Association methods, apparatus and devices for sensor-integrated networks
By constructing a heterogeneous graph using a graph neural network model, the association strategy between UAV base stations, communication users, and sensing targets is optimized, solving the problem of poor association decision-making in existing technologies and improving the performance and efficiency of the integrated sensing network.
Patent Information
- Application Number
- CN202611239424.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, the association decision between drone base stations, communication users, and sensing targets is poor, resulting in a decline in the performance of integrated communication and sensing, a decline in sensing performance, and an excessive backhaul load. Furthermore, the single-connection association model is difficult to achieve true communication and sensing collaboration and cannot meet the complex requirements of integrated communication and sensing networks.
A graph neural network model is adopted to construct a heterogeneous graph based on the features of UAV base stations, communication users and sensing targets. The node embedding features are updated through multiple computing layers, edge weights are calculated and the association relationship is determined, and the association strategy is optimized. The graph attention network model is used to perform feature aggregation and weight calculation to achieve a better association strategy.
It improves the overall performance of the integrated sensing network, fully leverages the communication and sensing capabilities of UAV base stations, enhances the effectiveness and efficiency of association strategies, and reduces service interruption time.
Smart Images

Figure CN123124135A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of communication technology, and more specifically, to an association method, apparatus, and device for a sensor-integrated network. Background Technology
[0002] With the rapid development of communication technology, integrated communication and sensing technology has become a core technology for realizing the synergy of communication and sensing functions, and has significant application value in typical scenarios such as UAV emergency communication and vehicle-to-everything (V2X) communication. Especially in post-disaster emergency scenarios, traditional ground communication facilities are easily damaged, and UAV base stations, with their advantages of rapid deployment and three-dimensional mobility, have become a key carrier for restoring communication and sensing capabilities. Summary of the Invention
[0003] A brief overview of this disclosure is given below to provide a basic understanding of some aspects of it. However, it should be understood that this overview is not an exhaustive summary of this disclosure. It is not intended to identify key or essential parts of this disclosure, nor is it intended to limit the scope of this disclosure. Its purpose is merely to present certain concepts of this disclosure in a simplified form as a prelude to the more detailed description that follows.
[0004] In related technologies, there is a problem of poor association decisions when associating UAV base stations with communication users and / or sensing targets. Specifically, the association strategies in these technologies may lead to UAV base stations being associated with incompatible communication users and sensing targets, resulting in decreased communication performance, decreased sensing performance, and excessive backhaul load in the integrated sensing and communication network. Furthermore, these technologies typically employ a single-connection association model, optimizing communication and sensing performance in isolation, which makes it difficult to achieve true sensing and communication collaboration and cannot meet the needs of complex integrated sensing and communication networks.
[0005] According to a first aspect of this disclosure, an association method for a sensing-integrated network is provided, comprising: acquiring a first feature of an unmanned aerial vehicle (UAV) base station, a second feature of a communication user, and a third feature of a sensing target; using a graph neural network model, obtaining target embedding features of nodes in a heterogeneous graph based on the first feature, the second feature, and the third feature, and obtaining a first edge weight and a second edge weight based on the target embedding features of nodes in the heterogeneous graph, wherein the UAV base station, the communication user, and the sensing target constitute nodes in the heterogeneous graph, the first edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the communication user in the heterogeneous graph, and the second edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the sensing target; determining an association between the UAV base station and the communication user in response to the first edge weight being greater than or equal to a first threshold; and determining an association between the UAV base station and the sensing target in response to the second edge weight being greater than or equal to the second threshold.
[0006] In some embodiments, the first feature includes at least one of the following: the location of the UAV base station, remaining communication resources, remaining sensing resources, maximum data transmission rate supported by the backhaul link, and occupied backhaul link capacity; the second feature includes at least one of the following: the location of the communication user, the communication rate quality of service requirement threshold, and the task priority weight coefficient; and the third feature includes at least one of the following: the location of the sensing target, the sensing rate quality of service requirement threshold, and the task priority weight coefficient.
[0007] In some embodiments, the graph neural network model includes multiple computational layers. Using the graph neural network model, target embedding features of nodes in a heterogeneous graph are obtained based on the first feature, the second feature, and the third feature. Obtaining first edge weights and second edge weights based on the target embedding features of nodes in the heterogeneous graph includes: in each computational layer of the multiple computational layers, for each node in the heterogeneous graph, using the embedding features of the node's neighboring nodes obtained in the previous computational layer, an aggregation result for the node is obtained; using the embedding features of the node obtained in the previous computational layer and the aggregation result of the node, the embedding feature of the node in that computational layer is obtained, wherein the embedding feature of the node in the last computational layer of the multiple computational layers is used as the target embedding feature of the node; and in the first computational layer of the multiple computational layers, the initial embedding feature of the node is used as the embedding feature obtained in the previous computational layer. The initial embedding feature of the node corresponding to the UAV base station is obtained based on the first feature, the initial embedding feature of the node corresponding to the communication user is obtained based on the second feature, and the initial embedding feature of the node corresponding to the sensing target is obtained based on the third feature.
[0008] In some embodiments, the association method further includes: obtaining a first initial association relationship between the UAV base station and the communication user, and a second initial association relationship between the UAV base station and the perception target based on the locations of the UAV base station, the communication user, and the perception target; and obtaining the neighboring nodes of the node corresponding to the communication user, the neighboring nodes of the node corresponding to the perception target, and the neighboring nodes of the node corresponding to the UAV base station based on the first initial association relationship and the second initial association relationship.
[0009] In some embodiments, the association method further includes: before obtaining target embedding features of nodes in a heterogeneous graph based on the first feature, the second feature, and the third feature using a graph neural network model, and obtaining the first edge weight and the second edge weight based on the target embedding features of nodes in the heterogeneous graph, for each UAV base station, controlling the UAV base station to establish a connection with the communication user that has established an initial association with the UAV base station according to a first initial association relationship, and controlling the UAV base station to perceive the sensing target that has established an initial association with the UAV base station according to a second initial association relationship.
[0010] In some embodiments, the graph neural network model is a graph attention network model. The aggregation result of a node is obtained by utilizing the embedding features of its neighboring nodes obtained in the previous computational layer. This includes: for each neighboring node, obtaining the attention weight coefficient between the node and its neighboring nodes based on the task priority weight vector, the attention weight vector, the embedding features of the node obtained in the previous computational layer, and the embedding features of the neighboring nodes obtained in the previous computational layer; wherein the task priority weight vector is obtained based on the task priority weight coefficient of the node and the task priority weight coefficient of the neighboring nodes; and obtaining the aggregation result of the node based on the embedding features of each neighboring node obtained in the previous computational layer and the attention weight coefficients between the node and each neighboring node.
[0011] In some embodiments, using a graph neural network model, target embedding features of nodes in a heterogeneous graph are obtained based on the first feature, the second feature, and the third feature. The first edge weight and the second edge weight are obtained based on the target embedding features of nodes in the heterogeneous graph, including: obtaining the first edge weight based on the target embedding features of the node corresponding to the UAV base station, the target embedding features of the node corresponding to the communication user, and the task priority weight coefficient of the communication user; and / or obtaining the second edge weight based on the target embedding features of the node corresponding to the UAV base station, the target embedding features of the node corresponding to the sensing target, and the task priority weight coefficient of the sensing target.
[0012] In some embodiments, the first threshold corresponds to the task priority weight coefficient of the communication user. When the task priority weight coefficient of the communication user is selected from multiple first priority weight coefficients, the first threshold is selected from multiple first decision thresholds, wherein the magnitudes of the multiple first priority weight coefficients are different, the magnitudes of the multiple first decision thresholds are different, and each first decision threshold corresponds to a first priority weight coefficient; and / or the second threshold corresponds to the task priority weight coefficient of the sensing target. When the task priority weight coefficient of the sensing target is selected from multiple second priority weight coefficients, the second threshold is selected from multiple second decision thresholds, wherein the magnitudes of the multiple second priority weight coefficients are different, the magnitudes of the multiple second decision thresholds are different, and each second decision threshold corresponds to a second priority weight coefficient.
[0013] In some embodiments, the association method further includes: for each drone base station, controlling the drone base station to establish a connection with a communication user associated with the drone base station, and controlling the drone base station to perceive a sensing target associated with the drone base station.
[0014] In some embodiments, the association method further includes: acquiring training samples, wherein the feature portion of the training samples includes a first feature of the sample UAV base station in the sample sensing network, a second feature of the sample communication user, and a third feature of the sample sensing target, and the label portion of the training samples includes a sample association label, which indicates the sample association relationship between the sample UAV base station and the sample communication user in the sensing network, and the sample association relationship between the sample UAV base station and the sample sensing target; and training a graph neural network model based on the training samples.
