Flight ad hoc network routing method and system based on graph neural network and mobile perception

Abstract

The invention discloses a flight ad hoc network routing method and system based on a graph neural network and mobile perception, and belongs to the technical field of wireless communication. The method comprises the steps that the unmanned aerial vehicle periodically uploads node information of the unmanned aerial vehicle to a high-altitude computing platform by using an information collection protocol; the high-altitude computing platform uses the uploaded node information to construct a network condition diagram as input of graph neural network model training, and the graph neural network model is trained and used for predicting a network condition diagram at the next moment; when the nodes perform routing requests, the high-altitude computing platform issues a command to collect information of all the nodes, and a network condition diagram is formed; and inputting the network condition diagram into the trained graph neural network model to obtain a network condition diagram at the next moment, and selecting an optimal route by using a shortest path algorithm and issuing the optimal route to the unmanned aerial vehicle requesting the route. The future network condition diagram is predicted through the graph neural network model, so that the accuracy and the intelligent level of routing selection are improved, the packet arrival rate is improved, and the network throughput is optimized.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology, and in particular relates to a flying ad hoc network routing method and system based on graph neural network and mobile perception. Background Art

[0002] Recently, drones or unmanned aerial vehicles have become increasingly popular in military and civilian applications such as aerial photography, communications, agriculture, and search and rescue. The use of drones in these applications requires reliable and fast wireless communications, which has led to the development of flying ad hoc networks, where multiple drones work together to achieve greater capabilities such as cost-effectiveness, scalability, survivability, and higher speed. Flying ad hoc networks (FANETs) have unique routing challenges because FANETs involve highly dynamic and three-dimensional mobility, which creates challenges in maintaining reliable and efficient routing paths as drones often change locations and encounter different network conditions.

[0003] In flying ad hoc networks (FANET), most existing FANET routing protocols are designed based on mobility or topology information, ignoring the traffic distribution in the network. As a result, nodes located in severely congested areas may cause large end-to-end delays, untimely path adjustments, and packet loss. However, although existing prediction-based FANET routing protocols perform congestion control, they often do not take into account the topological relationship between nodes, resulting in inconsistencies between the prediction results and the actual results.

[0004] To this end, when facing congestion control, the present invention proposes a flying ad hoc network routing method based on graph neural network and mobile perception to solve the above technical problems. Summary of the invention

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a flying ad hoc network routing method and system based on graph neural network and mobile perception, which solves the problems in the prior art.

[0006] The purpose of the present invention can be achieved through the following technical solutions: The flying ad hoc network routing method based on graph neural network and mobile perception includes the following steps: The drone periodically uses the information collection protocol to upload its own node information to the high-altitude computing platform; The high-altitude computing platform uses the uploaded node information to construct a network status graph as the input for graph neural network model training, trains the graph neural network model, updates the model parameters, and is used to predict the network status graph at the next moment; When a node makes a routing request, the high-altitude computing platform sends a command to collect all node information to form a network status graph; the network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment, and then the shortest path algorithm is used to select the best route and send it to the drone that made the routing request.

[0007] Furthermore, the flight ad hoc network includes: users, drones and a high-altitude computing platform; Each user is associated with the drone with the strongest received signal strength. When a user has a communication need, the data packet will be uploaded to the associated drone. Finding the route of the user data packet is to find the route of the associated drone, and the drone is regarded as a node; After receiving the data packet, the UAV sends a routing request to the high-altitude computing platform. The high-altitude computing platform sends an information collection command to the UAV. After collecting the UAV information, it pre-processes it to form a network status graph. The network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment. The shortest path algorithm is then used to select the best route and send it to the UAV that requested the route. After receiving the best route, the UAV forwards the data packet sent by the user.

[0008] Furthermore, the steps of uploading the drone's own node information to the high-altitude computing platform are: All nodes generate and send NLIP messages periodically; After receiving the NLIP message, the neighbor node checks the validity of the message to determine whether it is a duplicate message. If it is a non-duplicate message, it forwards it to other neighbor nodes and increments the hop count field in the message to indicate that the hop count of the path has increased. All nodes broadcast NLIP messages until the high-altitude computing platform collects complete information about all nodes in the flight ad hoc network. The high-altitude computing platform selects multi-point relay nodes based on the collected network information. Each node periodically sends source announcement data to the multi-point relay node according to the information collection protocol, and updates the node routing table according to the network status. If a neighbor node loses connection or does not receive the source announcement data, the neighbor node is removed from the node routing table.

