Satellite network traffic perception method based on spatiotemporal graph neural network

Abstract

The present invention provides a satellite network traffic perception method based on a spatiotemporal graph neural network, comprising the following steps: calculating the telemetry delay L from node i to node j along the path according to the satellite network topology and node link traffic; i,j , telemetry overhead C i,j and path reliability R i,j ; Based on the spectral clustering method, the satellite network is divided into K Sub-network G S (V S ,E S )={C1,C2,...,C K}; Find the telemetry path based on Hamiltonian circuit with the minimum weighted sum of telemetry delay and telemetry overhead in subnetwork k, and set the flow telemetry threshold T tr The method uses a spatiotemporal graph neural network to capture the spatiotemporal dependencies between network topology and service traffic, and predicts traffic data for the next moment. Based on the collected and predicted traffic data, a satellite network traffic status view is analyzed to achieve satellite network traffic perception. This method reduces telemetry overhead in satellite networks and improves traffic prediction accuracy, resolving the issues of high perception overhead and low accuracy caused by spatiotemporal coupling and random bursts of satellite network service traffic.

Description

Technical Field

[0001] The present invention relates to the field of satellite communications and network technology, and in particular to a satellite network traffic perception method based on a spatiotemporal graph neural network. Background Art

[0002] Satellite networks are complex systems that use satellite platforms as carriers, combined with ground-based gateways and data centers, to collect, process, and transmit information. Featuring wide-area coverage and ubiquitous access, they play an increasingly important role in ocean navigation, emergency communications, and other fields. They are a crucial component of future space communications infrastructure and have become a frontier of current research. However, with the rapid development of the internet, the volume of data in satellite networks has increased exponentially. Centralized control within the system control segment lacks flexibility for service management and cannot handle network failures in real time. Software-defined satellite networks (SDSNs) decouple the control and data planes. Satellite switching nodes only need to perform simple data forwarding and hardware configuration. The programmable control plane enables intelligent network management and control within a logically centralized controller without affecting data flow, providing a technical foundation for flexible satellite network management and control.

[0003] Network traffic awareness is fundamental to research on network performance monitoring, resource management, and fault diagnosis. However, methods such as active and passive measurement are no longer able to meet the low-latency and high-reliability performance requirements of satellite networks. In-band network telemetry (INT) has become a research hotspot in both industry and academia. It combines packet forwarding with network measurement, collecting state information of network nodes and inserting it into telemetry packets. It offers real-time, flexible, and fine-grained capabilities. Therefore, it is crucial to research INT telemetry methods in satellite networks to achieve real-time collection of satellite network traffic with minimal overhead and latency.

[0004] Satellite network traffic and network topology are tightly coupled, exhibiting strong spatiotemporal correlation and heterogeneity, making traffic forecasting difficult. Network traffic forecasting methods based on spatiotemporal graph neural networks (STGNNs) have been applied to ground traffic forecasting, weather forecasting, and other fields. Compared with traditional statistical and deep learning methods, STGNNs can more efficiently and accurately extract complex spatiotemporal dependencies, providing a data foundation for intelligent network management and control.

[0005] Telemetry latency, overhead, and reliability are important factors influencing telemetry path planning. Network latency affects the efficiency and real-time nature of telemetry data transmission. Satellite network resources are limited, and the bandwidth and other resources occupied by telemetry data increase linearly with the amount of data collected. The dynamic nature of satellite networks makes inter-satellite links susceptible to interruption, and the reliability of telemetry paths affects the time required for error-free transmission of telemetry data. Unlike terrestrial network traffic sensing, satellite networks are susceptible to the impact of the space environment and are highly dynamic. Furthermore, limited onboard resources make existing research inapplicable to satellite networks. Implementing traffic sensing in satellite networks suffers from excessive telemetry overhead and low accuracy. Summary of the Invention

[0006] In order to solve the technical problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a satellite network traffic perception method based on a spatiotemporal graph neural network, which reduces the telemetry overhead in the satellite network and improves the accuracy of traffic prediction, and solves the problems of large perception overhead and low accuracy caused by spatiotemporal coupling and random bursts of satellite network business traffic.

