A satellite routing method based on dynamic celestial sphere region division

Abstract

The present invention relates to the field of satellite communication technology, and particularly to a satellite routing method based on dynamic celestial sphere region division. The method includes: dividing multiple nodes by using a clustering algorithm to obtain multiple virtual regions with load balance; optimizing the multiple virtual regions by using a local flooding optimization strategy to generate a preset propagation range of data packets; predicting the delay and congestion conditions of the links in the preset propagation range by using a recurrent neural network model according to historical link delay data, and dynamically adjusting the path selection and traffic allocation strategies based on the prediction results; calculating the additional delay of each path based on the delay and load conditions of different paths in the path selection and traffic allocation strategies, and calculating the traffic to be allocated for each path based on the additional delay. Thereby, the communication bottleneck problems such as network delay and increased packet loss rate caused by the dynamic changes of the network in low-earth orbit satellites and the Internet of Things are solved, and the efficient operation of the network in a dynamic environment is realized.

Description

Technical Field

[0001] The present invention relates to the field of satellite communication technology, and particularly to a satellite routing method based on dynamic celestial sphere region division. Background Art

[0002] With the rapid development of satellite communication networks and Internet of Things technologies, global network connections have become increasingly complex. Especially in the application scenarios of Low Earth Orbit Satellite Networks (LEO) and Internet of Things (IoT), when low-orbit satellites orbit the Earth, the connections and disconnections of links occur frequently; and a large number of mobile nodes such as sensors, devices, and vehicles in the IoT also continuously change their positions and states at different times and in different spaces. These network dynamic characteristics directly affect the communication quality between nodes, such as problems like delay and packet loss.

[0003] In current technologies, common routing algorithms include the global flooding routing algorithm and the static node partitioning strategy. The basic idea of the global flooding routing algorithm is to forward data packets to each node in the network through broadcasting and discard them when duplicate data packets are received. However, this method often fails to adjust the routing in a timely manner when the network load suddenly changes, resulting in an increase in network delay and packet loss rate. On the other hand, the static node partitioning strategy attempts to achieve reasonable utilization of resources by pre-assigning network resources (such as frequency bands or time slots) to each node. But this strategy lacks flexibility and cannot be adaptively adjusted according to the dynamic changes of the network, resulting in low resource utilization and a decline in network efficiency.

[0004] Therefore, how to improve communication efficiency and effectively address problems such as delay and packet loss in an environment of network dynamic changes has become a technical problem to be urgently solved. Summary of the Invention

[0005] The present invention provides a satellite routing method based on dynamic celestial sphere region division to solve communication bottleneck problems such as network delay and increased packet loss rate that occur due to network dynamic changes in low-orbit satellites and the Internet of Things, and to achieve efficient operation of the network in a dynamic environment.

[0006] To achieve the above object, a first aspect embodiment of the present invention proposes a satellite routing method based on dynamic celestial sphere region division, including the following steps:

[0007] Using a first preset clustering algorithm to perform an initial division on multiple nodes to obtain multiple virtual regions with load balance;

[0008] Using a preset local flooding optimization strategy to optimize the multiple virtual regions with load balance to generate a preset propagation range of data packets;

[0009] According to the historical link delay data, use a preset recurrent neural network model to predict the delay and congestion conditions of the links in the preset propagation range, and based on the prediction results, dynamically adjust the path selection and traffic allocation strategies;

[0010] Based on the delay and load conditions of different paths in the path selection and traffic allocation strategies, calculate the additional delay of each path, and calculate the traffic to be allocated for each path based on the additional delay.

[0011] According to an embodiment of the present invention, the initial partitioning of satellites or nodes using a preset clustering algorithm includes:

[0012] Obtain the geographical coordinates and initial loads of the multiple nodes, and select a preset number of initial central points as virtual node centers;

[0013] Calculate the Euclidean distance from each node to each virtual node center, and based on the Euclidean distance, determine the virtual node center corresponding to each node;

[0014] Calculate the average position coordinates and load of each virtual node center, and re - execute the step of calculating the Euclidean distance from each node to each virtual node center and determining the virtual node center corresponding to each node based on the Euclidean distance until the multiple nodes meet the preset partitioning criteria.

[0015] According to an embodiment of the present invention, after the multiple nodes meet the preset partitioning criteria, it further includes:

[0016] Real - time monitor the load distribution of each virtual node center, and based on the monitoring results, judge whether the load of each virtual node center is balanced;

[0017] When the load is unbalanced, use a second preset clustering algorithm to dynamically adjust the node allocation in the corresponding virtual node center to obtain multiple load - balanced virtual regions.

[0018] According to an embodiment of the present invention, the use of a preset local flooding optimization strategy to optimize the multiple load - balanced virtual regions to generate a preset propagation range for data packets includes:

[0019] Calculate the shortest hop count between any two nodes among all nodes;

[0020] Based on the shortest hop count, determine the hop count limit for the corresponding nodes;

[0021] Obtain the hop count of the current data packet, select a node whose hop count limit is greater than or equal to the hop count of the current data packet as the target node, and generate a preset propagation range of the data packet based on the target node.

[0022] According to an embodiment of the present invention, the dynamically adjusting the path selection and traffic allocation strategy based on the prediction result includes:

[0023] Generate a delay matrix based on the prediction result;

[0024] Calculate the traffic allocation ratio of each path according to the delay matrix, and generate a traffic allocation matrix;

[0025] Dynamically adjust the path selection and traffic allocation strategy according to the traffic allocation matrix.