[0009] Furthermore, the graph neural network model includes multiple spatiotemporal convolution blocks, each of which includes two gated sequential convolution layers and a spatial graph convolution layer located between the two gated sequential convolution layers.

[0010] Furthermore, the training process of the graph neural network model is: The node features are preprocessed by MinMaxScaler normalization, and the adjacency matrix is ​​constructed according to the FANET topology; Initialize the weights and parameters of the graph neural network model, use the preprocessed node features and adjacency matrix to train the graph neural network model, define the loss function as the mean square error during the training process, and add L1 penalty to control the complexity of the graph neural network model. At the same time, the adaptive learning rate is determined using the Adam optimizer.

[0011] Furthermore, the specific steps of selecting the best route and sending it to the UAV that requested the route are: The business connects to the node, requests transmission, and the node reports the business type and routing request to the high-altitude computing platform; The high-altitude computing platform sends commands through relay nodes to let all nodes upload current node parameter information, construct the node feature matrix of FANET, and then weight the connection matrix of FANET network according to the collected link maintenance time in FANET network, expressed as an adjacency matrix, and obtain the network status graph; then the network status graph is input into the trained graph neural network model for prediction, and the network status graph at the next moment is obtained; The high-altitude computing platform obtains the predicted network status diagram at the next moment, normalizes the metric factors of the node, multiplies them by the influence factors of each metric, and weights them to obtain the node cost. The cost of the node with the larger cost value on both sides of the link is taken as the link cost, and the network cost diagram is obtained. The shortest algorithm is used to solve the minimum cost, that is, the optimal route; The high-altitude computing platform sends the best routing table to the routing request node and updates the node routing table.

[0012] The flying ad hoc network routing system based on graph neural network and mobile perception includes: Route discovery module: The drone periodically uses the information collection protocol to upload its own node information to the high-altitude computing platform; Model training module: The high-altitude computing platform uses the uploaded node information to construct a network status graph as the input for graph neural network model training, trains the graph neural network model, updates the model parameters, and is used to predict the network status graph at the next moment; And, the routing optimization module: when a node makes a routing request, the high-altitude computing platform sends a command to collect all node information to form a network status graph; the network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment, and then the shortest path algorithm is used to select the best route and send it to the drone that made the routing request.

[0013] A computer storage medium stores a readable program, which can execute the above-mentioned flying self-organizing network routing method based on graph neural network and mobile perception when the program is running.

[0014] An electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned flying self-organizing network routing method based on graph neural network and mobile perception.

[0015] A computer program product includes computer instructions, wherein the computer instructions instruct a computing device to perform operations corresponding to the above-mentioned flying self-organizing network routing method based on graph neural network and mobile perception.

[0016] Beneficial effects of the present invention: 1. The present invention provides a new node and link information flow data packet for network information transmission, route discovery and networking. When the network topology changes, the data packet structure and content are dynamically adjusted, and the information transmission path can be adaptively optimized according to the real-time status of the network, thereby significantly improving the efficiency and reliability of information transmission.

[0017] 2. The present invention introduces mobile perception and congestion control based on graph neural networks on the basis of traditional network routing maintenance. By integrating network performance evaluation parameters and combining the learning ability of graph neural networks, the system can predict the network status diagram at the next moment to build a more comprehensive and intelligent network performance evaluation model, thereby improving the adaptability and intelligence of the network.

[0018] 3. The present invention constructs a more comprehensive and accurate network performance evaluation model by considering node residual energy, node average link maintenance time and node residual buffer size as metrics for routing decisions, providing an important basis for improving network reliability, efficiency and intelligent management.