[0015] In some embodiments, training a graph neural network model based on training samples includes: training the graph neural network model based on training samples with the objective of reducing the value of a target loss function, wherein the target loss function includes one or more binary cross-entropy loss functions, each binary cross-entropy loss function corresponds to a candidate association edge in the heterogeneous graph corresponding to the sample synesthetic network, and each binary cross-entropy loss function is set based on the predicted edge weight of the corresponding candidate association edge in the heterogeneous graph corresponding to the sample synesthetic network, the sample association relationship of the corresponding candidate association edge in the sample association label, and the task priority weight coefficient of the corresponding candidate association edge. The predicted edge weight is obtained using the graph neural network model, and the edges between the sample UAV base station node and the sample communication user node, and the edges between the sample UAV base station node and the sample perception target node constitute candidate association edges.
[0016] In some embodiments, the graph neural network model is trained based on a Markov decision process, wherein the reward function in the Markov decision process is based on the state... implement The status is set based on the communication performance reward, perception performance reward, backhaul load penalty, and priority reward. This includes the relationships and characteristics of nodes in the integrated sensory network at time t, and actions. It is set to the predicted association relationship or predicted edge weight between nodes obtained based on the graph neural network model.
[0017] According to a second aspect of this disclosure, an association device for a sensing-integrated network is provided, comprising: an acquisition module configured to: acquire a first feature of an unmanned aerial vehicle (UAV) base station, a second feature of a communication user, and a third feature of a sensing target; a weight determination module configured to: use a graph neural network model to obtain target embedding features of nodes in a heterogeneous graph based on the first feature, the second feature, and the third feature, and obtain a first edge weight and a second edge weight based on the target embedding features of nodes in the heterogeneous graph, wherein the UAV base station, the communication user, and the sensing target constitute nodes in the heterogeneous graph, the first edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the communication user in the heterogeneous graph, and the second edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the sensing target; and an association determination module configured to: determine an association between the UAV base station and the communication user in response to the first edge weight being greater than or equal to a first threshold, and determine an association between the UAV base station and the sensing target in response to the second edge weight being greater than or equal to a second threshold.
[0018] According to a third aspect of this disclosure, an association device for a sensor-integrated network is provided, including a processor and a memory, wherein the memory stores instructions that, when executed by the processor, implement the steps of the association method as described above.
[0019] According to a fourth aspect of this disclosure, an aerial platform station is provided, comprising: the associated device as described above; or the associated equipment as described above.
[0020] According to a fifth aspect of this disclosure, a non-transitory computer-readable storage medium is provided, wherein instructions are stored on the non-transitory computer-readable storage medium, which, when executed by a processor, implement the steps of the association method as described above.
[0021] According to a sixth aspect of this disclosure, a computer program product is provided, including instructions that, when executed by a processor, implement the steps of the association method as described above.
[0022] According to embodiments of this disclosure, UAV base stations, communication users, and sensing targets are treated as nodes in a heterogeneous graph corresponding to the integrated sensing network. By using a graph neural network model, the embedding features of each node are updated to discover hidden correlation features in the integrated sensing network from a global perspective. This allows for the acquisition of better correlation strategies for the integrated sensing network, fully leveraging the communication and sensing capabilities of each UAV base station and improving the overall performance of the integrated sensing network.
[0023] Other features and advantages of this disclosure will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description
[0024] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.
[0025] This disclosure will become clearer with reference to the accompanying drawings and the following detailed description, wherein: Figure 1 A schematic diagram of a synergistic network according to some exemplary embodiments of the present disclosure is shown; Figure 2 A flowchart illustrating an association method for a sensor network according to some exemplary embodiments of the present disclosure is shown. Figure 3 A schematic diagram of a graph neural network model according to some exemplary embodiments of the present disclosure is shown; Figure 4 A schematic diagram illustrating a process for obtaining target embedding features of nodes using a graph neural network model according to some exemplary embodiments of the present disclosure is shown. Figure 5 A schematic flowchart illustrating the process of obtaining the aggregation result of nodes according to some exemplary embodiments of the present disclosure is shown; Figure 6 A flowchart illustrating a training method for a graph neural network model according to some embodiments of the present disclosure is shown. Figure 7 A schematic diagram of an association device for a sensor network according to some exemplary embodiments of the present disclosure is shown; Figure 8 A schematic diagram of an associated device for a sensor network according to some exemplary embodiments of the present disclosure is shown; Figure 9 A schematic block diagram of a computer system on which embodiments of the present disclosure may be implemented is shown.
[0026] Note that in the embodiments described below, the same reference numerals are sometimes used across different figures to denote the same parts or parts having the same function, and repeated descriptions are omitted. In this specification, similar reference numerals and letters are used to denote similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.
[0027] For ease of understanding, the positions, dimensions, and extents of the structures shown in the accompanying drawings and other materials may not represent actual positions, dimensions, and extents. Therefore, the disclosed invention is not limited to the positions, dimensions, and extents disclosed in the accompanying drawings and other materials. Furthermore, the drawings are not necessarily drawn to scale, and some features may be enlarged to show details of specific components. Detailed Implementation
[0028] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.
[0029] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this disclosure or its application or use. That is, the structures and methods herein are shown in an exemplary manner to illustrate different embodiments of the structures and methods in this disclosure. However, those skilled in the art will understand that they merely illustrate exemplary ways that can be used to implement this disclosure, and not exhaustive ways. Furthermore, the drawings are not necessarily drawn to scale, and some features may be enlarged to show details of specific components.
[0030] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.
[0031] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.
[0032] Figure 1 A schematic diagram of a sensory integration network according to some embodiments of the present disclosure is shown, such as... Figure 1 As shown, the integrated sensing network 100 may include a High Altitude Platform Station (HAPS) 110, a drone base station 120, a communication user 130, and a sensing target 140.
[0033] The high-altitude platform station 110, acting as a central node for data aggregation and command issuance, can establish connections with multiple UAV base stations 120. Here, the high-altitude platform station 110 can establish connections with the corresponding UAV base stations 120 by establishing a backhaul link 150. The high-altitude platform station 110 can be configured to receive association signaling requests and related data from each UAV base station 120 for centralized processing. Based on the processing results, it can generate association commands and issue them to the corresponding UAV base stations 120 via the backhaul link 150. Thus, the corresponding UAV base station 120 can establish communication connections with the corresponding communication user 130 and / or sense the corresponding sensing target 140 based on the received association commands.
[0034] As an aerial communication and sensing node, the UAV base station 120 can integrate communication and sensing functions and has high-speed data processing and transmission capabilities. The UAV base station 120 can be configured to collect communication data from communication users 130 and sensing and monitoring data of sensing targets 140, and transmit the collected data to the high-altitude platform station 110 through the backhaul link 150.
[0035] Communication user 130 may include, for example, rescue personnel or disaster victims, and communication user 130 has communication capabilities. In this disclosure, a communication connection with communication user 130 can refer to a communication connection with a terminal device that has communication capabilities connected to communication user 130. When a communication link 160 is established between drone base station 120 and communication user 130, communication user 130 can transmit data with drone base station 120 and perform functions such as voice communication and video communication through drone base station 120.
[0036] The sensing target 140 may include, for example, a vital signs detection area, an environmental monitoring point, or disaster-stricken people who do not have communication capabilities. When the UAV base station 120 establishes a sensing link 170 (indicated by a dashed line in the figure) with the sensing target 140, the UAV base station 120 can detect the sensing target 140 and collect sensing data through the sensing link 170.
[0037] Figure 2 A flowchart illustrating an association method for a sensor-integrated network according to some embodiments of this disclosure is shown. Figure 2 As shown, the association method according to some embodiments of this disclosure may include: Step S210: Obtain the first feature of the UAV base station, the second feature of the communication user, and the third feature of the perceived target.
[0038] In some embodiments, the first feature of the UAV base station may include at least one of the following: the location of the UAV base station, remaining communication resources, remaining sensing resources, the maximum data transmission rate supported by the backhaul link, and the occupied backhaul link capacity. Here, the location of the UAV base station can be represented by three-dimensional spatial coordinates. Remaining communication resources can indicate the remaining resource capacity of the UAV base station currently available to support communication functions. Remaining sensing resources can indicate the remaining resource capacity of the UAV base station currently available to support sensing functions. The maximum data transmission rate can indicate the maximum stable data transmission rate supported by the wireless backhaul link between the UAV base station and the high-altitude platform station. The occupied backhaul link capacity can indicate the backhaul link capacity already occupied by the UAV base station's communication data and sensing data.