[0026] A satellite routing method based on dynamic celestial sphere region division proposed according to an embodiment of the present invention can obtain multiple load-balanced virtual regions by using a clustering algorithm to divide multiple nodes; use a local flooding optimization strategy to optimize the multiple virtual regions to generate a preset propagation range of the data packet; predict the delay and congestion conditions of the links in the preset propagation range by using a recurrent neural network model according to historical link delay data, and dynamically adjust the path selection and traffic allocation strategy based on the prediction result; calculate the additional delay of each path based on the delay and load conditions of different paths in the path selection and traffic allocation strategy, and calculate the traffic to be allocated for each path based on the additional delay. Thereby, communication bottleneck problems such as network delay and increased packet loss rate caused by network dynamic changes in low-earth orbit satellites and the Internet of Things are solved, and efficient operation of the network in a dynamic environment is realized.

[0027] To achieve the above object, an embodiment of the second aspect of the present invention proposes a satellite routing device based on dynamic celestial sphere region division, including:

[0028] A division module, configured to perform an initial division on multiple nodes by using a first preset clustering algorithm to obtain multiple load-balanced virtual regions;

[0029] A generation module, configured to optimize the multiple load-balanced virtual regions by using a preset local flooding optimization strategy to generate a preset propagation range of the data packet;

[0030] An adjustment module, configured to predict the delay and congestion conditions of the links in the preset propagation range by using a preset recurrent neural network model according to historical link delay data, and dynamically adjust the path selection and traffic allocation strategy based on the prediction result;

[0031] A calculation module, configured to calculate the additional delay of each path based on the delays and loads of different paths in the path selection and traffic allocation policies, and calculate the traffic to be allocated for each path based on the additional delay.

[0032] According to an embodiment of the present invention, the partitioning module is specifically configured to:

[0033] Obtain the geographical coordinates and initial loads of the multiple nodes, and select a preset number of initial central points as virtual node centers;

[0034] Calculate the Euclidean distance from each node to each virtual node center respectively, and determine the corresponding virtual node center for each node based on the Euclidean distance;

[0035] Calculate the average position coordinates and loads of each virtual node center, and re - execute the step of calculating the Euclidean distance from each node to each virtual node center respectively, and determining the corresponding virtual node center for each node based on the Euclidean distance until the multiple nodes meet the preset partitioning criteria.

[0036] According to an embodiment of the present invention, after the multiple nodes meet the preset partitioning criteria, the partitioning module is further configured to:

[0037] Monitor the load distribution of each virtual node center in real - time, and determine whether the load of each virtual node center is balanced based on the monitoring results;

[0038] When the load is unbalanced, use a second preset clustering algorithm to dynamically adjust the node allocation in the corresponding virtual node center to obtain multiple load - balanced virtual regions.

[0039] According to an embodiment of the present invention, the generating module is specifically configured to:

[0040] Calculate the shortest hop count between any two nodes among all nodes;

[0041] Determine the hop count limit for the corresponding nodes based on the shortest hop count;

[0042] Obtain the hop count of the current data packet, select the nodes whose hop count limit is greater than or equal to the hop count of the current data packet as target nodes, and generate a preset propagation range for the data packet based on the target nodes.

[0043] According to an embodiment of the present invention, the adjustment module is specifically configured to:

[0044] Generate a delay matrix based on the prediction result;

[0045] Calculate the traffic allocation ratio for each path according to the delay matrix, and generate a traffic allocation matrix;

[0046] Dynamically adjust the path selection and traffic allocation strategy according to the traffic allocation matrix.

[0047] A satellite routing device based on dynamic celestial sphere region division according to an embodiment of the present invention can obtain multiple load-balanced virtual regions by using a clustering algorithm to divide multiple nodes; use a local flooding optimization strategy to optimize multiple virtual regions to generate a preset propagation range for data packets; according to historical link delay data, use a recurrent neural network model to predict the delay and congestion conditions of links in the preset propagation range, and dynamically adjust the path selection and traffic allocation strategy based on the prediction results; calculate the additional delay of each path based on the delay and load conditions of different paths in the path selection and traffic allocation strategy, and calculate the traffic to be allocated for each path based on the additional delay. Thus, communication bottleneck problems such as network delay and increased packet loss rate caused by network dynamic changes in low-earth orbit satellites and the Internet of Things are solved, and the efficient operation of the network in a dynamic environment is realized.

[0048] To achieve the above object, an embodiment of the third aspect of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to implement a satellite routing method based on dynamic celestial sphere region division as described in the above embodiment.

[0049] To achieve the above object, an embodiment of the fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and the program is executed by a processor to be used to implement a satellite routing method based on dynamic celestial sphere region division as described in the above embodiment.

[0050] To achieve the above object, an embodiment of the fifth aspect of the present invention provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it is used to implement a satellite routing method based on dynamic celestial sphere region division as described in the above embodiment.

[0051] The additional aspects and advantages of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The above and / or additional aspects and advantages of the present invention will become obvious and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0053] Figure 1 FIG. [FIGURE NUMBER] is a flowchart of a satellite routing method based on dynamic celestial sphere region division according to an embodiment of the present invention;

[0054] Figure 2a It is a flowchart of another satellite routing method based on dynamic celestial sphere region division provided according to an embodiment of the present invention;

[0055] Figure 2b It is a flowchart of a dynamic celestial sphere region division method, a local flooding optimization method, and an intelligent path prediction method provided according to an embodiment of the present invention;

[0056] Figure 2c It is a flowchart of a multi-path load balancing method and an adaptive adjustment and fault tolerance mechanism method provided according to an embodiment of the present invention;

[0057] Figure 3 It is a schematic block diagram of a satellite routing device based on dynamic celestial sphere region division provided according to an embodiment of the present invention;

[0058] Figure 4 It is a schematic structural diagram of an electronic device provided according to an embodiment of the present invention. Detailed implementation manners

[0059] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as a limitation to the present invention.

[0060] A satellite routing method based on dynamic celestial sphere region division proposed according to an embodiment of the present invention will be described below with reference to the accompanying drawings.

[0061] Figure 1 It is a flowchart of a satellite routing method based on dynamic celestial sphere region division according to an embodiment of the present invention.

[0062] Before introducing a satellite routing method based on dynamic celestial sphere region division proposed according to an embodiment of the present invention, the related technologies will be introduced first.