[0019] 4. The present invention uniformly allocates routes by a high-altitude computing platform, realizes the function of on-demand routing according to business types, and also runs on the data link layer, transmits routing information through Ethernet frames, reduces data packet processing delay, and compared with the routing strategy based on the shortest path in the prior art, it considers multiple channel parameters and adaptively selects routes. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 It is a schematic diagram of the structure of the flight ad hoc network of the present invention; Figure 2 It is a schematic diagram of the protocol routing discovery process of the present invention; Figure 3 It is a flow chart of the flight ad hoc network routing method based on graph neural network and mobile perception of the present invention; Figure 4 It is a curve diagram of the system packet transmission success rate and the number of drones achieved by the algorithm proposed in the present invention under the same packet grouping arrival rate and rate of drones; Figure 5 It is a curve diagram of the system packet transmission success rate and the packet grouping arrival rate achieved by the algorithm proposed in the present invention under the same number and rate of drones; Figure 6 It is a curve graph of the system packet transmission success rate and drone rate achieved by the proposed algorithm under the same number of drones and packet grouping arrival rate. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Example 1 like Figure 1 As shown, the present invention proposes a flying ad hoc network (FANET) based on graph neural network and mobile perception, including: users, drones and high-altitude computing platforms; Each user is associated with the drone with the strongest received signal strength (RSS). When a user has a communication need, the data packet will be uploaded to the associated communication drone. Because the user is associated with the drone, we find the route of the user's data packet is to find the route of the drone, so we directly regard the drone as a node to more conveniently find the data packet route.

[0024] After receiving the data packet, the communication drone sends a routing request to the high-altitude computing platform, and the high-altitude computing platform sends an information collection command to the drone.

[0025] In order to better implement the graph neural network model, the high-altitude computing platform defines a node and link information flow packet (NLIP) that shares node / link information throughout the FANET network. NLIP is defined as containing node and link information. Table 1 shows the structure of NLIP: Table 1 Node and link information flow data packet format It can be packed into a UDP packet. The checksum header field indicates that this is a routing message. The count field is used to avoid duplicate message delivery. It also helps to update the local mapping information saved in the buffer. The remaining fields contain two tables that contain the node and link information of the neighborhood.

[0026] The node information table contains the IP, buffer status, remaining energy, location information and clustering information of each node. The link information table contains the start and end nodes, received signal strength and estimated duration of each link.

[0027] After receiving the information collection command, the UAV uploads the NLIP containing the specific node and link information of its own node to the high-altitude computing platform.

[0028] After receiving the NLIP, the high-altitude computing platform collects the specific node and link information of FANET, pre-processes the node features and adjacency matrix to form a network status graph; the network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment, and then the shortest path algorithm is used to select the best route and send it to the drone that requested the route. After receiving the best route, the drone forwards the data packets sent by the user.

[0029] Example 2 like Figure 3 As shown, the flying ad hoc network routing method based on graph neural network and mobile perception includes the following steps: S1, the drone periodically uses the information collection protocol to upload its own node information to the high-altitude computing platform, which is the routing discovery process; The route discovery process in the flight ad hoc network is as follows: Figure 2 As shown, the following steps are included: S11 specifies a new node and link information flow packet (NLIP), which contains information about the node itself and its links with neighboring nodes, and is used for network information transmission, route discovery and networking; S12, all nodes periodically generate and send NLIP (Node and Link Information Flow Packet) messages; The NLIP message contains key information such as the unique identifier of the sending node, sequence number, hop count, and time-to-live (TTL); the purpose of the NLIP message is to let other nodes in the network know the existence of the source node and provide a basis for the route discovery process. The NLIP message is sent to the surrounding nodes by broadcasting, and the propagation range is controlled by the TTL field. When the TTL is reduced to 0, the NLIP message will be discarded, limiting its maximum propagation range and avoiding network overload.

[0030] S13, after receiving the NLIP message, the neighbor node checks the validity of the message and determines whether it is a duplicate message. If it is a non-duplicate message, it forwards it to other neighbor nodes and increments the hop count field in the message to indicate that the hop count of the path has increased; When checking the validity of the message, in particular, the NLIP sequence number is used to determine whether it is a duplicate message; if the message is new (i.e., the sequence number is received for the first time), the node forwards it to other neighboring nodes and appropriately increments it according to the hop count field in the message to indicate that the hop count of the path has increased.

[0031] S14, all nodes broadcast NLIP messages until the high-altitude computing platform collects complete information about all nodes in the flight ad hoc network; the high-altitude computing platform selects multi-point relay nodes based on the collected network information to reduce subsequent broadcast control overhead; S15, each node periodically sends source announcement data to the multi-point relay node according to the information collection protocol, and updates the node routing table according to the network status. If a neighbor node loses connection or does not receive the source announcement data, the neighbor node is removed from the node routing table; Each node gradually builds and updates its own routing table based on the NLIP messages received from neighboring nodes. This incremental forwarding of messages enables NLIP to traverse the entire network, and each node only needs to know the neighboring nodes directly connected to it, and does not need to understand the topology of the entire network.