[0039] In some embodiments, the second characteristic of a communication user may include at least one of the following: the user's location, a communication rate quality of service (QoS) requirement threshold, and a task priority weighting coefficient. Here, the user's location can be represented by three-dimensional spatial coordinates. The QoS requirement threshold indicates the critical communication rate value required to maintain or enable the user's communication functions. The task priority weighting coefficient indicates the priority level of the user's communication task; for example, the higher the urgency of the rescue task corresponding to the user, the higher the priority level of the communication task. In some embodiments, the priority level of a communication task may correspond to the magnitude of the user's task priority weighting coefficient. Specifically, the user's task priority weighting coefficient can be selected from multiple first priority weighting coefficients, wherein different magnitudes of the multiple first priority weighting coefficients indicate different communication task priority levels. For example, the priority levels of a communication task may include high, medium, and low priority levels, which can be indicated by three first priority weighting coefficients of different magnitudes, respectively.
[0040] In some embodiments, the third characteristic of the sensing target may include at least one of the following: the location of the sensing target, the sensing rate service quality requirement threshold, and the task priority weighting coefficient. Here, the location of the sensing target can be represented by three-dimensional spatial coordinates. The sensing rate service quality requirement threshold can indicate the critical value of the sensing rate used to maintain or achieve the sensing function of the sensing target. The task priority weighting coefficient of the sensing target can indicate the priority level of the sensing task of the sensing target; for example, the higher the disaster level corresponding to the sensing target, the higher the priority level of the sensing task. In some embodiments, the priority level of the sensing task can correspond to the magnitude of the task priority weighting coefficient of the sensing target. Specifically, the task priority weighting coefficient of the sensing target can be selected from multiple second priority weighting coefficients, wherein the magnitudes of the multiple second priority weighting coefficients are different to indicate different sensing task priority levels. For example, the priority level of the sensing task may include two priority levels, high and low, which can be indicated by two second priority weighting coefficients of different magnitudes, respectively.
[0041] In this disclosure, the first feature, the second feature, and the third feature can be represented by vectors.
[0042] For example, the first feature of drone base station n can be represented as ,in: Represents the three-dimensional spatial coordinates of the UAV base station n, for example ; This represents the remaining communication resources of drone base station n; This represents the remaining sensing resources of drone base station n; This represents the maximum stable data transmission rate that the wireless backhaul link between the UAV base station n and the high-altitude platform station can support. This indicates the backhaul link capacity already occupied by the communication and sensing data of drone base station n.
[0043] For example, communication users The second feature can be represented as .in: Indicates communication user The three-dimensional spatial coordinates; Indicates communication user QoS requirement threshold for communication rate; Indicates communication user Task priority weighting coefficient ,in , and These represent the high, medium, and low task priority weights for communication users, respectively. , .
[0044] For example, perceiving the target The third feature can be represented as .in: Representing the perceived target The three-dimensional spatial coordinates; Representing the perceived target The perceived rate QoS requirement threshold; Representing the perceived target Task priority weight coefficient, ,in These represent the high and low task priority weights of the perceived target, respectively. .
[0045] like Figure 2 As shown, the association method according to some embodiments of this disclosure may further include: Step S220: Using a graph neural network (GNN) model, the target embedding features of nodes in the heterogeneous graph are obtained based on the first feature, the second feature, and the third feature, and the first edge weight and the second edge weight are obtained based on the target embedding features of nodes in the heterogeneous graph.
[0046] Here, the heterogeneous graph corresponds to the synergistic sensing network. In some embodiments, graph modeling can be performed based on the synergistic sensing network to obtain the heterogeneous graph. The UAV base station, communication user, and sensing target can constitute nodes in the heterogeneous graph; that is, the heterogeneous graph can have three different types of nodes. In this disclosure, the first edge weight is the edge weight between the node corresponding to the UAV base station (also referred to herein as the UAV base station node) and the node corresponding to the communication user (also referred to herein as the communication user node) in the heterogeneous graph, used to indicate the suitability of association between the corresponding UAV base station and the corresponding communication user. The second edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the sensing target (also referred to herein as the sensing target node), used to indicate the suitability of association between the corresponding UAV base station and the corresponding sensing target. Figure 3 A schematic diagram of a graph neural network model according to some embodiments of the present disclosure is illustrated by way of example. Figure 3 As shown, in some embodiments, the graph neural network model 300 may include multiple computation layers 310 to perform layer-by-layer updates and computations using the features of nodes in the heterogeneous graph and the features of the node's neighboring nodes. Here, the processing of each computation layer 310 may include an aggregation step and an update step, wherein the aggregation step may be used to collect information about neighboring nodes, and the update step may be used to use the collected information and the computation results of the current node in the previous computation layer 310 to calculate the computation result of the current node in the current computation layer 310.
[0047] Specifically, such as Figure 3 and Figure 4 As shown, using the graph neural network model 300, the target embedding features of nodes in the heterogeneous graph are obtained based on the first feature, the second feature, and the third feature. The first edge weight and the second edge weight are obtained based on the target embedding features of nodes in the heterogeneous graph, which may include: Step S410: In each of the multi-layer computation layers 310, for each node in the heterogeneous graph, the aggregation result of the node is obtained by using the embedding features of the node's neighboring nodes obtained in the previous computation layer 310. Step S420: Using the embedding features obtained by the node in the previous computational layer 310 and the aggregation result of the node, the embedding features of the node in the current computational layer 310 are obtained. The embedding features of the node in the last computational layer 310 of the multi-layer computational layers are used as the target embedding features of the node.
[0048] Here, in the first computational layer 310 of the multi-layer computational layer 310, the initial embedding features of a node are the embedding features obtained by the node in the previous computational layer 310. The initial embedding features of the node corresponding to the UAV base station are obtained based on the first feature, the initial embedding features of the node corresponding to the communication user are obtained based on the second feature, and the initial embedding features of the node corresponding to the sensing target are obtained based on the third feature. In some embodiments, a multi-layer perceptron (MLP) can be used to obtain the initial embedding features of the UAV base station node based on the first feature of the UAV base station. In some embodiments, a multi-layer perceptron can be used to obtain the initial embedding features of the communication user node based on the second feature of the communication user. In some embodiments, a multi-layer perceptron can be used to obtain the initial embedding features of the sensing target node based on the third feature of the sensing target. Thus, the first feature, the second feature, and the third feature can be input into the multi-layer perceptron for feature encoding and dimension alignment to obtain the initial embedding features of the corresponding nodes.
[0049] In the above embodiments, a graph neural network model is used to effectively mine the hidden correlation features in the synesthetic network through the neighbor node aggregation and update mechanism and message passing, so that better first-side weights and second-side weights can be obtained in the future, and thus the optimal correlation strategy of the synesthetic network can be obtained.
[0050] In this disclosure, embedded features can be represented by vectors.
[0051] In some embodiments, the aggregation result of node i can be obtained based on the following formula: ;here, This represents the aggregation result obtained based on the embedding features of the neighboring nodes of node i in the previous layer (i.e., the (l-1)th layer) of computation layer 310; ( ) represents a nonlinear activation function; Represents the set of neighboring nodes of node i; Indicates the current number l The learnable weight matrix of layer 310 is calculated layer by layer; This represents the embedding feature of neighbor node j obtained in the previous layer (i.e., the (l-1)th layer) computation layer 310.
[0052] Here, the value of l can be from 1 to L, where L can represent the total number of layers in the calculation layer 310. L can represent the initial embedding features of node j. In some embodiments, L can be 2, 3, 4 or other positive integers.
[0053] In some embodiments, the graph neural network model 300 can be a graph attention network model. In the computation of each computation layer 310, the aggregation result of the node can be obtained based on the attention weight coefficients between the node and its neighboring nodes, thereby enabling more accurate message aggregation and obtaining better embedding features of the node.
[0054] Specifically, in some embodiments, such as Figure 5 As shown, the aggregation result of a node can be obtained by utilizing the embedding features of its neighboring nodes obtained in the previous computation layer 310, including: Step S510: For each neighboring node of the node, based on the task priority weight vector, the attention weight vector, the embedding features obtained by the node in the previous computation layer 310, and the embedding features obtained by the neighboring node in the previous computation layer 310, the attention weight coefficient between the node and the neighboring node is obtained. The task priority weight vector is obtained based on the task priority weight coefficient of the node and the task priority weight coefficient of the neighboring node. Step S520: Based on the embedding features of each neighboring node obtained in the previous computation layer 310 and the attention weight coefficients between the node and each neighboring node, the aggregation result of the node is obtained.
