Intelligent satellite network routing method and device based on position orientation

Abstract

The invention discloses an intelligent satellite network routing method and device based on position orientation, and relates to the technical field of satellite network communication, and the method comprises the steps: determining a satellite node data set according to a simulation analysis result of a satellite network, the satellite node data set comprises node data of all satellite nodes in the satellite network, and the node data is transmitted to the satellite network; the node data comprises geographical location information, lifting rail surface information, destination node geographical location information and a routing decision result; performing model training according to the satellite node data set to obtain a decision tree model; performing model compression on the decision tree model to obtain a target decision tree model; and converting the target decision tree model into a matching action entry suitable for a switch, and deploying the matching action entry into the switch, so that a data message is forwarded according to the matching action entry after entering the switch. According to the invention, occupation of network bandwidth resources can be greatly reduced.

Description

Technical Field

[0001] The present application relates to the field of satellite network communication technology, and in particular to a location-oriented intelligent satellite network routing method and device. Background Art

[0002] With the deployment and actual commercial operation of large-scale low-orbit satellite networks represented by Starlink, large-scale low-orbit satellite constellations have gradually become the mainstream development trend. By deploying a certain number of satellites, a satellite Internet with global coverage can be formed. Due to the characteristics of satellite motion, the intersatellite links between satellites in different orbits will switch when passing through polar regions (polar orbit constellations) or high latitudes (inclined orbit constellations). The intersatellite links between adjacent orbits will be disconnected and re-established twice in each orbital cycle, resulting in a highly dynamic satellite network topology.

[0003] Traditional routing algorithms are mainly based on relatively static network topology and configuration information. For example, when the OSPF protocol changes in the network, the corresponding node will send a link state notification protocol message to notify the neighboring node of its own state change. The neighboring node will then notify the next-level neighbor, and this will be repeated until all nodes in the network have synchronized the network state. All nodes need to run the routing algorithm to calculate the next-hop route of the entire network and update it to the routing table. In a highly dynamic satellite network, these notifications will occupy link bandwidth.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present application is to provide a location-oriented intelligent satellite network routing method and device, aiming to solve the technical problem of large link bandwidth occupancy caused by the strong dynamics of satellite network topology.

[0006] To achieve the above objectives, the present application proposes a location-oriented intelligent satellite network routing method, the location-oriented intelligent satellite network routing method comprising:

[0007] Determine a satellite node data set according to the simulation analysis result of the satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, ascending and descending track surface information, geographic location information of a destination node, and a routing decision result;

[0008] Performing model training according to the satellite node data set to obtain a decision tree model;

[0009] Performing model compression on the decision tree model to obtain a target decision tree model;

[0010] The target decision tree model is converted into matching action entries suitable for the switch, and the matching action entries are deployed in the switch, so that after the data message enters the switch, the message forwarding is completed according to the matching action entries.

[0011] In one embodiment, the step of determining the satellite node data set according to the simulation analysis result of the satellite network includes:

[0012] Expanding the simulation function on the NS3 simulator to obtain the position prediction information of all satellite nodes in the satellite network at different times;

[0013] Determine the angle between each satellite node and its neighboring node according to the position prediction information;

[0014] Determine the link status of each satellite node and the neighboring node according to the connection angle;

[0015] After synchronizing the link status of the entire network according to the link status, obtaining a routing decision result of each satellite node for the destination node;

[0016] A satellite node dataset is generated based on the geographic location information of each satellite node, the geographic location information of the destination node, and the routing decision results.

[0017] In one embodiment, the step of performing model training according to the satellite node data set to obtain a decision tree model includes:

[0018] Load satellite node dataset;

[0019] After converting the geographical location information of the node data in the satellite node data set and the geographical location information of the destination node into an area number, and converting the routing decision result of the node data into a classification result, a target satellite node data set is obtained;

[0020] Determine the current optimal feature based on information gain and Gini coefficient;

[0021] Based on the comparison result between the current optimal feature and the optimal feature threshold, the target satellite node data set is divided to obtain a decision tree model.

[0022] In one embodiment, the step of compressing the decision tree model to obtain a target decision tree model includes:

[0023] When the model is compressed into a model slice, a non-leaf node in the decision tree model that is inconsistent with the geographical location range of the slice is used as a target clipping node;

[0024] Delete the target clipping node and all subtree nodes under the target clipping node in the decision tree model to obtain a target decision tree model.

[0025] In one embodiment, the step of compressing the decision tree model to obtain a target decision tree model includes:

[0026] When the model is compressed into a next-hop merge, determining a merge mode;

[0027] When the merging mode is the first mode, if it is determined that the routing decision results of all leaf nodes under the parent node in the decision tree model are consistent, all leaf nodes under the parent node are merged, and the same routing decision results of all leaf nodes are set as the routing decision results of the parent node, and then the target decision tree model is obtained;

[0028] When the merging mode is the second mode, a group of non-sibling leaf nodes with the same routing decision results are merged into the same parent node of their respective parent nodes, and the same routing decision results of the non-sibling leaf nodes are set as the routing decision results of the same parent node, thereby obtaining the target decision tree model.