[0032] S2, the high-altitude computing platform uses the uploaded node information to construct a network status graph as the input for graph neural network model training, trains the graph neural network model, updates the model parameters, and is used to predict the network status graph at the next moment; The graph neural network model includes multiple spatiotemporal convolution blocks, each of which (ST-Conv block) forms a "sandwich" structure, including two gated sequence convolution layers and a spatial graph convolution layer in the middle. The spatial graph convolution layer in the middle is a bridge connecting the two gated sequence convolution layers, which can achieve fast propagation from graph convolution to spatial state through temporal convolution. The "sandwich" structure also helps the network to fully apply the bottleneck strategy, downscaling and upscaling the channel C through the graph convolution layer to achieve scale compression and feature compression. In addition, layer normalization is used in each ST-Conv block to prevent overfitting. The input and output of the ST-Conv block are both three-dimensional tensors.

[0033] The graph neural network model consists of two spatiotemporal convolutional blocks (ST-Conv blocks) and a fully connected output layer. Each ST-Conv block contains two temporal gated convolutional layers and a spatial graph convolutional layer in the middle. Residual connections and bottleneck strategies are applied inside each block. Input The ST-Conv blocks are used to process the features uniformly to explore spatial and temporal dependencies coherently. The comprehensive features are integrated through an output layer to generate the final prediction results. .

[0034] It takes multiple network status graphs at time instants (tT, t-T+1, . . . , t-1, t) as input. In this embodiment, sampling is performed at intervals of 1s, and the output is the predicted network status graph at time (t+1). Therefore, the graph neural network model can capture the characteristics of high-dimensional data; The specific steps of S2 are: S21, receiving information of nodes and links through the high-altitude computing platform and constructing a network status diagram; S22, the high-altitude computing platform processes the constructed network status graph, uses graph neural network for spatiotemporal modeling, models the spatiotemporal correlation in FANET through multiple spatiotemporal convolution blocks, extracts evolutionary patterns from a series of network status graphs, and then predicts the network status graph at the next moment; The specific construction process of the training process of the graph neural network model in S22 includes: S221, pre-processing node features (such as network traffic, speed, etc.) through MinMaxScaler normalization, and constructing an adjacency matrix according to the FANET network structure; S222, initialize the weights and parameters of the graph neural network model, use the preprocessed node features and adjacency matrix to train the graph neural network model, define the loss function as the mean square error during the training process, and add L1 penalty to control the complexity of the graph neural network model and reduce overfitting. At the same time, the adaptive learning rate is determined by the Adam optimizer, and the training is iterated until the specified number of iterations is reached or the specified threshold error is met; S223, after the training is completed, more accurate graph neural network model parameters are obtained to predict the network status graph at the next moment.

[0035] The output of the graph neural network model is usually a prediction result. In the network prediction task, the output of the graph neural network model is usually the traffic status (such as flow, speed, etc.) at a certain time step or multiple time steps in the future, expressed as ,in T ′ is the future time step, C ′ is the prediction feature.

[0036] S3, when a node makes a routing request, the high-altitude computing platform sends a command to collect all node information and form a network status graph; the network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment, and then the shortest path algorithm is used to select the best route and send it to the UAV that made the routing request; The specific steps of selecting the best route in S3 and sending it to the drone with the route request are: S31, business connection node, request transmission, the node reports the business type and routing request to the high-altitude computing platform; S32, the high-altitude computing platform sends a command through the relay node to let all nodes upload the current node parameter information, including the node's remaining energy, cache size, three-dimensional position information, and flight speed to construct the FANET node feature matrix, and then weights the connection matrix of the FANET network according to the collected link maintenance time in the FANET network, expressed as an adjacency matrix, and obtains a network status graph; then the network status graph is input into the trained graph neural network model for prediction to obtain the network status graph at the next moment; The steps of obtaining the network status graph at the next moment include: S321, the high-altitude computing platform receives the NLIP information packet of the node through the relay node and obtains the node parameter information of the FANET network; S322, record the three-dimensional position information of the drone for , remember the speed of the drone for , record the remaining energy of the drone for , the size of the drone's cache for ; form a feature vector, written as ; The connection matrix of FANET is weighted according to the link duration (LET) of each link and expressed as an adjacency matrix ; S323, the input data of the graph neural network model is usually a spatiotemporal graph, which provides the dynamic information of each node at different time steps, represented as node features ,in N is the number of FANET nodes; T is the time step, which can be customized; C is the number of features; S324, node features and the adjacency matrix Input the trained graph neural network model for prediction and obtain the network status graph at the next moment.