[0063] For the traditional global flooding routing algorithm, although it can collect complete link information in the network, in a complex dynamic network, especially in a low-earth orbit satellite network and a large-scale Internet of Things, the calculation and communication overheads are too large to meet the real-time requirements of the network. The global flooding routing algorithm needs to send routing information to each node in the network. As the network scale expands, data packets flood and propagate throughout the network, occupying a large amount of bandwidth and computing resources, resulting in an overloaded system. In addition, the traditional global flooding routing algorithm can only rely on the current link state and lacks the ability to predict future network changes. When the network load suddenly changes, it often cannot adjust the path in time, causing communication bottleneck problems such as network delay and packet loss rate increase.

[0064] At the same time, the static node partitioning strategy also has significant limitations in dealing with dynamic network changes. In satellite networks and the Internet of Things, the node load and topology change frequently, while the node regions divided statically cannot be adaptively adjusted according to these changes, resulting in low resource utilization and decreased network efficiency. The fixed partitioned routing regions are difficult to flexibly handle the load fluctuations on different paths, easily causing congestion on some paths, and further affecting the overall network performance. Especially in the case of high load, the overload problem of a single path or node will be exacerbated, and the traditional static partitioning strategy cannot perform effective load balancing adjustment, further increasing the risk of network communication bottlenecks.

[0065] Based on the above problems, the embodiments of the present invention propose a satellite routing method based on dynamic celestial sphere region partitioning, which combines the virtual node method based on dynamic celestial sphere region partitioning with the local flooding optimization strategy, aiming to solve the communication bottleneck problems such as network delay and increased packet loss rate due to network dynamic changes in low-earth orbit satellites and the Internet of Things, improve the routing efficiency in complex dynamic networks such as low-earth orbit satellite networks and large-scale Internet of Things, reduce communication delay and enhance the self-adaptability and fault tolerance of the network.

[0066] Next, a satellite routing method based on dynamic celestial sphere region partitioning proposed by the embodiments of the present invention will be elaborated in detail.

[0067] Exemplarily, as Figure 1 shown, the satellite routing method based on dynamic celestial sphere region partitioning includes the following steps:

[0068] In step S101, a first preset clustering algorithm is used to initially partition multiple nodes to obtain multiple load-balanced virtual regions.

[0069] Among them, the first preset clustering algorithm refers to the K-Means (K-Means Clustering Algorithm) clustering algorithm, which is an unsupervised learning algorithm. The purpose is to divide the samples in the dataset into multiple clusters (or classes) according to a certain similarity metric, so that the samples within the same cluster are highly similar to each other, while the samples in different clusters are less similar.

[0070] Specifically, the K-Means clustering algorithm is used to initially partition multiple satellites or nodes in the network. Here, K represents the number of virtual regions. Through an iterative optimization process, the satellites or nodes are assigned to K clusters, and each node can be assigned to different virtual node regions based on its geographical location and current load information. In this way, multiple nodes can be effectively partitioned into several clusters, with each cluster represented by a virtual node, thereby achieving the initial partitioning of the nodes and obtaining multiple load-balanced virtual regions. This partitioning helps optimize the load balancing and resource allocation of the network, improving the overall performance and efficiency of the system.

[0071] The following details how to initially partition satellites or nodes using a preset clustering algorithm.

[0072] As a possible implementation, in some embodiments, initially partitioning satellites or nodes using a preset clustering algorithm includes: obtaining the geographical coordinates and initial loads of multiple nodes, and selecting a preset number of initial center points as virtual node centers; calculating the Euclidean distance from each node to each virtual node center respectively, and based on the Euclidean distance, determining the virtual node center corresponding to each node; calculating the average position coordinates and load of each virtual node center, and re-executing the steps of calculating the Euclidean distance from each node to each virtual node center respectively and, based on the Euclidean distance, determining the virtual node center corresponding to each node until multiple nodes meet the preset partitioning criteria.

[0073] Among them, the Euclidean distance is a method for measuring the shortest distance between two points in geometry, and is commonly used to measure the straight-line distance between two points in a multi-dimensional space. The initial load refers to the initial workload or task volume of a node, which is usually related to the performance and resource allocation of the node.

[0074] Specifically, first input the geographical location coordinates and initial load conditions of each node in the network, then select K initial clustering center points (these points are preset and used to represent the central position of each cluster) from these nodes as virtual node centers to represent the central position of each cluster. Next, calculate the Euclidean distance from each node to these center points , and assign each node to the center point closest to it to form a preliminary clustering. Then, recalculate the new position of each clustering center point, that is, the average position of the coordinates of all nodes assigned to this center point, and update the load information of this cluster. Finally, use the updated virtual node center position and load conditions to repeat the process of calculating the Euclidean distance from each node to these centers and reassigning the nodes to the corresponding clustering centers according to this distance until the position of the center point and the assignment of the nodes no longer change (i.e., meet the preset partitioning criteria), indicating that the clustering process is complete.

[0075] Further, in some embodiments, after multiple nodes meet the preset partitioning criteria, the following steps are also included: real-time monitoring of the load distribution at the center of each virtual node, and based on the monitoring results, determining whether the load of each virtual node center is balanced; when the load is unbalanced, using a second preset clustering algorithm to dynamically adjust the node allocation in the corresponding virtual node center to obtain multiple load-balanced virtual regions.

[0076] Among them, the second preset clustering algorithm refers to the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm, which is an algorithm for data clustering. Its core idea is to group according to the density of data points. In the DBSCAN algorithm, data points are divided into three categories: core points, border points, and noise points. A core point is a point that has enough other points in its neighborhood. A border point is a point that is in the neighborhood of a core point but is not itself a core point, while a noise point is a point that does not have enough points in the neighborhood of any core point. The advantage of the DBSCAN algorithm is that it does not require pre-setting the number of clusters, can identify clusters of any shape, and can effectively handle noise points.