[0007] To achieve the above-mentioned object of the invention, the present invention provides a satellite network traffic perception method based on spatiotemporal graph neural network, comprising the following steps:

[0008] Step S1: Calculate the path according to the satellite network topology and node link traffic. Telemetry delay L from node i to node j i,j , telemetry overhead C i,j and path reliability R i,j ;

[0009] Step S2: Divide the satellite network into K sub-networks G based on the spectral clustering method. S (V S ,E S )={C1,C2,...,C K};

[0010] Step S3: Find a telemetry path based on a Hamiltonian circuit with the minimum weighted sum of telemetry delay and telemetry overhead in subnetwork k, and set the flow telemetry threshold T tr ;

[0011] Step S4: Use the spatiotemporal graph neural network to obtain the spatiotemporal dependency of network topology and service traffic, and predict the traffic data at the next moment;

[0012] Step S5: Analyze and obtain a satellite network traffic status view based on the collected traffic data and the predicted traffic data to achieve satellite network traffic perception.

[0013] According to a technical solution of the present invention, in step S1, the propagation delay between adjacent node a and node b is Among them, dis a,b is the distance between node a and node b, and c is the speed of light;

[0014] D a,b Indicates the amount of data sent between adjacent node a and node b, B a,b represents the data transmission rate, then the transmission delay from adjacent node a to node b is

[0015] The queue processing delay of node a is L a represents the queue length of node a, λ is the arrival rate, and the network is defined along the path The delay L from node i to node j i,j The expression is:

[0016]

[0017] in, Indicates that (a,b) is a path The number of links on the Link , Indicates that node a is a path The number of nodes on the Sat .

[0018] According to a technical solution of the present invention, in step S1, the network along the path Telemetry cost C from node i to node j i,j Expressed as:

[0019]

[0020] in, is the amount of data collected by node a, C header is the telemetry header overhead, C data For a single telemetry data overhead.

[0021] According to a technical solution of the present invention, in step S1, the path reliability is composed of node reliability and link reliability. The reliability of the path from node i to node j is R i,j Expressed as:

[0022]

[0023] Among them, the failure probability of node a is Failure probability of link (a, b)

[0024] According to a technical solution of the present invention, step S3 includes the following steps:

[0025] Step S31, determining the path repeatability and coverage;

[0026] Step S32: constructing an objective function;

[0027] Step S33: Set the flow telemetry threshold T tr ,Telemetry path planning based on Hamiltonian circuits.

[0028] According to a technical solution of the present invention, in step S31, sub-network C k Total nodes, telemetry paths The number of nodes included is n INT , then the telemetry path duplication of the sub-network is ovl k Expressed as:

[0029]

[0030] In subnetwork C k The path from node i to node j in The coverage is expressed as:

[0031]

[0032] According to a technical solution of the present invention, in step S32, the objective function χ i,j is the comprehensive optimization function of telemetry delay and telemetry overhead, expressed as:

[0033]

[0034] in, represents the normalized telemetry overhead, represents the normalized telemetry delay, α and β are the weight coefficients of the normalized overhead and delay respectively, and α + β = 1; constraint C1 ensures the reliability of the telemetry path R INT Greater than or equal to the threshold reliability R th ; Constraint C2 ensures the coverage of each subnetwork η k is 1 and the coverage of the entire network; the normalized path delay is expressed as:

[0035]

[0036] Among them, L MAX =max{L i,j} is the maximum telemetry delay of the network;

[0037] The normalized path cost is expressed as:

[0038]

[0039] Among them, C MAX =max{C i,j} represents the maximum telemetry overhead of the network.

[0040] According to a technical solution of the present invention, in step S4, it specifically includes:

[0041] The spatiotemporal convolutional layer consists of two GCNs and two TCNs connected alternately in series, and the spatiotemporal graph learning module consists of two spatiotemporal convolutional layers and one output layer;

[0042] Construct a spatiotemporal graph based on spatial similarity according to network topology characteristics and service traffic characteristics, and analyze the spatiotemporal dependency of network topology characteristics and service traffic characteristics in satellite networks;

[0043] The historical business traffic data is input into the spatiotemporal graph learning module, the spatiotemporal dependency is mapped to the next time period, and the future satellite network traffic changes are predicted.