[0037] S33, the high-altitude computing platform obtains the predicted network status diagram at the next moment, normalizes each metric factor of the node and multiplies it by the influence factor of each metric to obtain the node cost, takes the cost of the node with the larger cost value among the nodes on both sides of the link as the link cost, obtains the network cost diagram, and uses the shortest algorithm to solve the minimum cost, that is, the best route; By building a spatiotemporal graph through the high-altitude computing platform and training the graph neural network model, the best path can be dynamically selected to avoid network congestion and achieve traffic load balancing by jointly predicting the network status. At the same time, the prediction results help identify peak traffic periods in advance, adjust routing strategies or bandwidth allocation, and prevent network congestion and delays. Based on joint prediction and decision optimization, the intelligence and reliability of route selection can be improved, thereby improving the stability of the overall network and user experience: The steps to solve the optimal route are: S331, the measurement factors are respectively the remaining energy, LET, and the available size of the buffer area, and they are normalized and expressed as: in, Indicates drone At the moment The normalized energy cost value is Indicates drone At the moment The normalized link duration cost value of Indicates drone At the moment The normalized cache cost value of , , They represent the initial energy of the UAV, the maintenance time of the UAV system, and the buffer size. Indicates drone At the moment The remaining energy value, Indicates drone At the moment The remaining average link maintenance time is Indicates drone At the moment The remaining buffer size.

[0038] S332, according to the business type, normalize the metric factors of the node and multiply them by the influence factors of each metric to obtain the node cost. Take the cost of the node with the larger cost value on both sides of the link as the link cost, obtain the network cost graph, and use the shortest algorithm to solve the minimum cost, that is, the optimal route.

[0039] The calculation formula for node fees is: In the formula, is the node fee, , , They are the influencing factors of node remaining energy, LET, and available size of cache area.

[0040] The predicted FANET network status graph is represented as a network cost graph as follows: in, is a directed graph of FANET, is a collection of all drone nodes. Represents a collection of links. It is a slave node arrive Links, For the link fee in this link, Represents the set of link costs. In the equation above, the cost has a constant of 1, which means that the cost increases every time the path crosses a new hop. This helps in finding a short path (i.e. with fewer hops). is a matrix storing the predicted drone node cost values. It's a drone At the moment The predicted node cost value.

[0041] The objective function of the path search process can be expressed as: in, is the best path, l is the candidate path, L is the set of candidate paths, M For the system drone collection, i Candidate path l One end node of each link, j Candidate path l The other end node of each link; according to the business type, change , , The value of can be used to solve the above formula using the shortest algorithm to obtain the optimal route.

[0042] S34, the high-altitude computing platform sends the best routing table to the routing request node and updates the node routing table; When the node is in the routing maintenance phase, each node broadcasts NLIP information packets through multipoint relays every 1 second, which can regularly update the node neighbor routing table and ensure the stability of the network.

[0043] Based on similar inventive concepts, an embodiment of the present invention also provides a computer storage medium storing a readable program, which can execute the above-mentioned flying self-organizing network routing method based on graph neural network and mobile perception when the program is running.

[0044] Based on similar inventive concepts, an embodiment of the present invention provides an electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned flying self-organizing network routing method based on graph neural network and mobile perception.

[0045] Based on similar inventive concepts, an embodiment of the present invention also provides a computer program product, including computer instructions, which instruct a computing device to perform operations corresponding to the above-mentioned flying self-organizing network routing method based on graph neural network and mobile perception.

[0046] Example 3 In this embodiment, a specific embodiment is proposed to describe the solution of the present invention in detail. The specific embodiment is simulated using Python software; the specific parameters are set as shown in Table 2; Table 2 Simulation parameters Figure 4 The performance comparison results of different routing protocols in terms of packet delivery rate (PDR) are shown. Figure 4 As can be seen from the figure, the flow success rates of DSDV and GREEDY routing protocols are less than 50% in most cases, which is mainly due to their inability to effectively cope with the dynamic changes in network topology caused by the high-speed movement of nodes in 3D space. In contrast, DSDV performs relatively well in PDR, successfully delivering 40% of the packets on average. However, OPAR and JP protocols perform significantly better than other algorithms, both achieving an average packet transmission success rate of 70%, showing a significant advantage. This result shows that in a dynamically changing 3D network environment, the design of routing protocols needs to fully consider the real-time changes in network topology.

