module is further configured to:

[0169] S21. Input the embedding matrix of the node into the encoder to obtain the state vector e of each unit hidden layer of the embedding matrix in the encoder. i .

[0170] S22, input the state vector of the last unit hidden layer of the encoder into the decoder, and obtain the state vector d of each unit hidden layer in the decoder j .

[0171] S23, the state vector d of each unit hidden layer in the decoder j Respectively with the state vector e of each unit hidden layer in the encoder i Perform similarity calculation to obtain the state vector with the highest similarity in the encoder, and use the node path corresponding to the state vector with the highest similarity as the output of the decoder to generate a complete path.

[0172] S24. Calculate the sum of the cross entropy loss functions of all units in the decoder, use the optimization algorithm to iteratively solve it, and update the weights until convergence to obtain the trained encoder-decoder architecture.

[0173] Optionally, the similarity calculation formula is as follows (3):

[0174]

[0175] in, is the similarity value; tanh is the nonlinear activation function; W1 and W2 are the weight matrices to be learned; m is the number of label vectors.

[0176] Optionally, the output module is further configured to:

[0177] The node embedding matrix is ​​input into the trained encoder-decoder architecture, and the node with the highest output probability is selected in turn as the next node of the current path, thereby obtaining the path optimization result of the relay node in the wireless relay communication system.

[0178] Optionally, the performance evaluation index of the wireless relay communication system is minimization of information delay.

[0179] The information delay is calculated as follows:

[0180]

[0181] Among them, t r The time it takes for the relay node to receive data; t m is the inter-node movement time; k is the number of fixed nodes.

[0182] In an embodiment of the present invention, a wireless relay communication method based on graph neural network path optimization solves the problem of poor information transmission performance in multi-node wireless relay communication methods due to the mobility of unmanned systems and the instability of wireless communications. When the network structure does not meet the information transmission conditions, the wireless relay node path can be adjusted in a timely manner, allowing information to be transmitted without distortion and reducing communication latency.

[0183] The present invention utilizes graph neural networks to autonomously generate the movement path of the unmanned system during wireless relay communication by combining the power of fixed nodes and the relative distance between fixed nodes and relay nodes.

[0184] The present invention does not impose artificial path restrictions on relay nodes and is a more generalized wireless relay communication method that is not limited by distance and the number of fixed nodes, and achieves a wireless relay communication method with high end-to-end reliability and a wider range of applicability.

[0185] Figure 5is a schematic structural diagram of an electronic device 500 provided in an embodiment of the present invention. The electronic device 500 may vary significantly due to different configurations or performances, and may include one or more processors (central processing units, CPUs) 501 and one or more memories 502. The memories 502 store at least one instruction, which is loaded and executed by the processor 501 to implement the following wireless relay communication method based on graph neural network path optimization:

[0186] S1. Generate an embedding matrix of nodes in a wireless relay communication system.

[0187] S2. Based on the embedding matrix of the node, the encoder-decoder architecture is trained to obtain a trained encoder-decoder architecture.

[0188] S3. Input the node embedding matrix into the trained encoder-decoder architecture to obtain the path optimization result of the relay node in the wireless relay communication system.

[0189] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions. The instructions can be executed by a processor in a terminal to implement the wireless relay communication method based on graph neural network path optimization. For example, the computer-readable storage medium can be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0190] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0191] 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 in the scope of protection of the present invention.

Claims

1. A wireless relay communication method based on graph neural network path optimization, characterized in that: The method comprises: S1. Generate an embedding matrix of nodes in a wireless relay communication system; wherein the wireless relay communication system uses a dual-hop single-node or multi-hop multi-node wireless relay as a wireless communication mode; S2. Training an encoder-decoder architecture based on the embedding matrix of the node to obtain a trained encoder-decoder architecture; S3. Inputting the embedding matrix of the node into the trained encoder-decoder architecture, minimizing information delay as a performance evaluation indicator of the wireless relay communication system, and using a graph neural network to characterize the unmanned system's multi-node wireless relay communication process and implement path optimization of the unmanned system in the wireless relay communication system by adjusting the power of each fixed node and the relative position of the fixed node and the relay node, thereby obtaining a path optimization result of the relay node in the wireless relay communication system; The step of generating an embedding matrix of a node in the wireless relay communication system in S1 includes: S11. Acquire nodes in the wireless relay communication system; wherein the nodes include a plurality of fixed nodes and relay nodes; S12. Constructing a graph neural network based on the nodes; S13. Acquire k+1-dimensional information of each of the multiple nodes; wherein the k+1-dimensional information includes the relative power of the node and the distance between the fixed node and the relay node; S14. Aggregating the k+1-dimensional information of each node based on the graph neural network to generate an embedding matrix of the node in the wireless relay communication system; The fixed node in S11 is an energy-constrained node; The energy limit condition of any fixed node i among the multiple fixed nodes is as follows: t ri ×p i ≤E i (1) Among them, t ri is the transmission time used in fixed node t; p i Refers to the power transmitted by the fixed node; E i refers to the energy limit of a fixed node i; The information delay is calculated as follows: Among them, t r The time it takes for the relay node to receive data refers to the time it takes for the relay node to transmit all data to the unmanned system within the limited energy range. The unmanned system will move directly above all nodes to receive data to ensure the best channel quality. m It is the time taken by the UAV system to move to the position above a node at the maximum speed. The maximum speed of the UAV system is v max , the distance d between the relay node and the fixed node iu Calculation; k is the number of fixed nodes.