[0044] According to a technical solution of the present invention, the flow telemetry threshold T is set tr , only when the traffic volume T of node i i Compared with the benchmark traffic T base When the difference between the two is greater than the threshold traffic flow value, the traffic flow data of node i is collected and the benchmark traffic flow T is updated. base =T i ;

[0045] Otherwise, the node traffic is considered to be consistent with the previous moment.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention proposes a satellite network traffic perception method based on a spatiotemporal graph neural network, which reduces the telemetry overhead in the satellite network and improves the accuracy of traffic prediction, solving the problems of high perception overhead and low accuracy caused by spatiotemporal coupling and random bursts of satellite network service traffic.

[0048] The present invention addresses the difficulty in perceiving satellite network service traffic. By planning satellite network telemetry paths with set thresholds to collect traffic data and constructing a spatiotemporal graph neural network to predict service traffic changes, the invention achieves low-overhead and high-accuracy satellite network traffic perception, providing a data foundation for real-time and flexible satellite network management and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort.

[0050] Figure 1 1 is a flow chart of a satellite network traffic perception method based on a spatiotemporal graph neural network in one embodiment of the present invention;

[0051] Figure 2 A schematic diagram of a satellite network scenario simulation in one embodiment of the present invention;

[0052] Figure 3 This is an in-band network telemetry data collection process in one embodiment of the present invention;

[0053] Figure 4 is threshold in-band network telemetry in one embodiment of the present invention;

[0054] Figure 5 A spatiotemporal graph neural network traffic prediction process in one embodiment of the present invention;

[0055] Figure 6 Comparison of satellite network telemetry overhead and latency in one embodiment of the present invention;

[0056] Figure 7 This is an actual satellite network service traffic view in one embodiment of the present invention;

[0057] Figure 8 This is a diagram of predicting satellite network service traffic in one embodiment of the present invention. DETAILED DESCRIPTION

[0058] Reference will now be made in detail to various embodiments of the present invention, examples of which are illustrated in the accompanying drawings and described below. Although the present invention will be described in conjunction with exemplary embodiments of the present invention, it should be understood that this description is not intended to limit the present invention to those exemplary embodiments. On the other hand, the present invention is intended to cover not only the exemplary embodiments of the present invention, but also various alternatives, modifications, equivalents and other embodiments, which may be included within the spirit and scope of the present invention as defined by the appended claims.

[0059] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. The specific structures and functions described in the exemplary embodiments of the present invention are for illustrative purposes only. The embodiments according to the concepts of the present invention may be implemented in various forms, and it should be understood that they should not be interpreted as being limited to the exemplary embodiments described in the exemplary embodiments, but include all modifications, equivalents or alternatives included in the spirit and scope of the present invention.

[0060] Throughout the specification, the technical terms used herein are for the purpose of describing various exemplary embodiments only and are not intended to be limiting. It will be further understood that the terms "comprises," "comprising," "having," etc., when used in exemplary embodiments, specifically refer to the presence of the stated parts, steps, operations, or elements, but do not preclude the presence or addition of one or more other parts, steps, operations, or elements.

[0061] The satellite network traffic perception method based on spatiotemporal graph neural network of the present invention involves a satellite network scenario based on software defined network (SDN) G(V,E)=G S ∪G T , including the spatial segment network G S (V S ,E S ) and ground segment network G T (V T ,E T ), the two are connected through a satellite-to-ground link, such as Figure 2 The ground segment includes satellite gateways, switch nodes, and user nodes. Switch nodes are connected via ground links, and users are connected to switch nodes via user ground links. Satellite gateway deploys the main controller N docon Formulate satellite network forwarding strategy and upload it to the slave controller N in the space segment sercon At the same time, the user node transmits data to the satellite network through the satellite gateway.