[0029] In one embodiment, the step of compressing the decision tree model to obtain a target decision tree model includes:

[0030] When the model is compressed into a default route selection, determining the route decision result with the largest number in the decision tree model as the target route decision result;

[0031] Delete the node corresponding to the target routing decision result in the decision tree model to obtain the target decision tree model, and use the target routing decision result as the default route so that when the destination address of the data message does not match the entry in the target decision tree model, the message forwarding is completed based on the default route.

[0032] In one embodiment, the step of compressing the decision tree model to obtain a target decision tree model includes:

[0033] When the model is compressed into classification merging, the Value arrays of all child nodes under the satellite node in the decision tree model are determined, and the maximum value and the sum of the arrays are determined according to the Value arrays of all child nodes;

[0034] Determine a satellite node whose array maximum value is greater than a preset maximum value threshold and whose array sum is greater than a preset array threshold as a target satellite node;

[0035] The left and right subtrees of the target satellite node are deleted, and the routing decision result corresponding to the maximum value of the array is set as the routing decision result of the target satellite node to obtain a target decision tree model.

[0036] In one embodiment, the method further comprises:

[0037] Define a terminal information extension area, wherein the terminal information extension area is centered on the ground terminal and is a circular area;

[0038] When the data message enters the last hop problem range, determining whether the current satellite belongs to the terminal information extension area;

[0039] If the current satellite belongs to the terminal information diffusion area and the satellite does not have the information of the terminal accessing the satellite, the destination terminal orbital plane in the data message is modified to the opposite orbital plane and forwarded.

[0040] In one embodiment, the method further comprises:

[0041] After the data message enters the head node of the satellite network, a satellite routing message header is added to the data message, wherein the satellite routing message header includes the destination terminal orbital plane, the destination terminal longitude, the destination terminal latitude, the next hop direction of the different orbit, and the next hop direction of the same orbit.

[0042] In addition, to achieve the above purpose, the present application also proposes a location-oriented intelligent satellite network routing device, the location-oriented intelligent satellite network routing device comprising:

[0043] A determination module, used to determine a satellite node data set according to a simulation analysis result of a satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, ascending and descending track information, geographic location information of a destination node, and a routing decision result;

[0044] A training module, used for performing model training according to the satellite node data set to obtain a decision tree model;

[0045] A compression module, used for compressing the decision tree model to obtain a target decision tree model;

[0046] The deployment module is used to convert the target decision tree model into a matching action entry suitable for the switch, and deploy the matching action entry to the switch, so that after the data message enters the switch, the message forwarding is completed according to the matching action entry.

[0047] In addition, to achieve the above-mentioned purpose, the present application also proposes a location-oriented intelligent satellite network routing device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the location-oriented intelligent satellite network routing method as described above.

[0048] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the location-oriented intelligent satellite network routing method as described above are implemented.

[0049] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the location-oriented intelligent satellite network routing method as described above.

[0050] One or more technical solutions proposed in this application have at least the following technical effects:

[0051] The present application proposes a location-oriented intelligent satellite network routing method and device, which determines a satellite node data set according to the simulation analysis results of the satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, elevation track information, destination node geographic location information and routing decision results; performs model training according to the satellite node data set to obtain a decision tree model; performs model compression on the decision tree model to obtain a target decision tree model; converts the target decision tree model into a matching action entry suitable for a switch, and deploys the matching action entry to the switch, so that after the data message enters the switch, the matching action entry is deployed to the switch, so that the target decision tree model is converted into a matching action entry suitable for a switch, and the matching action entry is deployed to the switch, so that after the data message enters the switch, the matching action entry is deployed to the switch, and the matching action entry is deployed to the switch, so that ... According to the matching action entries, message forwarding is completed, which solves the technical problem of large link bandwidth occupancy caused by the strong dynamics of satellite network topology. Compared with the existing technology, the present application can simulate the satellite network for a long time on a simulation platform to obtain simulation analysis results, and then train the decision tree model based on the simulation analysis results. Through the decision tree model, a huge global network routing table is no longer required, and the routing problem on the satellite network is converted into a routing problem of nodes within the constellation, avoiding real-time updating of synchronous link status information of the entire network, thereby greatly reducing the occupancy of network bandwidth and saving satellite network bandwidth resources. In addition, the present application also achieves the streamlining of matching entry resources by compressing the decision tree model. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0053] Figure 2 A data set acquisition flow chart provided for the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0054] Figure 3The original data set conversion process provided for the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0055] Figure 4 The decision tree model training process provided for the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0056] Figure 5 A model slicing compression flow chart provided for the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0057] Figure 6 A schematic diagram of aggregation of brother leaf nodes provided in Embodiment 1 of the location-oriented intelligent satellite network routing method of the present application;

[0058] Figure 7 A schematic diagram of aggregation of non-sibling nodes provided in Embodiment 1 of the location-oriented intelligent satellite network routing method of the present application;