[0055] Here, the attention weight vector is a trainable model parameter in the graph attention network model. The corresponding attention weight vector in each computational layer of the graph attention network model is obtained by training the graph attention network model.
[0056] In some embodiments, the task priority weight coefficient of the current node and the task priority weight coefficients of its neighboring nodes can be concatenated and mapped to obtain a task priority weight vector. For example, a multilayer perceptron can be used to obtain the task priority weight vector based on the task priority weight coefficients of the current node and its neighboring nodes. As some implementations, when the current node does not have a task priority weight coefficient, for example, when the current node is a drone base station node, the task priority weight coefficients of its neighboring nodes can be mapped to obtain the task priority weight vector. As some implementations, when the neighboring nodes of the current node do not have task priority weight coefficients, for example, when the neighboring nodes of the current node are drone base station nodes, the task priority weight coefficients of the current node can be mapped to obtain the task priority weight vector. As some implementations, when neither the current node nor its neighboring nodes have task priority weight coefficients, for example, when both the current node and its neighboring nodes are drone base station nodes, the task priority weight vector can be a zero vector.
[0057] In some embodiments, the result can be obtained based on the following formula. l The attention weight coefficients between node i and its neighbor j in the layer are calculated. : Here, exp( LeakyReLU represents the exponential function. () represents the linear rectified activation function with leakage. Represents the attention weight vector. express The transpose of , Indicates the first l The learnable weight matrix of the layer is calculated layer by layer. This represents the embedding feature of node i obtained in the previous computational layer (i.e., the (l-1)th layer). This represents the embedding features of neighbor node j obtained in the previous computation layer (i.e., the (l-1)th layer). Let i represent the set of neighboring nodes. This represents the task priority weight vector obtained by mapping the task priority weight coefficient of node i to the task priority weight coefficient of neighboring node j, and || represents the concatenation operation.
[0058] Thus, an attention mechanism can be introduced to obtain the aggregate weights (i.e., attention weight coefficients) of each neighbor node, so as to perform feature weighting based on the corresponding aggregate weights.
[0059] As one implementation method, among the current node's neighboring nodes, those with higher task priority weight coefficients have higher attention weight coefficients. This allows for more accurate message aggregation, resulting in better embedding features for the node.
[0060] In some embodiments, the aggregation result of node i can be obtained based on the following formula: ;here, Indicates the current number l The aggregation results obtained in the layer calculation layer; This represents the attention weight coefficient between node i and its neighbor node j in the current layer l computation; ( ) represents a nonlinear activation function; Represents the set of neighboring nodes of node i; This represents the learnable weight matrix of the current l-th computational layer; Indicates the previous level (i.e., the first level) l The embedding features of neighbor node j obtained in the -1 layer of the computation layer.
[0061] In some embodiments, obtaining the embedding features of a node in the current computational layer 310 by utilizing the embedding features of the node obtained in the previous computational layer 310 and the aggregation result of the node may include obtaining the embedding features of node i in the current computational layer 310 based on the following formula. l Layer 310 computes embedded features : ;here, This represents the embedding feature of node i in the previous layer (i.e., the (l-1)th layer) of computation layer 310. This represents the aggregation result obtained by node i in the current l-th computation layer 310. ( () indicates an update operation. In some embodiments, an update operation includes... After splicing, the data is input into a multilayer perceptron to obtain... .
[0062] Thus, after computation by the Lth computation layer 310, the embedding features of node i in the last (i.e., the Lth) computation layer 310 can be obtained. Thus, the target embedding features of node i can be obtained, i.e. .
[0063] After obtaining the target embedding features of nodes in the heterogeneous graph, a graph neural network model can be used to obtain the first edge weights and second edge weights based on these features. Specifically, the graph neural network model can be used to obtain the first edge weights between the UAV base station node and the communication user node based on their target embedding features, and the graph neural network model can be used to obtain the second edge weights between the UAV base station node and the sensing target node based on their target embedding features.
[0064] In some embodiments, for each candidate association edge in a heterogeneous graph, the weight of the candidate association edge can be obtained based on the target embedding features of the node corresponding to the candidate association edge. The candidate association edges in the heterogeneous graph include edges between UAV base station nodes and communication user nodes, and edges between UAV base station nodes and sensing target nodes. In this way, the weight of the first edge between UAV base station nodes and communication user nodes and the weight of the second edge between UAV base station nodes and sensing target nodes in the heterogeneous graph can be obtained.
[0065] As some implementation methods, any UAV base station node and any communication user node can be connected in pairs, and any UAV base station node and any sensing target node can be connected in pairs, thereby obtaining candidate association edges of the heterogeneous graph.
[0066] Alternatively, as some implementation methods, candidate association edges of the heterogeneous graph can be obtained based on the first initial association relationship between the UAV base station and the communication user, and the second initial association relationship between the UAV base station and the sensing target. In this way, the association relationship can be quickly pre-established based on the initial association relationship and then optimized, thereby improving the association efficiency.
[0067] Specifically, the association method according to some embodiments of this disclosure may further include: obtaining a first initial association relationship between the UAV base station and the communication user, and a second initial association relationship between the UAV base station and the perception target based on the locations of the UAV base station, the communication user, and the perception target; and obtaining the neighboring nodes of the node corresponding to the communication user, the neighboring nodes of the node corresponding to the perception target, and the neighboring nodes of the node corresponding to the UAV base station based on the first initial association relationship and the second initial association relationship.
[0068] As some implementation methods, for communication users You can connect with the nearest one. Each drone can communicate with a base station, meaning it can choose the nearest one. Each drone base station establishes an initial association, specifically with the communication user. Initially associated set of base stations It can be represented as: ;in, Indicates communication user With drone base stations The Euclidean distance between them; For communication users The three-dimensional spatial coordinates, ; For drone base stations The three-dimensional spatial coordinates, .
[0069] As some implementation methods, for the perceived target It can be reached by the nearest one Each drone base station can be used for sensing, allowing it to select the nearest one. Each drone base station establishes an initial association, specifically with the perceived target. Initially associated set of base stations It can be represented as: ;in, Representing the perceived target With drone base stations The Euclidean distance between them; To perceive the target The three-dimensional spatial coordinates, .
[0070] In this way, we can obtain the first initial association between each UAV base station and each communication user in the heterogeneous graph, and the second initial association between each UAV base station and each sensing target in the heterogeneous graph. Here, the first initial association indicates whether an initial association has been established between the corresponding UAV base station and the corresponding communication user, and the second initial association indicates whether an initial association has been established between the corresponding UAV base station and the corresponding sensing target. Thus, nodes that have established initial associations can be neighboring nodes, and nodes that have established initial associations can form candidate association edges in the heterogeneous graph.
[0071] In some embodiments, the association method according to some embodiments of the present disclosure may further include: before obtaining target embedding features of nodes in a heterogeneous graph based on a first feature, a second feature, and a third feature using a graph neural network model, and obtaining a first edge weight and a second edge weight based on the target embedding features of nodes in the heterogeneous graph, for each UAV base station, controlling the UAV base station to establish a connection with the communication user that has established an initial association with the UAV base station according to a first initial association relationship, and controlling the UAV base station to perceive the sensing target that has established an initial association with the UAV base station according to a second initial association relationship.
[0072] In the above embodiments, a fast and low-complexity initial association strategy is implemented based on the first and second initial association relationships. Subsequently, based on the initial association relationships, a graph neural network model is used to optimize each association relationship to determine the final association relationship. In this way, "connect first, optimize later" can be achieved, that is, the most basic multi-connection service can be established for each communication user and each sensing target in the shortest time, forming an initial and feasible network association state, which provides a good starting point for subsequent fine-grained optimization. Specifically, the neighbor nodes of each node can be obtained based on the current initial association relationships for subsequent optimization calculations using the graph neural network model. Compared with directly using the graph neural network model for calculation, this improves computational efficiency, reduces time consumption, and reduces the possibility of excessively long service interruption time.
[0073] In some embodiments, considering that different UAV base stations may share the same communication user or the same communication resources, or that the sensing beams may interfere with each other, different UAV base station nodes can form interference edges of a heterogeneous graph to be used to accurately obtain the neighboring nodes of the UAV base station node and thus more accurately calculate the target embedding features of the UAV base station node.
[0074] In some embodiments, depending on the node type in the heterogeneous graph, the set of neighbor nodes can be represented as follows.
[0075] For drone base station nodes Its neighboring nodes It can be represented as: .
[0076] For communication user nodes Its neighboring nodes It can be represented as: .
[0077] For the sensing target node Its neighboring nodes It can be represented as: .