[0062] Satellite Network G S (V S ,E S ) is composed of a mixture of polar orbit and inclined orbit low earth orbit (LEO) satellites. Adjacent satellites are connected by inter-satellite links, which are divided into K sub-networks G S (V S ,E S )={C1,C2,...,C K}, and deploy the slave controller in sub-network k Simultaneously, telemetry paths are planned within the subnetwork, collecting data from the nodes traversed and reporting it to the controller. Time-varying traffic and network topology features are extracted to construct a spatiotemporal graph, predicting traffic trends and analyzing them to generate a localized traffic status view. Telemetry paths are planned within subnetwork k, and their data plane information collection eliminates the need for interaction with the control plane, enabling real-time, fine-grained collection and reporting of device port, data volume, and queue information.

[0063] The telemetry process used in this patent is as follows Figure 3 As shown, Switch 1 inserts an INT header into the data packet, which contains instructions for metadata collection. After acquiring telemetry information, it encapsulates it into an INT packet. The telemetry packet, after passing through Switches 2 and 3, reaches Switch 4, which then sends the metadata to the controller via an INTreport message. The controller extracts telemetry data and network topology features to construct a spatiotemporal graph, predicting service traffic trends and providing a network status view.

[0064] like Figures 1 to 3 As shown, the present invention provides a satellite network traffic perception method based on a spatiotemporal graph neural network, comprising the following steps:

[0065] Step S1: Calculate the path according to the satellite network topology and node link traffic. Telemetry delay L from node i to node j i,j , telemetry overhead C i,j and path reliability R i,j ;

[0066] Step S2: Divide the satellite network into K sub-networks G based on the spectral clustering method. S (V S ,E S )={C1,C2,...,C K};

[0067] Step S3: Find a telemetry path based on a Hamiltonian circuit with the minimum weighted sum of telemetry delay and telemetry overhead in subnetwork k, and set the flow telemetry threshold T tr ;

[0068] Step S4: Use the spatiotemporal graph neural network to obtain the spatiotemporal dependency of network topology and service traffic, and predict the traffic data at the next moment;

[0069] Step S5: Analyze and obtain a satellite network traffic status view based on the collected traffic data and the predicted traffic data to achieve satellite network traffic perception.

[0070] STGNN consists of a data processing module, a spatiotemporal graph learning module, and a task perception prediction module, such as Figure 5As shown in the figure, telemetry data is processed and input into the spatiotemporal graph learning module to obtain the spatiotemporal relationship between network topology and service traffic, and to predict future network traffic. First, satellite network traffic information collected by the telemetry path is input into the data processing module. A spatiotemporal graph based on spatial similarity is constructed based on the satellite network topology and service traffic characteristics. The spatiotemporal graph learning module consists of two spatiotemporal convolutional layers and one output layer. It analyzes the dependencies between service traffic and network topology in complex satellite networks based on the spatiotemporal graph. The spatiotemporal convolutional layer uses a Generative Network (GCN) to capture the spatial correlations of satellite network topology, while a Transitional Network (TCN) captures the temporal correlations of service traffic. Spatial and temporal learning are coupled through a factored neural architecture to obtain the spatial characteristics of time-varying satellite network traffic. Finally, the perception prediction module maps the spatiotemporal dependencies of satellite network service traffic obtained through spatiotemporal graph learning into service traffic forecasts for the next time period, analyzing and obtaining a satellite network traffic status view.

[0071] According to a technical solution of the present invention, in step S1, the propagation delay between adjacent node a and node b is Among them, dis a,b is the distance between node a and node b, and c is the speed of light;

[0072] D a,b Indicates the amount of data sent between adjacent node a and node b, B a,b represents the data transmission rate, then the transmission delay from adjacent node a to node b is

[0073] The queue processing delay of node a is L a represents the queue length of node a, λ is the arrival rate, and the network is defined along the path The delay L from node i to node j i,j The expression is:

[0074]

[0075] in, Indicates that (a,b) is a path The number of links on the Link , Indicates that node a is a path The number of nodes on the Sat .