[0077] Specifically, the system will monitor the load distribution of each virtual node center in real time, and then use Equation (1) combined with the monitored data to determine whether the workload of these virtual node centers has reached a balanced state.

[0078]

[0079] Among them, is a node, is the load of the node.

[0080] When the load is unbalanced, the second preset clustering algorithm can be used to dynamically adjust the node allocation in the corresponding virtual node center to ensure that the workload of all virtual regions reaches a balanced state.

[0081] It should be noted that if the loads of all nodes are close to their average values, it indicates that the load distribution is balanced; if the loads of some nodes are much higher or lower than those of other nodes, it indicates that the load is unbalanced.

[0082] The specific calculation method for determining whether the load is balanced is as follows:

[0083] (1) Calculate the average load of all nodes, as shown in Equation (2):

[0084]

[0085] Among them, is a node, is the load, is the number of nodes participating in the calculation in the network. Among them, the node load is defined as the total traffic (number of data packets or bandwidth consumption) processed by the node per unit time.

[0086] (2)Calculate the standard deviation of the node load, as shown in Equation (3):

[0087]

[0088] When the standard deviation meets the preset threshold range it can be considered that the load distribution is balanced. If the standard deviation exceeds the preset threshold range then it is considered that the load distribution is unbalanced.

[0089] (3)Use the second preset clustering algorithm (DBSCAN algorithm) to dynamically adjust the node allocation in the corresponding virtual node center.

[0090] ① Define the node position: The position of each node can be defined by its Euclidean distance from the virtual node center, forming the spatial distribution of the data, that is, the distribution of the nodes in space;

[0091] ② Run DBSCAN clustering: Use the load and position of the nodes as input data and perform clustering using DBSCAN. The main steps of the DBSCAN algorithm are as follows:

[0092] a. Select appropriate neighborhood radius and minimum number of points as the parameters of the DBSCAN algorithm.

[0093] b. Run the DBSCAN algorithm on all nodes and mark the core points, boundary points and noise points.

[0094] c. In the clustering, nodes with similar loads are divided into a virtual node cluster. By dynamically adjusting the division of the virtual node center, try to balance the loads of each node.

[0095] ③ Adjust the virtual node center: According to the clustering results of the DBSCAN algorithm, adjust the position of the virtual node center to make it close to the area where the load is concentrated, ensuring that each virtual node can better handle the node loads around it to improve the efficiency and response speed of the entire system.

[0096] ④ Loop monitoring and adjustment: Continuously monitor the node load distribution, and regularly recalculate the average load and load standard deviation of the nodes to evaluate the load balance situation.

[0097] ⑤ Dynamically adjust node partitioning: If the load distribution is unbalanced, the DBSCAN algorithm can be re-run to adjust the partitioning of virtual nodes until an equilibrium state is reached, thereby obtaining multiple load-balanced virtual regions.

[0098] In step S102, using a preset local flooding optimization strategy, optimize multiple load-balanced virtual regions to generate a preset propagation range for data packets.

[0099] It can be understood that to prevent a large amount of unpurposed propagation, i.e., global flooding, when sending data packets in the network, which leads to a waste of network resources, the preset local flooding optimization strategy can be used to limit the area where data packets propagate to improve network efficiency. By optimizing multiple load-balanced virtual regions, it is ensured that data packets only propagate within the preset propagation range, rather than spreading unrestrictedly throughout the network, so as to reduce communication overhead.

[0100] Next, a detailed description will be given on how to use the preset local flooding optimization strategy to optimize multiple load-balanced virtual regions to generate a preset propagation range for data packets.

[0101] As a possible implementation method, in some embodiments, using the preset local flooding optimization strategy to optimize multiple load-balanced virtual regions to generate a preset propagation range for data packets includes: calculating the shortest hop count between any two nodes among all nodes; determining the hop count limit for the corresponding nodes based on the shortest hop count; obtaining the hop count of the current data packet, selecting the nodes whose hop count limit is greater than or equal to the hop count of the current data packet as target nodes, and generating a preset propagation range for the data packet based on the target nodes.

[0102] Specifically, to avoid the high overhead of global flooding, embodiments of the present invention can use hop count limits to control the propagation range of data packets. First, a comprehensive calculation can be performed on all nodes in the network to determine all possible shortest paths between them. This step involves complex algorithms such as the Dijkstra algorithm (also known as Dijkstra's algorithm or Dijkstra's shortest path algorithm) or the Floyd-Warshall algorithm (also known as the Floyd-Warshall algorithm), to ensure that the shortest hop count between any two nodes can be accurately found. After completing this step, set a local hop count limit for each node , which is used to control the propagation range of data packets in the network.

[0103] When a data packet propagates in the network, it can start from the source node and send it to its adjacent nodes in a flooding manner. By introducing a hop count limit , which means that after each node receives the data packet, it will check its hop count counter. If the value of the hop count counter is less than or equal to , then the data packet will continue to be flooded and propagated to adjacent nodes; on the contrary, if the value of the hop counter exceeds , then the data packet will no longer be propagated, thus avoiding unnecessary communication overhead.

[0104] In this way, fast routing selection can be achieved within a local area. Since the data packet is only propagated within a limited hop range, redundant data transmission in the network can be significantly reduced, improving the overall network efficiency.

[0105] Among them, is calculated from the shortest hop count between nodes:

[0106]

[0107] In step S103, according to the historical link delay data, use a preset recurrent neural network model to predict the delay and congestion conditions of the links in the preset propagation range, and based on the prediction results, dynamically adjust the path selection and traffic allocation strategies.

[0108] Among them, the preset recurrent neural network model can be LSTM (Long Short-Term Memory), which is a special structure of recurrent neural network.

[0109] That is to say, through the training and learning of historical link delay data, the recurrent neural network can capture the time series characteristics of link delay and predict the future delay and congestion status according to these characteristics.