[0078] here, This represents the set of communicating user nodes in a heterogeneous graph. This represents the set of drone base station nodes in a heterogeneous graph. This represents the set of sensing target nodes in a heterogeneous graph. This represents a interference edge in a heterogeneous graph. This represents a candidate associated edge in a heterogeneous graph.
[0079] like Figure 3 As shown, in some embodiments, the graph neural network model 300 may further include a decoding layer 320 disposed after the multi-layer computation layer 310. The decoding layer 320 may be used to output a first edge weight between the UAV base station node and the communication user node based on the target embedding features of the UAV base station node and the target embedding features of the communication user node. The decoding layer 320 may also be used to output a second edge weight between the UAV base station node and the perceived target node based on the target embedding features of the UAV base station node and the target embedding features of the perceived target node. In some embodiments, the decoding layer 320 may include a multilayer perceptron.
[0080] In some embodiments, edge weights can also be determined based on the task priority weight coefficients of nodes. Specifically, in some embodiments, using a graph neural network model, obtaining the target embedding features of nodes in a heterogeneous graph based on a first feature, a second feature, and a third feature, and obtaining the first edge weight and the second edge weight based on the target embedding features of nodes in the heterogeneous graph can include: obtaining the first edge weight based on the target embedding features of the node corresponding to the UAV base station, the target embedding features of the node corresponding to the communication user, and the task priority weight coefficient of the communication user. In some embodiments, using a graph neural network model, obtaining the target embedding features of nodes in a heterogeneous graph based on a first feature, a second feature, and a third feature, and obtaining the first edge weight and the second edge weight based on the target embedding features of nodes in the heterogeneous graph can include: obtaining the second edge weight based on the target embedding features of the node corresponding to the UAV base station, the target embedding features of the node corresponding to the sensing target, and the task priority weight coefficient of the sensing target.
[0081] After obtaining the first and second side weights, the final association strategy for the sensor-interactive network can be determined based on these weights. Specifically, as follows: Figure 2 As shown, the association method according to some embodiments of this disclosure may further include: Step S230: In response to the first side weight being greater than or equal to the first threshold, determine the association between the drone base station and the communication user; Step S240: In response to the second side weight being greater than or equal to the second threshold, determine the association between the UAV base station and the perceived target.
[0082] Here, in some embodiments, in response to the first side weight being less than a first threshold, it is determined that the drone base station and the communication user are not associated. In some embodiments, in response to the second side weight being less than a second threshold, it is determined that the drone base station and the sensing target are not associated.
[0083] In some embodiments, for each candidate association edge in the heterogeneous graph, the edge weight of the candidate association edge can be compared with a corresponding threshold. For example, if the candidate association edge is between a UAV base station and a communication user, the first edge weight of the candidate association edge is compared with a corresponding first threshold; if the candidate association edge is between a UAV base station and a sensing target, the second edge weight of the candidate association edge is compared with a corresponding second threshold. In this way, the final association relationship between the nodes in the heterogeneous graph can be obtained, that is, the final association relationship between the UAV base station, the communication user, and the sensing target in the corresponding integrated sensing network can be obtained.
[0084] In some embodiments, the first threshold and / or the second threshold may be preset. Alternatively, in some embodiments, the first threshold and / or the second threshold may be determined based on the task priority weight coefficient of the corresponding node. Specifically, the first threshold may be determined based on the magnitude of the task priority weight coefficient of the communication user corresponding to the first side weight, and the second threshold may be determined based on the magnitude of the task priority weight coefficient of the sensing target corresponding to the second side weight.
[0085] In some embodiments, the first threshold corresponds to the task priority weight coefficient of the communication user. Where the task priority weight coefficient of the communication user is selected from multiple first priority weight coefficients, the first threshold is selected from multiple first decision thresholds. Here, the multiple first priority weight coefficients have different values, the multiple first decision thresholds have different values, and each first decision threshold corresponds to one first priority weight coefficient. Thus, in response to a first side weight being greater than or equal to the first decision threshold corresponding to the first priority weight coefficient of the communication user, an association between the UAV base station and the communication user is determined; in response to a first side weight being less than the first decision threshold corresponding to the first priority weight coefficient of the communication user, an unrelated association between the UAV base station and the communication user is determined.
[0086] As one implementation, the sum of multiple first priority weight coefficients can be one. For example, the multiple first priority weight coefficients may include a first weight coefficient, a second weight coefficient less than the first weight coefficient, and a third weight coefficient less than the second weight coefficient, wherein the sum of the first weight coefficient, the second weight coefficient, and the third weight coefficient is one.
[0087] In some embodiments, the second threshold corresponds to the task priority weight coefficient of the perceived target. Where the task priority weight coefficient of the perceived target is selected from multiple second priority weight coefficients, the second threshold is selected from multiple second decision thresholds. Here, the multiple second priority weight coefficients have different values, the multiple second decision thresholds have different values, and each second decision threshold corresponds to one second priority weight coefficient. Thus, in response to a second side weight being greater than or equal to the second decision threshold corresponding to the second priority weight coefficient of the perceived target, an association between the UAV base station and the perceived target is determined; in response to a second side weight being less than the second decision threshold corresponding to the second priority weight coefficient of the perceived target, an unrelated association between the UAV base station and the perceived target is determined.
[0088] As one implementation, the sum of multiple second priority weight coefficients can be one. For example, the multiple second priority weight coefficients may include a fourth weight coefficient and a fifth weight coefficient that is less than the fourth weight coefficient, wherein the sum of the fourth weight coefficient and the fifth weight coefficient is one.
[0089] In the above embodiments, different decision thresholds can be set for different priority weight coefficients. In this way, the corresponding decision thresholds can be adjusted by adjusting the task priority weight coefficients to control the sparsity of the association between the nodes in the sensor network. This can indirectly meet the load constraints of the UAV base station and can be applied to sensor networks with different load constraints, thus realizing flexible association between different sensor networks.
[0090] In some embodiments, the association method according to some embodiments of this disclosure may further include: for each UAV base station, controlling the UAV base station to establish a connection with a communication user associated with the UAV base station, and controlling the UAV base station to sense a sensing target associated with the UAV base station. Thus, based on the final association relationship determined using a graph neural network model, each UAV base station can be controlled to establish a communication network connection with the corresponding communication user and to sense the corresponding sensing target. As some implementations, if the UAV base station has already been initially associated with the corresponding communication user and / or sensing target based on the initial connection relationship, the association relationships between each UAV base station and each communication user, and between each UAV base station and each sensing target, can be updated based on the final association relationship obtained using a graph neural network model.
[0091] In the above embodiments, after receiving instructions to establish associations with corresponding communication users and / or corresponding sensing targets, each UAV base station can establish or adjust communication links with the corresponding communication users and / or establish or adjust sensing links with the corresponding sensing targets. Furthermore, network status can be continuously monitored to periodically or based on event triggers re-run the above association method, achieving dynamic association.
[0092] Figure 6 A flowchart illustrating a training method for a graph neural network model according to some embodiments of the present disclosure is shown. It should be understood that the training method for the graph neural network model according to some embodiments of the present disclosure can be used as a method applied in any of the above-described association methods to obtain the association relationship between a UAV base station and a communication user and a sensing target. That is, any of the above-described association methods may include the training method for the graph neural network model according to some embodiments of the present disclosure; or, the training method for the graph neural network model according to some embodiments of the present disclosure can also be used as a training method independent of any of the above-described association methods.
[0093] like Figure 6 As shown, a method for training a graph neural network model according to some embodiments of this disclosure may include: Step S610: Obtain training samples. The feature part of the training samples includes the first feature of the sample UAV base station in the integrated sensor network, the second feature of the sample communication user, and the third feature of the sample sensing target. The label part of the training samples includes sample association labels, which indicate the sample association relationship between the sample UAV base station and the sample communication user, and the sample association relationship between the sample UAV base station and the sample sensing target in the integrated sensor network. Step S620: Train the graph neural network model based on the training samples.
[0094] Here, the graph neural network model can have corresponding neural network layers (e.g. Figure 3 The computational layer 310 and decoding layer 320 shown can update the learnable parameters of the corresponding neural network layers by training the graph neural network model, so that the trained graph neural network model can obtain the target embedding features of the nodes in the heterogeneous graph corresponding to the synesthetic network based on the first feature, the second feature and the third feature, and obtain the first side weight and the second side weight based on the target embedding features of the nodes in the heterogeneous graph.
[0095] In some embodiments, a graph neural network model can be trained based on training samples and a target loss function. Specifically, the graph neural network model can be trained based on training samples with the goal of reducing the value of the target loss function. For example, appropriate optimization algorithms can be used to update the parameters of the graph neural network model to minimize the value of the target loss function, thereby completing the training of the graph neural network model.