[0076] Regarding telemetry latency, network telemetry needs to complete data collection and transmission with low latency and report it to the controller node to meet the satellite network's real-time perception of traffic data. Therefore, this paper considers the impact of propagation delay, transmission delay, and queuing processing delay on the real-time performance of telemetry perception.

[0077] According to a technical solution of the present invention, in step S1, the network along the path Telemetry cost C from node i to node j i,j Expressed as:

[0078]

[0079] in, is the amount of data collected by node a, C header is the telemetry header overhead, C data For a single telemetry data overhead.

[0080] As for telemetry overhead, the more nodes and links an INT packet passes through, the greater the overhead. The bandwidth and other resources it occupies increase linearly with the information collected by the packet. The INT header in INT occupies 8 bytes, and each telemetry data occupies 4 bytes. For a 2000-byte packet, bandwidth usage increases dramatically with the number of nodes on the telemetry path. If two data points are collected from each node over 10 hops, the bandwidth usage is 4.4%. Furthermore, bandwidth overhead increases linearly with the growth of telemetry demand. Therefore, it is necessary to optimize INT overhead to reduce bandwidth usage.

[0081] According to a technical solution of the present invention, in step S1, the path reliability is composed of node reliability and link reliability. The reliability of the path from node i to node j is R i,j Expressed as:

[0082]

[0083] Among them, the failure probability of node a is Failure probability of link (a, b)

[0084] As for path reliability, the higher the path reliability, the longer the time it takes for error-free transmission of telemetry data between network nodes. Therefore, telemetry paths should choose paths with high reliability to ensure accurate collection of telemetry data. Path reliability consists of node reliability and link reliability.

[0085] According to a technical solution of the present invention, step S3 includes the following steps:

[0086] Step S31, determining the path repeatability and coverage;

[0087] Step S32: constructing an objective function;

[0088] Step S33: Set the flow telemetry threshold T tr ,Telemetry path planning based on Hamiltonian circuits.

[0089] A spectral clustering-based approach is used to partition the satellite network into K subnetworks. This reduces the complexity of the algorithm design by breaking the complex global problem into multiple, relatively simple local problems. A backtracking-pruning approach is then used within subnetwork k to recursively search the solution space with a limited depth, reducing the number of invalid traversals. If no Hamiltonian circuit exists in a subnetwork, the backtracking-pruning search is continued by gradually increasing the number of nodes that can be traversed in the subnetwork.

[0090] Set the traffic telemetry threshold T tr ,like Figure 4 As shown, if and only if the service flow of node i is T i Compared with the benchmark traffic T base When the difference between the two is greater than the threshold traffic flow value, the traffic flow data of node i is collected, and the current benchmark traffic flow is updated to the traffic flow of the current node i, that is, T base =T i If the difference between the two is less than the threshold, no traffic data is collected, and the node traffic is considered to be consistent with the previous moment. The above telemetry path planning and threshold setting mechanism can effectively reduce telemetry data redundancy and achieve lightweight network traffic data acquisition.

[0091] According to a technical solution of the present invention, in step S31, sub-network C k Total nodes, telemetry paths The number of nodes included is n INT , then the telemetry path duplication of the sub-network Expressed as:

[0092]

[0093] In subnetwork C k The path from node i to node j in The coverage is expressed as:

[0094]

[0095] Regarding path repetition and coverage, telemetry paths may collect node information through the same node multiple times. Repeated node data collection leads to information redundancy and increased telemetry overhead. Coverage is the ratio of the number of nodes traversed by a path to the total number of nodes in the network, reflecting the satellite network's coverage capacity for the telemetry path.