[0110] The predicted values of link path delay and congestion status The calculation method is as follows:

[0111]

[0112] Among them, is the historical link delay data, is the link path.

[0113] According to the prediction results, the path selection follows the following formula:

[0114]

[0115] Among them, is the actual delay of the current link, is the weight to adjust the influence of future delay prediction, is a small constant used to avoid the denominator being zero, is the expected distance from the source point O to the target point A, is the actual distance from the source point O to node of.

[0116] Based on these prediction results, corresponding measures can be taken to optimize network performance. That is, the path selection strategy can be dynamically adjusted to ensure that data packets can be transmitted through paths with lower latency and smoothness, thereby improving the overall network transmission efficiency. At the same time, according to the predicted congestion situation, traffic can be reasonably allocated to avoid overloading of certain links, further enhancing the stability and reliability of the network. Through this prediction-based dynamic adjustment mechanism, the uncertainties and changes in the network can be effectively addressed, ensuring the optimal configuration and utilization of network resources.

[0117] Among them, during the training process of the preset long short-term memory (LSTM) recurrent neural network model, in order to measure the prediction effect of the model, the mean squared error can be used as the loss function to measure the difference between the predicted path latency and the actual latency. At the same time, the Adam (Adaptive Moment Estimation) algorithm has the advantage of an adaptive learning rate and can accelerate convergence. Therefore, the LSTM model can use the Adam optimizer for gradient descent and parameter update, and set the initial learning rate to 0.001. In addition, during the data processing stage, the collected historical data needs to be scientifically and reasonably divided into a training set, a validation set, and a test set. The specific ratio can be: 80% of the data is used for training, 10% of the data is used for validation, and the remaining 10% of the data is used for testing.

[0118] The specific model training steps are as follows:

[0119] (1) Initialize the LSTM model and its parameters.

[0120] (2) Input the historical latency data into the LSTM model for training according to the time window.

[0121] (3) Calculate the loss (mean squared error) and update the model weights through the Adam optimizer.

[0122] (4) After each round of training, use the validation set data to validate the model to avoid overfitting.

[0123] (5) Determine the hyperparameters (number of hidden units, learning rate, etc.) through cross-validation.

[0124] (6) Continue training until the loss function converges (the error no longer decreases significantly).

[0125] After the LSTM model training is completed, assuming there is the latest historical latency data of a certain path currently, after inputting it into the LSTM model, the model will output the latency prediction values for a period of time in the future. According to the prediction results of the LSTM model, the system can dynamically adjust the path selection and traffic allocation strategies.

[0126] The following elaborates in detail on how to dynamically adjust the path selection and traffic allocation strategies based on the prediction results.

[0127] As a possible implementation method, in some embodiments, dynamically adjusting the path selection and traffic allocation strategies based on the prediction results includes: generating a delay matrix based on the prediction results; calculating the traffic allocation ratio for each path according to the delay matrix, and generating a traffic allocation matrix; and dynamically adjusting the path selection and traffic allocation strategies according to the traffic allocation matrix.

[0128] Specifically, assume that there are N paths in the network, and the delay of each path can be predicted by the LSTM model to generate a delay matrix.

[0129] Delay matrix is a matrix composed of the predicted future path delays by the LSTM model, with a dimension of 1×N, is the th path's delay prediction value.

[0130] According to the delay prediction value of each path in the delay matrix , the traffic allocation ratio for the corresponding path can be calculated : :

[0131]

[0132] Based on the traffic allocation ratio of each path, a traffic allocation matrix can be generated.

[0133] Traffic allocation matrix has a dimension of 1×N and satisfies the following constraints:

[0134]

[0135] That is, the sum of the traffic allocation ratios of all paths is 1.

[0136] It can be seen from Equation (7) that when allocating traffic according to the delay values predicted by the LSTM model, the following principle is followed:

[0137] If the delay of a certain network path is relatively high, the data traffic allocated to this path should be appropriately reduced to prevent network congestion; conversely, if the delay of a certain path is relatively low, its traffic allocation should be appropriately increased to improve the overall network load balancing efficiency.

[0138] It is understandable that during the actual deployment process, various dynamic changes often occur, such as the addition or removal of devices, fluctuations in links, etc. To cope with these changes, the LSTM model can be continuously optimized through its online learning ability. Specifically, whenever new path delay data is generated in the network, this data can be captured in real time and added to the model as new training samples. In this way, the LSTM model can update its internal parameters in real time, thereby better adapting to changes in the network state. This continuous online learning mechanism enables the LSTM model to maintain high prediction accuracy and robustness when facing a complex network environment.

[0139] In step S104, based on the delays and loads of different paths in the path selection and traffic allocation strategy, calculate the additional delay of each path, and calculate the traffic to be allocated for each path based on the additional delay.

[0140] Specifically, the system can calculate the additional delay of each path according to the delays and loads on different paths :

[0141]

[0142] where is the packet length, is the bandwidth rate of the inter-satellite link.

[0143] According to the additional delay situation and the current load of each path, the system can dynamically adjust and allocate network traffic in real time. Thereby, ensuring that the load of each path is evenly distributed, avoiding performance bottlenecks due to overload on some paths, and thus improving the overall stability and efficiency of the network.

[0144] The path load balancing formula is:

[0145]

[0146] where is the load of the path, is the adjustment coefficient.

[0147] The optimization formula for path selection is as follows:

[0148]

[0149] The dynamic allocation method is:

[0150] (1) Initial traffic allocation: The system first makes an initial allocation according to the delays and loads of the paths.

[0151] (2) Iterative adjustment: In each iteration, according to the actual delay changes and the effect of load allocation, make a small adjustment to the traffic to avoid drastic fluctuations.

[0152] (3)Convergence condition: Set a convergence condition such that when the standard deviation of the load on all paths is less than a certain threshold the traffic adjustment stops.