[0047] When analyzing the PDR performance, the simulation experiment set the average PAR (Packet Arrival Rate) to 10KB / s and the average velocity (AV) to 20 m / s, and observed the changes in protocol performance by changing the number of drones, such as Figure 4As shown in the figure. The PDR of the four protocols shows an obvious three-stage change feature. This change pattern can be explained as follows: in the initial stage, increasing the number of drones will improve network connectivity, increase the number of available paths, and thus improve the packet delivery rate; but as the number of drones continues to increase, the network load gradually increases, resulting in a decrease in the packet arrival rate; and finally reaching a balanced state. This three-stage change reflects the complex balance between network density, connectivity, and load, with different factors playing a leading role in different stages.

[0048] The reason why the JP protocol can show the best PDR performance is mainly due to its advanced routing decision mechanism. By predicting and evaluating the network status, the protocol can effectively avoid selecting relay links with high traffic, high mobility and low persistence, while comprehensively considering the current network indicators and possible future changes. This forward-looking routing strategy makes JP superior to other routing protocols in all test scenarios. In contrast, DSDV and GREEDY performed the worst due to the lack of sufficient consideration of the network environment. This result verifies the importance of routing protocols having predictive and adaptive capabilities in dynamic network environments.

[0049] from Figure 5 It can be seen that under low PAR conditions, the performance difference between JP and OPAR protocols is not significant. However, as PAR increases, the packet arrival rate of all protocols shows a downward trend. It is worth noting that although the packet success rate of JP and OPAR both decreased, the decline rate of JP was significantly slower. This advantage mainly comes from the fact that the JP protocol considers the remaining capacity of the drone cache when making routing decisions and predicts the congestion of the next hop, thereby effectively avoiding the occurrence of data packet congestion. This design enables JP to maintain good performance under high load conditions.

[0050] Figure 6 The effect of drone speed on PDR is shown. As the drone speed increases, the PDR of all protocols shows a downward trend. However, JP and OPAR protocols not only maintain a high PDR in high-speed mobile scenarios because they take link expiration time into account, but also their performance degrades more slowly. This result shows that in a high-speed mobile network environment, a routing strategy that takes link stability into account can significantly improve protocol performance.

[0051] Example 4 Based on the flying ad hoc network routing method based on graph neural network and mobile perception proposed in Example 1, in this embodiment, a flying ad hoc network routing system based on graph neural network and mobile perception is proposed, which specifically includes: Route discovery module: The drone periodically uses the information collection protocol to upload its own node information to the high-altitude computing platform; Model training module: The high-altitude computing platform uses the uploaded node information to construct a network status graph as the input for graph neural network model training, trains the graph neural network model, updates the model parameters, and is used to predict the network status graph at the next moment; And, the route optimization module: when a node makes a route request, the high-altitude computing platform sends a command to collect all node information and form a network status graph; the network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment, and then the shortest path algorithm is used to select the best route and send it to the UAV that made the route request; The method of the present invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CDROM, RAM, floppy disk, hard disk or magneto-optical disk), or as a computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and will be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a special-purpose processor or programmable or special-purpose hardware (such as an ASIC or FPGA). It is understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown here, the execution of the code converts the general-purpose computer into a special-purpose computer for executing the method shown here. The basic principles, main features and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above-mentioned embodiments. The above-mentioned embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may be subject to various changes and improvements, and these changes and improvements shall all fall within the scope of the present invention claimed for protection.

Claims

1. A flying ad hoc network routing method based on graph neural network and mobile perception, characterized in that: The following steps are involved: The drone periodically uses the information collection protocol to upload its own node information to the high-altitude computing platform; The high-altitude computing platform uses the uploaded node information to construct a network status graph as the input for graph neural network model training, trains the graph neural network model, updates the model parameters, and is used to predict the network status graph at the next moment; When a node makes a routing request, the high-altitude computing platform sends a command to collect all node information to form a network status graph; the network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment, and then the shortest path algorithm is used to select the best route and send it to the drone that made the routing request.