[0096] According to a technical solution of the present invention, in step S32, the objective function χ i,j is a comprehensive optimization function of telemetry delay and telemetry overhead, aiming to minimize the weighted sum of the two, expressed as:

[0097]

[0098] in, represents the normalized telemetry overhead, represents the normalized telemetry delay, α and β are the weight coefficients of the normalized overhead and delay respectively, and α + β = 1; constraint C1 ensures the reliability of the telemetry path R INT Greater than or equal to the threshold reliability R th ; Constraint C2 ensures the coverage of each subnetwork η k is 1 and the coverage of the entire network; the normalized path delay is expressed as:

[0099]

[0100] Among them, L MAX =max{L i,j} is the maximum telemetry delay of the network;

[0101] The normalized path cost is expressed as:

[0102]

[0103] Among them, C MAX =max{C i,j} represents the maximum telemetry overhead of the network.

[0104] According to a technical solution of the present invention, in step S4, it specifically includes:

[0105] The spatiotemporal convolutional layer for traffic prediction consists of two GCNs and two TCNs alternately connected in series. It uses fewer parameters to achieve faster training and easier convergence, and has certain flexibility and scalability. The spatiotemporal graph learning module consists of two spatiotemporal convolutional layers and one output layer.

[0106] Construct a spatiotemporal graph based on spatial similarity according to network topology characteristics and service traffic characteristics, and analyze the spatiotemporal dependency of network topology characteristics and service traffic characteristics in satellite networks;

[0107] The historical business traffic data is input into the spatiotemporal graph learning module, the spatiotemporal dependency is mapped to the next time period, and the future satellite network traffic changes are predicted.

[0108] For a network G(V,E) with N nodes, its node set is V = {v1,v2,...,v N If nodes i and j are connected, then w ij =1, otherwise w ij =0, the adjacency matrix is: W = (w ij ); i, j = 1, ..., N. The degree of node i is The degree matrix of the graph is:

[0109]

[0110] In this application, the satellite network is based on the adjacency matrix A of spatial similarity = (a ij );i,j=1,...,N, and a ij The definition is as follows,

[0111]

[0112] Among them, σ is a hyperparameter that controls the distribution. Spatial learning uses GCN, and the expression is approximately written as

[0113]

[0114] Among them, Θ is the graph convolution kernel, θ is the learnable parameter, I N is the identity matrix, D is the degree matrix, A is the adjacency matrix, and The temporal learning of spatiotemporal graph neural network adopts TCN, whose causal convolution is expressed as

[0115]

[0116] Among them, Γ is the temporal convolution kernel, d is the scaling factor, and k is the filter f in TCN. i The size of , s is the time series input.

[0117] The spatiotemporal convolution layer of this application is composed of 2 GCNs and 2 TCNs connected in series alternately. For the spatiotemporal convolution layer l, its input x l After the graph convolution kernel Temporal convolution kernel Graph convolution kernel and temporal convolution kernels Output x l+1 , the mapping relationship can be expressed as

[0118]

[0119] Among them, ReLU{·} is the rectified linear unit function.

[0120] According to a technical solution of the present invention, the flow telemetry threshold T is set tr , only when the traffic volume T of node i i Compared with the benchmark traffic T base When the difference between the two is greater than the threshold traffic flow value, the traffic flow data of node i is collected and the benchmark traffic flow T is updated. base =T i ;

[0121] Otherwise, the node traffic is considered to be consistent with the previous moment.

[0122] The following is a simulation analysis of the satellite network traffic perception method based on spatiotemporal graph neural network involved in the present invention.

[0123] The satellite network used in the simulation analysis references the "Iridium" and "Starlink" constellation configurations, consisting of a mix of inclined and polar orbit satellites, with 293 LEO satellites. This was used to verify the feasibility and effectiveness of the STGNN-based satellite network traffic perception algorithm. The dataset used for STGNN training, validation, and testing was based on the service traffic of a backbone network collected by the WIDE Project. This traffic data was loaded into the satellite network and node traffic information was collected to form a satellite network service traffic dataset. This dataset was then divided into training, test, and validation sets at a ratio of 0.7:0.15:0.15.