[0153] Thus, by selecting paths with lower latency, load balancing on multiple paths can be achieved, reducing network congestion.

[0154] In addition, by monitoring the network status in real time and predicting potential failures, the algorithm's adaptive adjustment and fault tolerance capabilities can automatically adjust the routing when a failure occurs to ensure the continuity and reliability of data transmission.

[0155] Specifically, the system can use the LSTM model to monitor the status of nodes and links in real time and predict potential failures. The results of the failure prediction are as follows:

[0156]

[0157] The system has the ability to automatically discover the network topology. Through network scanning, routing protocol interaction, etc., it can construct a topology view of the entire network. This topology view includes all nodes, links, and their connection relationships in the network, providing a basic network map for path adjustment.

[0158] When the failure probability of the system exceeds a pre-set threshold, the system can automatically perform path adjustment based on the network topology structure and use a path search algorithm to find alternative paths. This process ensures that the data flow can effectively avoid the faulty nodes, thereby ensuring the stability and reliability of the entire network. In this way, the system can quickly respond when a failure occurs, re-plan the data transmission path, avoid data staying or being lost at the faulty nodes, and ensure the smoothness and efficiency of the data flow.

[0159] When the faulty node or faulty link returns to normal, the system will re-evaluate the network topology structure, identify the repaired nodes or links, and re-integrate them into the virtual node cluster. By updating the path selection, the system can re-distribute the data traffic to avoid passing through the still faulty parts, thereby ensuring the continuity and stability of data transmission.

[0160] To facilitate those skilled in the art to further understand a satellite routing method based on dynamic celestial sphere region division proposed in the embodiments of the present invention, the following is combined with Figures 2a - 2c for further explanation.

[0161] As Figure 2a shown, another satellite routing method based on dynamic celestial sphere region division includes the following steps:

[0162] Step S201, Dynamic Celestial Sphere Region Division. That is, through dynamic celestial sphere region division, the network structure is adaptively adjusted according to the geographical location and real-time load of nodes to ensure load balancing.

[0163] Step S202, Local Flooding Optimization. That is, by adopting local flooding optimization technology, the propagation range of data packets is restricted, the link status is quickly collected, and the communication overhead is reduced.

[0164] Step S203, Intelligent Path Prediction. That is, the LSTM model is used to predict the link delay, and the optimal path is selected in advance to avoid future congestion risks.

[0165] Step S204, Multi-Path Load Balancing. That is, the system dynamically adjusts the traffic distribution through multi-path load balancing to avoid overloading of a single path and improve the transmission efficiency.

[0166] Step S205, Adaptive Adjustment and Fault Tolerance Mechanism. That is, by real-time monitoring the network status and predicting potential faults, based on the adaptive adjustment and fault tolerance capabilities of the algorithm, the routing is automatically adjusted when a fault occurs to ensure the continuity and reliability of data transmission.

[0167] Furthermore, as shown in (a) of Figure 2b the dynamic celestial sphere region division includes the following steps:

[0168] Step S2011, Input the geographical coordinates and initial load of the nodes.

[0169] Step S2012, Select K initial centers as virtual node centers.

[0170] Step S2013, Calculate the Euclidean distance from all nodes to the virtual node centers.

[0171] Step S2014, Allocate the nodes to the virtual node centers closest to them respectively.

[0172] Step S2015, Calculate the average position and load of each virtual node center.

[0173] Step S2016, Recalculate the average position and load of each virtual node center.

[0174] Step S2017, Determine whether the positions of the virtual node centers and the node allocation have changed. If so, execute Step S2013; otherwise, execute Step S2018.

[0175] Step S2018, First, define the node load, secondly, calculate the average load of all nodes, and finally, calculate the standard deviation of all nodes.

[0176] Step S2019, Use the DBSCAN clustering algorithm to adjust the virtual node division.

[0177] Further, as shown in (b) of Figure 2b , the local Hongfan optimization includes the following steps:

[0178] Step S2021, calculate the shortest hop count between all node pairs in the network.

[0179] Step S2022, determine whether the shortest hop count is less than or equal to the hop count limit. If so, execute Step S2031; otherwise, execute Step S2013.

[0180] Further, as shown in (c) of Figure 2b , the intelligent path prediction includes the following steps:

[0181] Step S2031, train the LSTM model.

[0182] Step S2032, use the LSTM model to predict the delay and congestion conditions of future paths.

[0183] Step S2033, input the delay matrix predicted by the LSTM model.

[0184] Step S2034, calculate the traffic allocation ratio for each path.

[0185] Step S2035, dynamically adjust the traffic allocation matrix.

[0186] Step S2036, dynamic traffic adjustment.

[0187] Further, as shown in (d) of Figure 2c , the multi-path load balancing includes the following steps:

[0188] Step S2041, calculate the additional delay.

[0189] Step S2042, perform traffic scheduling.

[0190] Step S2043, determine whether the load standard deviation of all paths is less than the preset threshold. If so, execute Step S2044; otherwise, execute Step S2042.

[0191] Further, as shown in (e) of Figure 2c , the adaptive adjustment and fault tolerance mechanism includes the following steps:

[0192] Step S2051, perform fault prediction.

[0193] Step S2052, fault recovery.

[0194] In summary, a satellite routing method based on dynamic celestial sphere region division proposed in an embodiment of the present invention has the following advantages:

[0195] (1) High - efficiency adaptability: By dynamically adjusting the division of the celestial sphere region and real - time updating the virtual node division, the system can adapt to changes in the network structure, accommodate load fluctuations in a dynamic network, and ensure the efficient operation of the network.

[0196] (2) Reducing network overhead: The local flooding optimization mechanism reduces the excessive computational and communication overhead caused by traditional global flooding by restricting the hop count of data packets, improving the routing efficiency of the system.

[0197] (3) Forward - looking path selection: Intelligent path prediction is not only based on the current link state but also can predict future link delays through machine learning to avoid potential network congestion, thereby optimizing the data transmission path and improving network performance.