2. The flying ad hoc network routing method based on graph neural network and mobile perception according to claim 1 is characterized in that: The flying ad hoc network includes: users, drones and high-altitude computing platforms; Each user is associated with the drone with the strongest received signal strength. When a user has a communication need, the data packet will be uploaded to the associated drone. Finding the route of the user data packet is to find the route of the associated drone, and the drone is regarded as a node; After receiving the data packet, the drone sends a routing request to the high-altitude computing platform. The high-altitude computing platform sends an information collection command to the drone. After collecting the drone information, it preprocesses it to form a network status graph. The network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment. The shortest path algorithm is then used to select the best route and send it to the drone that requested the route. After receiving the best route, the drone forwards the data packet sent by the user.

3. The flying ad hoc network routing method based on graph neural network and mobile perception according to claim 2 is characterized in that: The steps of uploading the drone's own node information to the high-altitude computing platform are as follows: All nodes generate and send NLIP messages periodically; After receiving the NLIP message, the neighbor node checks the validity of the message to determine whether it is a duplicate message. If it is a non-duplicate message, it forwards it to other neighbor nodes and increments the hop count field in the message to indicate that the hop count of the path has increased. All nodes broadcast NLIP messages until the high-altitude computing platform collects complete information about all nodes in the flight ad hoc network. The high-altitude computing platform selects multi-point relay nodes based on the collected network information. Each node periodically sends source announcement data to the multi-point relay node according to the information collection protocol, and updates the node routing table according to the network status. If a neighbor node loses connection or does not receive the source announcement data, the neighbor node is removed from the node routing table.

4. The flying ad hoc network routing method based on graph neural network and mobile perception according to claim 2 is characterized in that: The graph neural network model includes multiple spatiotemporal convolution blocks, each of which includes two gated sequential convolution layers and a spatial graph convolution layer located between the two gated sequential convolution layers.

5. The flying ad hoc network routing method based on graph neural network and mobile perception according to claim 4 is characterized in that: The training process of the graph neural network model is as follows: The node features are preprocessed by MinMaxScaler normalization, and the adjacency matrix is ​​constructed according to the FANET topology; Initialize the weights and parameters of the graph neural network model, use the preprocessed node features and adjacency matrix to train the graph neural network model, define the loss function as the mean square error during the training process, and add L1 penalty to control the complexity of the graph neural network model. At the same time, the adaptive learning rate is determined using the Adam optimizer.

6. The flying ad hoc network routing method based on graph neural network and mobile perception according to claim 2 is characterized in that: The specific steps of selecting the best route and sending it to the drone requesting the route are: The business connects to the node, requests transmission, and the node reports the business type and routing request to the high-altitude computing platform; The high-altitude computing platform sends commands through relay nodes to let all nodes upload current node parameter information, construct the node feature matrix of FANET, and then weight the connection matrix of FANET network according to the collected link maintenance time in FANET network, expressed as an adjacency matrix, and obtain the network status graph; then the network status graph is input into the trained graph neural network model for prediction, and the network status graph at the next moment is obtained; The high-altitude computing platform obtains the predicted network status diagram at the next moment, normalizes the metric factors of the node, multiplies them by the influence factors of each metric, and weights them to obtain the node cost. The cost of the node with the larger cost value on both sides of the link is taken as the link cost, and the network cost diagram is obtained. The shortest algorithm is used to solve the minimum cost, that is, the optimal route; The high-altitude computing platform sends the best routing table to the routing request node and updates the node routing table.

7. A flying ad hoc network routing system based on graph neural network and mobile perception, characterized in that: include: Route discovery module: The drone periodically uses the information collection protocol to upload its own node information to the high-altitude computing platform; Model training module: The high-altitude computing platform uses the uploaded node information to construct a network status graph as the input for graph neural network model training, trains the graph neural network model, updates the model parameters, and is used to predict the network status graph at the next moment; And, the routing optimization module: when a node makes a routing request, the high-altitude computing platform sends a command to collect all node information to form a network status graph; the network status graph is input into the trained graph neural network model to obtain the network status graph at the next moment, and then the shortest path algorithm is used to select the best route and send it to the drone that made the routing request.

8. A computer storage medium storing a readable program, characterized in that: When the program is running, it can execute the flying self-organizing network routing method based on graph neural network and mobile perception as described in any one of claims 1-6.

9. An electronic device, characterized in that: include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the flying self-organizing network routing method based on graph neural network and mobile perception as described in any one of claims 1-6.

10. A computer program product comprising computer instructions, characterized in that The computer instructions instruct the computing device to perform operations corresponding to the flying self-organizing network routing method based on graph neural network and mobile perception as described in any one of claims 1-6.

Images