[0096] As some implementation methods, the target loss function can be set based on the sample association labels of the training samples and the first and second predicted edge weights output by the graph neural network model based on the first feature of the sample, the second feature of the sample communication user, and the third feature of the sample sensing target. Here, the first predicted edge weight is the edge weight between the sample UAV base station node and the sample communication user node in the heterogeneous graph corresponding to the sample integrated sensing network, and the second predicted edge weight is the edge weight between the sample UAV base station node and the sample sensing target node. In some embodiments, the target loss function may include a binary cross-entropy loss function set based on the sample association labels of the training samples and the predicted edge weights output by the graph neural network model.
[0097] As some implementation methods, the target loss function may include one or more binary cross-entropy loss functions, wherein each binary cross-entropy loss function corresponds to a candidate association edge in the heterogeneous graph corresponding to the sample synesthetic network, and each binary cross-entropy loss function is set based on the predicted edge weight of the corresponding candidate association edge in the heterogeneous graph corresponding to the sample synesthetic network, the sample association relationship of the corresponding candidate association edge in the sample association label, and the task priority weight coefficient of the corresponding candidate association edge. Here, the edges between the sample UAV base station node and the sample communication user node, and the edges between the sample UAV base station node and the sample sensing target node can constitute candidate association edges in the heterogeneous graph corresponding to the sample synesthetic network.
[0098] For example, the target loss function It can be set based on the following formula: ; here, This represents the candidate association edges in the heterogeneous graph corresponding to the sample synesthetic network. This indicates the number of candidate associated edges. The edge (i,j) between node i and node j is considered a candidate associated edge. Let (i,j) be the task priority weight coefficient, where... The priority weight coefficients can be determined based on the task priority weight coefficients of the communication users connected by edge (i,j) or the task priority weight coefficients of the sensing target. Let (i,j) be the sample association relationship indicated by the sample association label. The predicted edge weights (e.g., first predicted edge weights or second predicted edge weights) of edge (i,j) obtained based on the graph neural network model.
[0099] In some embodiments according to this disclosure, the training of a graph neural network model can be based on a Markov Decision Process (MDP).
[0100] Specifically, the state at time t The graph state at time t can be set, where the graph state can include the relationships between nodes in the integrated sensing network and node characteristics. Here, when the node is a UAV base station node, the node characteristic can be the first characteristic; when the node is a communication user node, the node characteristic can be the second characteristic; and when the node is a sensing target node, the node characteristic can be the third characteristic.
[0101] Action at time t It can be set to predict the association between nodes or predict the edge weights (e.g., the first predicted edge weight or the second predicted edge weight) based on the graph neural network model. Here, the predicted association can be based on the predicted edge weights. For example, when the predicted edge weight is greater than the corresponding decision threshold, it can be determined that the corresponding nodes are associated.
[0102] Reward function at time t It can be based on the state implement The rewards are set based on the communication performance rewards, perception performance rewards, backhaul load penalties, and priority rewards.
[0103] In some embodiments, the reward function It can be set based on the following formula: .
[0104] here, The weights are adjustable, meaning they can be adjusted according to the specific scenario. Used to encourage improvements in communication and sensing performance. Used to punish return overload. In some embodiments, this is used to reward high-priority task assurance. It can be greater than .
[0105] Priority-weighted communication rates The perception rate is a priority-weighted value, where, and They represent Time Communication Users Communication rate and sensing target The rate of perception.
[0106] The return load function can be set as follows: . It can represent the communication user at time t. The backhaul link capacity occupied by the communication data This can represent the perception of the target at time t. The backhaul link capacity occupied by the sensing data This can represent the backhaul link load capacity of drone base station n at time t. N is the total number of drone base stations in the integrated sensing and communication network.
[0107] To ensure that the indicator function is satisfied for high-priority tasks, it can be set as follows: .in, For indicator functions, if If the expression inside is true, then ,otherwise . and Let these represent the set of communication users with high task priority and the set of sensing targets with high task priority, respectively. For example, This refers to the set of communication users whose task priority weight coefficient is greater than or equal to the specified weight coefficient. This refers to the set of perceived targets whose task priority weight coefficient is greater than or equal to the specified weight coefficient. Indicates communication user The communication rate QoS requirement threshold, Representing the perceived target The perceived rate QoS requirement threshold.
[0108] Thus, the parameters of a graph neural network can be updated based on a Markov decision process with the goal of maximizing the cumulative reward obtained from the reward function. As some implementation methods, policy gradient algorithms (such as PPO and DDPG) can be used to continuously update the parameters of the GNN.
[0109] In the above embodiments, based on Markov decision processes, graph neural network models can make high-reward decisions under various complex network states, i.e., output high-reward decision results. This end-to-end online training method can eliminate the need for manual annotation, such as manually labeling related information, thus reducing labor costs.
[0110] Embodiments of this disclosure also provide an association device for a sensor-integrated network, which can be configured to perform the association method of any of the above embodiments.
[0111] In some embodiments, the association apparatus may include a module for performing the association method of any of the above embodiments.
[0112] Figure 7 A schematic diagram of an associated device according to some embodiments of the present disclosure is shown. For example... Figure 7 As shown, the association device 700 according to some embodiments of the present disclosure may include an acquisition module 710, a weight determination module 720 and an association determination module 730.
[0113] The acquisition module 710 can be configured to acquire a first feature of the UAV base station, a second feature of the communication user, and a third feature of the perceived target.
[0114] The weight determination module 720 can be configured to use a graph neural network model to obtain target embedding features of nodes in a heterogeneous graph based on the first feature, the second feature, and the third feature, and to obtain a first edge weight and a second edge weight based on the target embedding features of nodes in the heterogeneous graph. Here, the UAV base station, the communication user, and the sensing target constitute nodes in the heterogeneous graph. The first edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the communication user in the heterogeneous graph, and the second edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the sensing target. The association determination module 730 can be configured to determine the association between the UAV base station and the communication user in response to the first edge weight being greater than or equal to a first threshold, and to determine the association between the UAV base station and the sensing target in response to the second edge weight being greater than or equal to a second threshold.
[0115] In some embodiments, the association device 700 may further include other modules to perform the association method of any of the above embodiments.
[0116] For details on the operation of the various modules in the association device 700, please refer to the detailed explanation of the association method above, which will not be repeated here.
[0117] Figure 8 A schematic diagram of an associated device according to some embodiments of the present disclosure is shown. For example... Figure 8 As shown, the association device 800 may include a memory 810 and a processor 820 coupled to the memory 810. The processor 820 may be configured to execute the association method of any of the foregoing embodiments based on instructions stored in the memory 810.
[0118] Specifically, processor 820 can perform various actions and processes according to instructions stored in memory 810. Processor 820 can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor can be a microprocessor or any conventional processor, and can be an x86 architecture or an ARM architecture, etc.
[0119] Memory 810 stores executable instructions that, when executed by processor 820, implement the associated methods described above. Memory 810 may be volatile memory or non-volatile memory, or may include both. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.
[0120] Embodiments of this disclosure also provide an aerial platform station. In some embodiments, the aerial platform station can be configured to perform the association method described in any of the above embodiments. As some implementations, the aerial platform station may include the association device 700 described in any of the above embodiments. As some implementations, the aerial platform station may include the association device 800 described in any of the above embodiments.
[0121] Embodiments of this disclosure also provide a sensing-integrated network. In some embodiments, the sensing-integrated network may include the high-altitude platform station described above. In some embodiments, the sensing-integrated network may further include unmanned aerial vehicle (UAV) base stations, communication users, and sensing targets.
[0122] This disclosure also proposes a non-transitory computer-readable storage medium storing instructions that, when executed by a processor, can implement the steps of the association method as described above.
[0123] Similarly, the non-transitory computer-readable storage media in the embodiments of this disclosure are intended to include, but are not limited to, the above and any other suitable types of memory.
[0124] This disclosure also proposes a computer program product that may include instructions that, when executed by a processor, implement the steps of the associated method as described above.
[0125] Instructions can be any set of instructions that will be executed directly by one or more processors, such as machine code, or any set of instructions that will be executed indirectly, such as a script. The terms “instruction,” “application,” “procedure,” “step,” “program,” and “computer program” used herein are used interchangeably. Instructions can be stored in object code format for direct processing by one or more processors, or stored in any other computer language, including scripts or sets of independent source code modules that are interpreted on demand or compiled in advance. The function, methods, and routines of instructions are explained in more detail in other parts of this document.