[0198] (4) Fault - tolerance mechanism: Through fault prediction and adaptive adjustment, the system can automatically switch paths when a fault occurs, ensuring the stability and fault - tolerance of the network.

[0199] A satellite routing method based on dynamic celestial sphere region division proposed according to an embodiment of the present invention can obtain multiple load - balanced virtual regions by using a clustering algorithm to divide multiple nodes; use a local flooding optimization strategy to optimize multiple virtual regions to generate a preset propagation range for data packets; predict the delay and congestion conditions of links in the preset propagation range based on historical link delay data by using a recurrent neural network model, and dynamically adjust the path selection and traffic allocation strategy based on the prediction results; calculate the additional delay of each path based on the delay and load conditions of different paths in the path selection and traffic allocation strategy, and calculate the traffic to be allocated for each path based on the additional delay. Thus, it solves the communication bottleneck problems such as network delay and rising packet loss rate due to network dynamic changes in low - earth - orbit satellites and the Internet of Things, and realizes the efficient operation of the network in a dynamic environment.

[0200] Next, a satellite routing device based on dynamic celestial sphere region division proposed according to an embodiment of the present invention is described with reference to the accompanying drawings.

[0201] Figure 3 It is a block diagram of a satellite routing device based on dynamic celestial sphere region division according to an embodiment of the present invention.

[0202] As Figure 3 shown, the satellite routing device 10 based on dynamic celestial sphere region division includes: a division module 100, a generation module 200, an adjustment module 300, and a calculation module 400.

[0203] Among them, the division module 100 is used to initially divide multiple nodes by using a first preset clustering algorithm to obtain multiple load - balanced virtual regions;

[0204] The generation module 200 is configured to optimize multiple load-balanced virtual regions by using a preset local flooding optimization strategy, and generate a preset propagation range of data packets;

[0205] The adjustment module 300 is configured to predict the delay and congestion conditions of the links in the preset propagation range according to historical link delay data by using a preset recurrent neural network model, and dynamically adjust the path selection and traffic allocation strategies based on the prediction results;

[0206] The calculation module 400 is configured to calculate the additional delay of each path based on the delay and load conditions of different paths in the path selection and traffic allocation strategies, and calculate the traffic to be allocated for each path based on the additional delay.

[0207] Furthermore, in some embodiments, the partitioning module 100 is specifically configured to:

[0208] Obtain the geographical coordinates and initial loads of multiple nodes, and select a preset number of initial central points as the virtual node centers;

[0209] Calculate the Euclidean distance from each node to each virtual node center respectively, and determine the corresponding virtual node center for each node based on the Euclidean distance;

[0210] Calculate the average position coordinates and loads of each virtual node center, and re-execute the steps of calculating the Euclidean distance from each node to each virtual node center respectively, and determining the corresponding virtual node center for each node based on the Euclidean distance until multiple nodes meet the preset partitioning criteria.

[0211] Furthermore, in some embodiments, after multiple nodes meet the preset partitioning criteria, the partitioning module 100 is further configured to:

[0212] Monitor the load distribution of each virtual node center in real time, and determine whether the load of each virtual node center is balanced based on the monitoring results;

[0213] When the load is unbalanced, dynamically adjust the node allocation in the corresponding virtual node center by using a second preset clustering algorithm to obtain multiple load-balanced virtual regions.

[0214] Furthermore, in some embodiments, the generation module 200 is specifically configured to:

[0215] Calculate the shortest hop count between any two nodes among all nodes;

[0216] Determine the hop count limit value of the corresponding node based on the shortest hop count;

[0217] Obtain the hop count of the current data packet, select a node whose hop count limit is greater than or equal to the hop count of the current data packet as the target node, and generate a preset propagation range of the data packet based on the target node.

[0218] Further, in some embodiments, the adjustment module 300 is specifically configured to:

[0219] Generate a delay matrix based on the prediction result;

[0220] Calculate the traffic allocation ratio of each path according to the delay matrix, and generate a traffic allocation matrix;

[0221] Dynamically adjust the path selection and traffic allocation strategies according to the traffic allocation matrix.

[0222] It should be noted that the foregoing explanation of the embodiments of a satellite routing method based on dynamic celestial sphere region division also applies to a satellite routing device based on dynamic celestial sphere region division in this embodiment, and will not be elaborated here.

[0223] According to a satellite routing device based on dynamic celestial sphere region division provided by an embodiment of the present invention, by using a clustering algorithm to divide multiple nodes, multiple load-balanced virtual regions can be obtained; by using a local flooding optimization strategy to optimize multiple virtual regions, a preset propagation range of data packets can be generated; according to historical link delay data, a recurrent neural network model is used to predict the delay and congestion conditions of links in the preset propagation range, and the path selection and traffic allocation strategies are dynamically adjusted based on the prediction results; the additional delay of each path is calculated based on the delay and load conditions of different paths in the path selection and traffic allocation strategies, and the traffic to be allocated for each path is calculated based on the additional delay. Thus, the communication bottleneck problems such as network delay and rising packet loss rate caused by network dynamic changes in low-earth orbit satellites and the Internet of Things are solved, and the efficient operation of the network in a dynamic environment is realized.

[0224] Figure 4 The structural schematic diagram of an electronic device provided by an embodiment of the present invention. The electronic device may include:

[0225] A memory 401, a processor 402, and a computer program stored on the memory 401 and executable on the processor 402.

[0226] When the processor 402 executes the program, it implements a satellite routing method based on dynamic celestial sphere region division provided in the foregoing embodiment.

[0227] Further, the electronic device further includes:

[0228] A communication interface 403 for communication between the memory 401 and the processor 402.

[0229] A memory 401 for storing a computer program that can run on a processor 402.