2. The method according to claim 1, characterized in that The embedding matrix expression of the nodes in S1 is as follows (3): Among them, H (l+1) is the embedding matrix of k+1-dimensional information in the l+1th layer of the graph neural network; σ is the nonlinear activation function; is the diagonal matrix after the self-loop; is the adjacent order matrix after the self-loop; H (l) is the embedding matrix of k+1-dimensional information in the lth layer of the graph neural network; W (l) is the weight matrix of the lth layer.

3. The method according to claim 1, characterized in that The encoder-decoder architecture is trained based on the embedding matrix of the node in S2, and the trained encoder-decoder architecture is obtained, including: S21, input the embedding matrix of the node into the encoder, and obtain the state vector e of each unit hidden layer of the embedding matrix in the encoder i ; S22, input the state vector of the last unit hidden layer of the encoder into the decoder, and obtain the state vector d of each unit hidden layer in the decoder j ; S23, the state vector d of each unit hidden layer in the decoder j The state vector e of each hidden layer of the encoder is i Perform similarity calculation to obtain the state vector with the highest similarity in the encoder, and use the node path corresponding to the state vector with the highest similarity as the output of the decoder to generate a complete path; S24. Calculate the sum of the cross entropy loss functions of all units in the decoder, use the optimization algorithm to iteratively solve it, and update the weights until convergence to obtain the trained encoder-decoder architecture.

4. The method according to claim 3, characterized in that The similarity calculation formula in S23 is as follows (4): in, is the similarity value; tanh is the nonlinear activation function; W1 and W2 are the weight matrices to be learned; m is the number of label vectors.

5. The method according to claim 1, wherein Inputting the embedding matrix of the node into the trained encoder-decoder architecture in S3 to obtain a path optimization result of the relay node in the wireless relay communication system includes: The embedding matrix of the node is input into the trained encoder-decoder architecture, and the node with the highest output probability is selected in turn as the next node of the current path, thereby obtaining the path optimization result of the relay node in the wireless relay communication system.

6. A wireless relay communication device based on graph neural network path optimization, characterized in that: The device comprises: A generation module is used to generate an embedding matrix of nodes in a wireless relay communication system; wherein the wireless relay communication system uses a dual-hop single-node or multi-hop multi-node wireless relay as a wireless communication mode; A training module, configured to train an encoder-decoder architecture based on the embedding matrix of the node to obtain a trained encoder-decoder architecture; An output module is configured to input the node embedding matrix into the trained encoder-decoder architecture, minimize information delay as a performance evaluation indicator of the wireless relay communication system, and utilize a graph neural network to characterize the unmanned system's multi-node wireless relay communication process and implement path optimization of the unmanned system in the wireless relay communication system by adjusting the power of each fixed node and the relative position of the fixed node and the relay node, thereby obtaining a path optimization result of the relay node in the wireless relay communication system; Generating an embedding matrix of a node in a wireless relay communication system includes: S11. Acquire nodes in the wireless relay communication system; wherein the nodes include a plurality of fixed nodes and relay nodes; S12. Constructing a graph neural network based on the nodes; S13. Acquire k+1-dimensional information of each of the multiple nodes; wherein the k+1-dimensional information includes the relative power of the node and the distance between the fixed node and the relay node; S14. Aggregating the k+1-dimensional information of each node based on the graph neural network to generate an embedding matrix of the node in the wireless relay communication system; The fixed node in S11 is an energy-constrained node; The energy limit condition of any fixed node i among the multiple fixed nodes is as follows: t ri ×p i ≤E i (1) Among them, t ri is the transmission time used in fixed node t; p i Refers to the power transmitted by the fixed node; E i Refers to the energy limit of a fixed node i The information delay is calculated as follows: Among them, t r The time it takes for the relay node to receive data refers to the time it takes for the relay node to transmit all data to the unmanned system within the limited energy range. The unmanned system will move directly above all nodes to receive data to ensure the best channel quality. m It is the time taken by the UAV system to move to the position above a node at the maximum speed. The maximum speed of the UAV system is v max , the distance d between the relay node and the fixed node iu Calculation; k is the number of fixed nodes.

Images