[0126] Figure 9A schematic block diagram of a computer system 900 on which embodiments of the present disclosure may be implemented is shown. The computer system 900 includes a bus 910 or other communication mechanism for transmitting information, and a processing means 920 coupled to the bus 910 for processing information. The computer system 900 also includes a storage means 930 coupled to the bus 910 for storing instructions to be executed by the processing means 920. The storage means 930 may include random access memory (RAM) or other dynamic storage devices. The storage means 930 may be used to store temporary variables or other intermediate information during the execution of instructions to be executed by the processing means 920. The storage means 930 may include a read-only memory (ROM) or other static storage device for storing static information and instructions for the processing means 920. The storage means 930 may include means such as a magnetic disk or optical disk for storing information and instructions. The computer system 900 may be coupled via the bus 910 to an output device 940 for providing output to a user, such as, but not limited to, a display (such as a cathode ray tube (CRT) or liquid crystal display (LCD)), a speaker, etc. Input devices 950, such as a keyboard, mouse, and microphone, are coupled to bus 910 for transmitting information and command selections to processing device 920. Computer system 900 can execute embodiments of this disclosure. Consistent with certain implementations of this disclosure, results are provided by computer system 900 in response to processing device 920 executing one or more sequences of one or more instructions contained in storage device 930. Such instructions may be read into storage device 930 from another computer-readable medium, such as storage device 930. Execution of the sequence of instructions contained in storage device 930 causes processing device 920 to perform the methods described herein. Alternatively, hard-wired circuitry may be used in place of or in combination with software instructions to implement the teachings. Therefore, implementations of this disclosure are not limited to any particular combination of hardware circuitry and software. In various embodiments, computer system 900 may be connected across a network via network interface 960 to one or more other computer systems, such as computer system 900, to form a networked system. This network may include a private network or a public network such as the Internet. In a networked system, one or more computer systems may store data and supply data to other computer systems. As used herein, the term "computer-readable medium" refers to any medium that participates in providing instructions to the processing device 920 for execution. Such media can take many forms, including but not limited to non-volatile media, volatile media, and transmission media. Non-volatile media include, for example, optical discs or magnetic disks. Volatile media include dynamic memory such as RAM. Transmission media include coaxial cables, copper wires, and optical fibers, including wiring that includes bus 910.Common forms of computer-readable media or computer program products include, for example, floppy disks, flexible disks, hard disks, magnetic tapes, or any other magnetic media, CD-ROMs, digital video discs (DVDs), Blu-ray discs, any other optical media, thumb drives, memory cards, RAM, PROMs and EPROMs, fast EPROMs, any other memory chips or cartridges, or any other tangible media from which a computer can read. Various forms of computer-readable media may be involved when carrying one or more sequences of one or more instructions to processing device 920 for execution. For example, instructions may initially be carried on a disk of a remote computer. The remote computer may load the instructions into its dynamic memory and transmit the instructions over a telephone line using a modem. A modem local to computer system 900 may receive data over a telephone line and convert the data into an infrared signal using an infrared transmitter. An infrared detector coupled to bus 910 may receive the data carried in the infrared signal and place the data on bus 910. Bus 910 carries the data to storage device 930, where processing device 920 retrieves and executes the instructions. Optionally, the instructions received by the storage device 930 may be stored on the storage device 930 before or after they are executed by the processing device 920.
[0127] According to various embodiments, instructions configured to be executed by processing device 920 to perform a method are stored on a computer-readable medium. The computer-readable medium may be a device for storing digital information. For example, the computer-readable medium includes a compact disc read-only memory (CD-ROM) as known in the art for storing software. The computer-readable medium is accessed by a processor adapted to execute the instructions configured to be executed.
[0128] In the technical solution disclosed herein, communication users, sensing targets, and UAV base stations are uniformly modeled as heterogeneous graph nodes. Candidate association edges are constructed between "UAV base station - communication user" and "UAV base station - sensing target," breaking through the limitations of the traditional single-connection mode and supporting flexible and redundant many-to-many connections between users and base stations, providing a structured data foundation for global resource collaboration. The solution disclosed herein effectively solves the complex association decision-making problem caused by the coexistence of multiple communication users and multiple sensing targets in post-disaster scenarios. According to some embodiments of this disclosure, the message passing and aggregation mechanism of GNN is used to automatically learn the complex high-dimensional coupling relationships between network nodes and output corresponding association weights for each candidate association edge, so as to achieve end-to-end joint optimization of communication and sensing performance. This overcomes the shortcomings of traditional algorithms that rely on precise mathematical models and are difficult to capture nonlinear relationships, significantly improving the global optimality of the strategy and the model's generalization ability. Furthermore, according to some embodiments of this disclosure, a flexible association decision-making scheme based on priority-aware thresholds is proposed. Specifically, differentiated decision thresholds are set for communication users and sensing targets with different priorities, so as to perform threshold-based decisions based on the weights predicted by the GNN, generating the final many-to-many association strategy. In this way, while ensuring that nodes of high-priority tasks are associated, the interpretability of the decision and the system's adaptability in resource-constrained scenarios are enhanced. According to the scheme of this disclosure, load balancing and resource fragmentation integration can be dynamically achieved under conditions of limited backhaul links, while meeting the differentiated QoS requirements of communication and sensing, effectively improving resource utilization efficiency, service reliability, and the ability to guarantee high-urgent tasks.
[0129] The embodiments disclosed herein can be applied to electronic devices such as terminal devices, computer systems, and servers, and can operate together with a wide range of other general-purpose or special-purpose computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments, and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, and servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments including any of the above systems, etc.
[0130] In this disclosure, electronic devices such as terminal devices, computer systems, and servers can be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. Typically, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in a distributed cloud computing environment, where tasks are performed by remote processing devices linked via a communication network. In a distributed cloud computing environment, program modules can reside on local or remote computing system storage media, including storage devices.
[0131] The technical solutions of the various embodiments of this application can be applied to various existing or future-developed communication systems, such as: 5th generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, and sixth generation mobile communication systems.
[0132] The technical solutions provided in this application can also be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems.
[0133] In this disclosure, terminal equipment, also referred to as user terminal, terminal, etc., is a device with wireless transceiver capabilities that can communicate with one or more core networks (CNs) via access network devices in a radio access network (RAN). Terminal equipment can be deployed on land, including indoors or outdoors, handheld, wearable, or vehicle-mounted; it can also be deployed on water, such as on ships; and it can be deployed in the air, such as on airplanes, balloons, or satellites. Terminal equipment can be a mobile phone, tablet computer, computer with wireless transceiver capabilities, virtual reality (VR) terminal equipment, augmented reality (AR) terminal equipment, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical care, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, etc.
[0134] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0135] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.
[0136] As used herein, the term “exemplary” means “used as an example, instance, or illustration” and not as a “model” to be exactly copied. Any implementation described herein by example is not necessarily to be construed as preferred or advantageous over other implementations.
[0137] Furthermore, terms such as “first,” “second,” etc., may be used in this document for reference purposes only and are not intended to be limiting. For example, unless the context clearly indicates otherwise, the words “first,” “second,” and other such numerical terms relating to structures or elements do not imply order or sequence.
[0138] It should also be understood that when the term “including / contains” is used herein, it indicates the presence of the indicated feature, whole, step, operation, unit and / or component, but does not preclude the presence or addition of one or more other features, wholes, steps, operations, units and / or components and / or combinations thereof.
[0139] In this disclosure, the term “provide” is used broadly to cover all ways of obtaining an object, and therefore “provide an object” includes, but is not limited to, “purchasing,” “preparing / manufacturing,” “arranging / setting up,” “installing / assembling,” and / or “ordering” an object.
[0140] As used herein, the term “and / or” includes any and all combinations of one or more of the listed items in association. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure. As used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise.
[0141] Those skilled in the art will recognize that the boundaries between the above operations are merely illustrative. Multiple operations may be combined into a single operation, a single operation may be distributed among additional operations, and operations may be performed with at least partial overlap in time. Moreover, alternative embodiments may include multiple instances of a particular operation, and the order of operations may be changed in various other embodiments. However, other modifications, variations, and substitutions are equally possible. Aspects and elements of all the embodiments disclosed above may be combined in any way and / or in combination with aspects or elements of other embodiments to provide multiple additional embodiments. Therefore, this specification and the accompanying drawings should be considered illustrative rather than restrictive.
[0142] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. The various embodiments disclosed herein can be combined in any way without departing from the spirit and scope of this disclosure. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.