[0230] The memory 401 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0231] If the memory 401, the processor 402, and the communication interface 403 are implemented independently, the communication interface 403, the memory 401, and the processor 402 can be interconnected through a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 4 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0232] Optionally, in a specific implementation, if the memory 401, the processor 402, and the communication interface 403 are integrated on a chip, the memory 401, the processor 402, and the communication interface 403 can communicate with each other through an internal interface.

[0233] The processor 402 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0234] The embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, a satellite routing method based on dynamic celestial sphere region division as described above is implemented.

[0235] The embodiments of the present invention also provide a computer program product, which includes a computer program, and when the computer program is executed by a processor, a satellite routing method based on dynamic celestial sphere region division as described above is implemented.

[0236] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0237] In the description of this specification, descriptions with reference to terms such as "an embodiment", "some embodiments", "examples", "specific examples", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0238] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A satellite routing method based on dynamic celestial sphere region division, characterized in that: The following steps are involved: Using a first preset clustering algorithm to initially divide the multiple nodes to obtain multiple load-balanced virtual areas; Utilizing a preset local flooding optimization strategy, the plurality of load-balanced virtual areas are optimized to generate a preset propagation range of a data packet; According to the historical link delay data, a preset recurrent neural network model is used to predict the delay and congestion of the links in the preset propagation range, and based on the prediction results, the path selection and traffic distribution strategies are dynamically adjusted; Based on the delay and load conditions of different paths in the path selection and traffic distribution strategy, calculate the additional delay of each path, and calculate the traffic to be distributed on each path based on the additional delay; The method of optimizing the multiple load-balanced virtual areas and generating a preset propagation range of a data packet by using a preset local flooding optimization strategy includes: calculating the shortest hop number between any two nodes among all nodes, determining the hop number limit of the corresponding node based on the shortest hop number, obtaining the hop number of the current data packet, selecting a node whose hop number limit is greater than or equal to the hop number of the current data packet as a target node, and generating the preset propagation range of the data packet based on the target node; The additional delay of each path is: in, is the additional delay for each path, is the packet length, is the bandwidth rate of the intersatellite link.

2. The satellite routing method based on dynamic celestial sphere region division according to claim 1, characterized in that: The using a first preset clustering algorithm to initially divide the multiple nodes includes: Obtaining geographic coordinates and initial loads of the plurality of nodes, and selecting a preset number of initial center points as virtual node centers; Calculate the Euclidean distance from each node to each virtual node center respectively, and determine the virtual node center corresponding to each node based on the Euclidean distance; Calculate the average position coordinates and load of each virtual node center, and re-execute the steps of respectively calculating the Euclidean distance from each node to each virtual node center, and determining the virtual node center corresponding to each node based on the Euclidean distance, until the multiple nodes meet the preset division criteria; The average position coordinates of each virtual node center are the average values ​​of the geographic coordinates of all nodes assigned to the current virtual node center.

3. The satellite routing method based on dynamic celestial sphere region division according to claim 2, characterized in that: After the plurality of nodes meet the preset division criteria, the method further includes: Monitor the load distribution of each virtual node center in real time, and determine whether the load of each virtual node center is balanced based on the monitoring results; When the load is unbalanced, the node allocation in the corresponding virtual node center is dynamically adjusted using a second preset clustering algorithm to obtain the multiple load-balanced virtual areas.

4. The satellite routing method based on dynamic celestial sphere region division according to claim 1, characterized in that: The method of dynamically adjusting the path selection and traffic distribution strategy based on the prediction result includes: Based on the prediction results, generating a delay matrix; According to the delay matrix, the flow distribution ratio of each path is calculated, and a flow distribution matrix is ​​generated; The path selection and traffic distribution strategy are dynamically adjusted according to the traffic distribution matrix.

5. A satellite routing device based on dynamic celestial sphere region division, characterized in that: include: A partitioning module, used to perform initial partitioning on multiple nodes using a first preset clustering algorithm to obtain multiple load-balanced virtual areas; A generating module, configured to optimize the plurality of load-balanced virtual areas by using a preset local flooding optimization strategy, and generate a preset propagation range of a data packet; An adjustment module is used to predict the delay and congestion of the links in the preset propagation range according to the historical link delay data using a preset recurrent neural network model, and dynamically adjust the path selection and traffic distribution strategy based on the prediction results; A calculation module, configured to calculate an additional delay of each path based on the delay and load conditions of different paths in the path selection and traffic distribution strategy, and calculate the traffic to be distributed on each path based on the additional delay; The generation module is specifically used to: calculate the shortest hop number between any two nodes among all nodes, determine the hop number limit of the corresponding node based on the shortest hop number, obtain the hop number of the current data packet, select a node whose hop number limit is greater than or equal to the hop number of the current data packet as a target node, and generate a preset propagation range of the data packet based on the target node; The additional delay of each path is: in, is the additional delay for each path, is the packet length, is the bandwidth rate of the intersatellite link.

6. The satellite routing device based on dynamic celestial sphere region division according to claim 5, characterized in that: The division module is specifically used for: Obtaining geographic coordinates and initial loads of the plurality of nodes, and selecting a preset number of initial center points as virtual node centers; Calculate the Euclidean distance from each node to each virtual node center respectively, and determine the virtual node center corresponding to each node based on the Euclidean distance; Calculate the average position coordinates and load of each virtual node center, and re-execute the steps of respectively calculating the Euclidean distance from each node to each virtual node center, and determining the virtual node center corresponding to each node based on the Euclidean distance, until the multiple nodes meet the preset division criteria; The average position coordinates of each virtual node center are the average values ​​of the geographic coordinates of all nodes assigned to the current virtual node center.

7. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a satellite routing method based on dynamic celestial region division as described in any one of claims 1 to 4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement a satellite routing method based on dynamic celestial region division as described in any one of claims 1 to 4.

9. A computer program product, characterized in that It comprises a computer program, which, when executed by a processor, is used to implement a satellite routing method based on dynamic celestial sphere area division as described in any one of claims 1 to 4.

Images