[0124] The comparison method of telemetry delay and overhead is the telemetry path planning method based on the Euler circuit, which starts from a node, passes through all edges in the network once, and then returns to the starting point. The telemetry path based on the Hamiltonian circuit is a path that starts from a node, passes through all nodes in the network only once, and finally returns to the starting point. The Euler method will repeatedly collect data from the same node multiple times, resulting in a sharp increase in overhead, while the Hamiltonian method only aims to pass through each node, which can reduce the redundancy of information collection to a certain extent and reduce telemetry overhead. Figure 6 As shown in the figure, the blue dashed line represents the telemetry overhead of the Euler method, and the orange dashed line represents the telemetry overhead of the Hamiltonian method. The proposed method reduces the overhead by over 20% compared to the Euler method. The blue bar graph represents the telemetry latency of the Euler method, and the orange bar graph represents the telemetry latency of the Hamiltonian method. The proposed method reduces the latency by over 40%. In other words, the proposed Hamiltonian-based telemetry path achieves satellite network node traffic data acquisition with lower telemetry latency and overhead.

[0125] The LSTM-based method was used for network traffic prediction. The accuracy of the two traffic prediction results is shown in Table 1. The LSTM's root mean square error (RMSE) and mean absolute error (MAE) were 1.9654 and 1.3438, respectively, while the STGNN's RMSE and MAE were 0.7893 and 0.2846. The STGNN-based satellite network traffic prediction method used in this paper outperformed the LSTM-based method in both metrics, with improvements of 59.8% and 78.8%, respectively, enabling more accurate prediction of traffic flow for satellite network nodes in subsequent time slices.

[0126] Table 1 Comparison of prediction accuracy

[0127]

[0128] The global status view includes the current network traffic and network topology, and provides information about the satellite network traffic distribution. Each square represents a satellite node. Yellower blocks indicate higher traffic volume, while smaller blocks indicate lower load. Figure 7 This is the actual satellite network traffic view. Figure 8 For the predicted traffic view, the predicted satellite traffic distribution is consistent with the actual one. For satellite nodes without business traffic, the proposed algorithm still performs micro-traffic prediction, which can effectively deal with the impact of micro-burst traffic on the real-time perception of the network and improve the adaptability of the satellite network.

[0129] The present invention proposes a satellite network traffic perception method based on a spatiotemporal graph neural network. This method proposes a Hamiltonian circuit-based satellite network telemetry path and sets up traffic control node data collection. Based on the satellite network topology and traffic characteristics, a spatiotemporal graph model based on spatial similarity is constructed. A GCN and TCN are coupled in series to construct an STGNN to obtain spatiotemporal dependencies and predict satellite network traffic for the next time period. Based on current and future traffic data, a satellite network traffic status view is analyzed to achieve satellite network traffic perception. This method addresses the high overhead and low accuracy of satellite network traffic perception by planning satellite network telemetry paths to reduce telemetry and latency. Furthermore, a STGNN-based method is used to achieve high-accuracy satellite network traffic prediction based on telemetry data, thus achieving satellite network traffic perception.

[0130] The foregoing description of specific exemplary embodiments of the present invention has been presented for purposes of illustration and description. It is not intended to exclude or limit the invention to the precise form disclosed, and it is apparent that many modifications and variations are possible in light of the above teachings. The exemplary embodiments have been selected and described to explain certain principles of the present invention and their practical application so as to enable others skilled in the art to make or utilize the various exemplary embodiments of the present invention, and various alternatives and modifications thereof. It is intended that the scope of the present invention be defined by the claims appended hereto and their equivalents.

[0131] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is merely a specific embodiment of the present invention and is not intended to limit the present invention. It can also be a reasonable combination of the features described in the above embodiments. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A satellite network traffic perception method based on spatiotemporal graph neural network, characterized in that: The following steps are involved: Step S1: Calculate the path according to the satellite network topology and node link traffic. Telemetry delay L from node i to node j i,j , telemetry overhead C i,j and path reliability R i,j ; Step S2: Divide the satellite network into K sub-networks G based on the spectral clustering method. S (V S ,E S )={C1,C2,...,C K }; Step S3: Find a telemetry path based on a Hamiltonian circuit with the minimum weighted sum of telemetry delay and telemetry overhead in subnetwork k, and set the flow telemetry threshold T tr ; The setting flow telemetry threshold T tr , only when the traffic volume T of node i i Compared with the benchmark traffic T base The difference is greater than the flow telemetry threshold T tr When the service flow data of node i is collected, the benchmark service flow T is updated. base =T i Otherwise, the node traffic is considered to be consistent with the previous moment; Step S4: Use the spatiotemporal graph neural network to obtain the spatiotemporal dependency of network topology and service traffic, and predict the traffic data at the next moment; Step S5: Analyze and obtain a satellite network traffic status view based on the collected traffic data and the predicted traffic data to achieve satellite network traffic perception.