Claims
1. A correlation method for a sensor-integrated network, comprising: Acquire the first feature of the UAV base station, the second feature of the communication user, and the third feature of the perceived target; Using a graph neural network model, target embedding features of nodes in a heterogeneous graph are obtained based on the first feature, the second feature, and the third feature. First edge weights and second edge weights are obtained based on the target embedding features of nodes in the heterogeneous graph. The heterogeneous graph corresponds to the integrated sensing network. UAV base stations, communication users, and sensing targets constitute nodes in the heterogeneous graph. The first edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the communication user in the heterogeneous graph. The second edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the sensing target. In response to the first side weight being greater than or equal to the first threshold, the association between the drone base station and the communication user is determined; In response to the second side weight being greater than or equal to the second threshold, the association between the drone base station and the perceived target is determined.
2. The association method according to claim 1, wherein, The first feature includes at least one of the following: the location of the UAV base station, remaining communication resources, remaining sensing resources, the maximum data transmission rate supported by the backhaul link, and the occupied backhaul link capacity. The second feature includes at least one of the following: the location of the communication user, the communication rate quality of service requirement threshold, and the task priority weight coefficient; The third feature includes at least one of the following: the location of the perceived target, the perception rate service quality requirement threshold, and the task priority weight coefficient.
3. The association method according to claim 1, wherein, The graph neural network model includes multiple computational layers. Using the graph neural network model, target embedding features of nodes in the heterogeneous graph are obtained based on the first feature, the second feature, and the third feature. The first edge weight and the second edge weight are obtained based on the target embedding features of nodes in the heterogeneous graph, including: In each of the multi-layered computational layers, for each node in the heterogeneous graph, the aggregation result of the node is obtained by using the embedding features of the node's neighboring nodes obtained in the previous computational layer. By utilizing the embedding features obtained by the node in the previous computational layer and the aggregation result of the node, the embedding features of the node in this computational layer are obtained. Among them, the embedding features of the node in the last computational layer of the multi-layered computational system are used as the target embedding features of the node. In the first computing layer of the multi-layer computing layers, the initial embedding feature of a node is used as the embedding feature obtained by the node in the previous computing layer. The initial embedding feature of the node corresponding to the UAV base station is obtained based on the first feature, the initial embedding feature of the node corresponding to the communication user is obtained based on the second feature, and the initial embedding feature of the node corresponding to the sensing target is obtained based on the third feature.
4. The association method according to claim 3, wherein, The association method further includes: Based on the locations of the UAV base station, communication users, and sensing targets, a first initial association between the UAV base station and the communication users, and a second initial association between the UAV base station and the sensing targets are obtained; Based on the first and second initial association relationships, the neighboring nodes of the node corresponding to the communication user, the neighboring nodes of the node corresponding to the sensing target, and the neighboring nodes of the node corresponding to the UAV base station are obtained.
5. The association method according to claim 4, wherein, The association method further includes: Before using a graph neural network model to obtain the target embedding features of nodes in a heterogeneous graph based on the first feature, the second feature, and the third feature, and before obtaining the first edge weight and the second edge weight based on the target embedding features of nodes in a heterogeneous graph, for each UAV base station, the system controls the UAV base station to establish a connection with the communication user that has established an initial association with the UAV base station according to the first initial association relationship, and controls the UAV base station to perceive the sensing target that has established an initial association with the UAV base station according to the second initial association relationship.
6. The association method according to claim 3, wherein, The graph neural network model is a graph attention network model. It utilizes the embedding features of the node's neighboring nodes obtained in the previous computational layer to obtain the aggregation result for that node, including: For each neighboring node of the node, the attention weight coefficient between the node and the neighboring node is obtained based on the task priority weight vector, the attention weight vector, the embedding features obtained by the node in the previous computation layer, and the embedding features obtained by the neighboring node in the previous computation layer. The task priority weight vector is obtained based on the task priority weight coefficient of the node and the task priority weight coefficient of the neighboring node. Based on the embedding features of each neighboring node obtained in the previous computation layer and the attention weight coefficients between the node and each neighboring node, the aggregation result of the node is obtained.
7. The association method according to claim 1, wherein, Using a graph neural network model, target embedding features of nodes in a heterogeneous graph are obtained based on the first feature, the second feature, and the third feature. The first edge weight and the second edge weight are then obtained based on the target embedding features of nodes in the heterogeneous graph, including: Based on the target embedding features of the node corresponding to the UAV base station, the target embedding features of the node corresponding to the communication user, and the task priority weight coefficient of the communication user, the first side weight is obtained; and / or The second-side weights are obtained based on the target embedding features of the node corresponding to the UAV base station, the target embedding features of the node corresponding to the perceived target, and the task priority weight coefficient of the perceived target.
8. The association method according to claim 7, wherein: The first threshold corresponds to the task priority weight coefficient of the communication user. When the task priority weight coefficient of the communication user is selected from multiple first priority weight coefficients, the first threshold is selected from multiple first decision thresholds. The multiple first priority weight coefficients have different values, the multiple first decision thresholds have different values, and each first decision threshold corresponds to one first priority weight coefficient; and / or The second threshold corresponds to the task priority weight coefficient of the perceived target. When the task priority weight coefficient of the perceived target is selected from multiple second priority weight coefficients, the second threshold is selected from multiple second decision thresholds. The magnitudes of the multiple second priority weight coefficients are different, the magnitudes of the multiple second decision thresholds are different, and each second decision threshold corresponds to a second priority weight coefficient.
9. The association method according to claim 1, wherein, The association method further includes: For each drone base station, control the drone base station to establish connections with the communication users associated with the drone base station, and control the drone base station to perceive the sensing targets associated with the drone base station.
10. The association method according to claim 1, wherein, The association method further includes: The training samples are obtained. The feature part of the training samples includes the first feature of the sample UAV base station in the integrated sensor network, the second feature of the sample communication user, and the third feature of the sample sensing target. The label part of the training samples includes the sample association label, which indicates the sample association relationship between the sample UAV base station and the sample communication user in the integrated sensor network, and the sample association relationship between the sample UAV base station and the sample sensing target. A graph neural network model is trained based on training samples.
11. The association method according to claim 10, wherein, The graph neural network models trained based on training samples include: The graph neural network model is trained based on training samples, with the goal of reducing the value of the target loss function. The target loss function includes one or more binary cross-entropy loss functions. Each binary cross-entropy loss function corresponds to a candidate association edge in the heterogeneous graph corresponding to the sample synesthetic network. Each binary cross-entropy loss function is set based on the predicted edge weight of the corresponding candidate association edge in the heterogeneous graph corresponding to the sample synesthetic network, the sample association relationship of the corresponding candidate association edge in the sample association label, and the task priority weight coefficient of the corresponding candidate association edge. The predicted edge weight is obtained using a graph neural network model. The edges between the sample UAV base station node and the sample communication user node, and the edges between the sample UAV base station node and the sample perception target node constitute candidate association edges.
12. The association method according to claim 1, wherein, Graph neural network models are trained based on Markov decision processes, where the reward function in a Markov decision process is based on the state... implement The status is set based on the communication performance reward, perception performance reward, backhaul load penalty, and priority reward. This includes the relationships and characteristics of nodes in the integrated sensory network at time t, and actions. It is set to the predicted association relationship or predicted edge weight between nodes obtained based on the graph neural network model.
13. An association device for a sensor-integrated network, comprising: The acquisition module is configured to acquire the first feature of the UAV base station, the second feature of the communication user, and the third feature of the perceived target; The weight determination module is configured to: use a graph neural network model to obtain the target embedding features of nodes in the heterogeneous graph based on the first feature, the second feature and the third feature, and obtain the first edge weight and the second edge weight based on the target embedding features of nodes in the heterogeneous graph, wherein the UAV base station, the communication user and the sensing target constitute the nodes in the heterogeneous graph, the first edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the communication user in the heterogeneous graph, and the second edge weight is the edge weight between the node corresponding to the UAV base station and the node corresponding to the sensing target. The association determination module is configured to: determine the association between the UAV base station and the communication user in response to the weight of the first side being greater than or equal to the first threshold, and determine the association between the UAV base station and the sensing target in response to the weight of the second side being greater than or equal to the second threshold.
14. An association device for a sensor-integrated network, comprising a processor and a memory, wherein the memory stores instructions that, when executed by the processor, implement the steps of the association method according to any one of claims 1 to 12.
15. An aerial platform station, comprising: The associated device according to claim 13; or The associated device according to claim 14.
16. A non-transitory computer-readable storage medium, wherein, The non-transitory computer-readable storage medium stores instructions that, when executed by a processor, implement the steps of the association method according to any one of claims 1 to 12.
17. A computer program product comprising instructions that, when executed by a processor, implement the steps of the association method according to any one of claims 1 to 12.