2. The satellite network traffic perception method based on spatiotemporal graph neural network according to claim 1 is characterized in that: In step S1, the propagation delay between adjacent node a and node b is Among them, dis a,b is the distance between node a and node b, and c is the speed of light; D a,b Indicates the amount of data sent between adjacent node a and node b, B a,b represents the data transmission rate, then the transmission delay from adjacent node a to node b is The queue processing delay of node a is L a represents the queue length of node a, λ is the arrival rate, and the network is defined along the path The delay L from node i to node j i,j The expression is: in, Indicates that (a,b) is a path The number of links on the Link , Indicates that node a is a path The number of nodes on the Sat .

3. The satellite network traffic perception method based on spatiotemporal graph neural network according to claim 1 is characterized in that: In step S1, the network along the path Telemetry cost C from node i to node j i,j Expressed as: in, is the amount of data collected by node a, C header is the telemetry header overhead, C data For a single telemetry data overhead.

4. The satellite network traffic perception method based on spatiotemporal graph neural network according to claim 1 is characterized in that: In step S1, path reliability consists of node reliability and link reliability. The reliability of the path from node i to node j is R i,j Expressed as: Among them, the failure probability of node a is Failure probability of link (a, b) 5. The satellite network traffic perception method based on spatiotemporal graph neural network according to claim 1 is characterized in that: The step S3 includes the following steps: Step S31, determining the path repeatability and coverage; Step S32: constructing an objective function; Step S33: Set the flow telemetry threshold T tr ,Telemetry path planning based on Hamiltonian circuits.

6. The satellite network traffic perception method based on spatiotemporal graph neural network according to claim 5 is characterized in that: In step S31, sub-network C k Total nodes, telemetry paths The number of nodes included is n INT , then the telemetry path duplication of the sub-network is ovl k Expressed as: In subnetwork C k The path from node i to node j in The coverage is expressed as:

7. The satellite network traffic perception method based on spatiotemporal graph neural network according to claim 5 is characterized in that: In step S32, the objective function χ i,j is the comprehensive optimization function of telemetry delay and telemetry overhead, expressed as: subject to:C1:R INT ≥R th C2:∑η k =K&h k =1 in, represents the normalized telemetry overhead, represents the normalized telemetry delay, a and β are the weight coefficients of the normalized overhead and delay respectively, and α + β = 1; constraint C1 ensures the reliability of the telemetry path R INT Greater than or equal to the threshold reliability R th ; Constraint C2 ensures the coverage of each subnetwork η k is 1 and the coverage of the entire network; the normalized telemetry delay is expressed as: Among them, L MAX =max{L i,j } is the maximum telemetry delay of the network; The normalized telemetry overhead is expressed as: Among them, C MAX =max{C i,j } represents the maximum telemetry overhead of the network.

8. The satellite network traffic perception method based on spatiotemporal graph neural network according to claim 1 is characterized in that: In the step S4, it specifically includes: The spatiotemporal convolutional layer consists of two GCNs and two TCNs connected alternately in series, and the spatiotemporal graph learning module consists of two spatiotemporal convolutional layers and one output layer; Construct a spatiotemporal graph based on spatial similarity according to network topology characteristics and service traffic characteristics, and analyze the spatiotemporal dependency of network topology characteristics and service traffic characteristics in satellite networks; The historical business traffic data is input into the spatiotemporal graph learning module, the spatiotemporal dependency is mapped to the next time period, and the future satellite network traffic changes are predicted.

Images