[0059] Figure 8 A next hop merging and compression flow chart provided for the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0060] Fig. 9 A schematic diagram of the default route compression provided in the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0061] Fig.10 A compression flow chart of default route selection provided in the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0062] Fig.11 A classification, merging and compression flow chart provided for the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0063] Fig.12 A flow chart of routing decision for message forwarding provided in the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0064] Fig.13 A system description diagram provided for the first embodiment of the location-oriented intelligent satellite network routing method of the present application;

[0065] Fig.14 A flow chart of the second embodiment of the location-oriented intelligent satellite network routing method of the present application is provided;

[0066] Fig.15 This is a schematic diagram of the module structure of a location-oriented intelligent satellite network routing device according to an embodiment of the present application;

[0067] Fig.16The present invention is a schematic diagram of the device structure of the hardware operating environment involved in the location-oriented intelligent satellite network routing method in the embodiment of the present application. DETAILED DESCRIPTION

[0068] It should be understood that the specific embodiments described herein are only used to explain the technical solution of the present application and are not used to limit the present application. Detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0069] The main solution of the embodiment of the present application is: determine a satellite node data set according to the simulation analysis results of the satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, elevation and descending orbit information, destination node geographic location information and routing decision results; perform model training according to the satellite node data set to obtain a decision tree model; perform model compression on the decision tree model to obtain a target decision tree model; convert the target decision tree model into a matching action entry suitable for a switch, and deploy the matching action entry to the switch, so that after the data message enters the switch, the message forwarding is completed according to the matching action entry.

[0070] It can be seen from the above embodiments that the present application solves the technical problem of large link bandwidth occupancy caused by the strong dynamics of satellite network topology. Compared with the prior art, the present application can perform long-term simulation of the satellite network on a simulation platform to obtain simulation analysis results, and then train based on the simulation analysis results to obtain a decision tree model. Through the decision tree model, a huge global network routing table is no longer required, and the routing problem on the satellite network is converted into a routing problem of nodes within the constellation, avoiding real-time updating of synchronous link status information across the entire network, thereby greatly reducing the occupancy of network bandwidth and saving satellite network bandwidth resources. In addition, the present application also achieves the streamlining of matching entry resources by compressing the decision tree model.

[0071] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a location-oriented intelligent satellite network routing device, etc. The following takes location-oriented intelligent satellite network routing as an example to illustrate this embodiment and the following embodiments.

[0072] Based on this, the embodiment of the present application provides a location-oriented intelligent satellite network routing method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the location-oriented intelligent satellite network routing method of the present application.

[0073] In this embodiment, the location-oriented intelligent satellite network routing method includes steps S10 to S40:

[0074] Step S10, determining a satellite node data set according to the simulation analysis result of the satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, ascending and descending track information, geographic location information of the destination node, and routing decision results;

[0075] In the specific implementation, the NS3 simulation software can be expanded to add a satellite constellation operation simulation module, and the simulation analysis results can be obtained after the simulation analysis of the entire satellite network for a certain period of time. Each satellite node is loaded with the routing algorithm (such as OSPF routing) that needs to be learned and referenced, so as to obtain the satellite node data set required for machine learning. The satellite node data set contains the geographical location information of all satellite nodes at a certain moment (i.e., the geographical location information of the forwarding node), whether the satellite is in the ascending orbit plane or the descending orbit plane (i.e., the ascending and descending orbit plane information), the geographical location information of the destination satellite node corresponding to the message at this moment (i.e., the destination node geographical location information), and the routing decision results of the routing algorithm (such as OSPF) that needs to be referenced.

[0076] It should be noted that when designing location-oriented intelligent satellite network routing, the following prerequisites must be followed:

[0077] 1. All satellites know their own location information. This is a known condition in the actual satellite network. Satellites can determine their own geographic location and altitude by receiving global navigation satellite systems (GPS or BeiDou, etc.); they can also measure satellite radio signals or track satellite motion through ground tracking stations, calculate their current location based on known orbital mechanics models, and transmit it back to the satellite.

[0078] 2. The number of network interfaces of satellites is limited. Due to on-board resource limitations, satellites in current satellite networks have only a few link interfaces for communicating with neighboring satellites. Taking Starlink as an example, its satellites are designed with four intersatellite links, which are used to connect the two front and rear satellites in the same orbit, and the two closest satellites in the left and right adjacent orbits. The constellation of this algorithm also follows a similar setting. This patent does not require the number of interfaces to be four, but the satellite network design determines that the number of intersatellite links for each satellite is fixed, and the satellites and intersatellite links in the network are equal, and there is no difference in their roles in the topology. If the intersatellite links in the same orbit are regarded as vertical connections and the intersatellite links in different orbits are regarded as horizontal connections, the entire network can be regarded as a Manhattan-like network.

[0079] 3. There are two isolated communication planes in the satellite network. In the satellite constellation, if the satellite moves toward a higher latitude, we believe that the satellite is in the ascending orbit plane; conversely, if the satellite moves toward a lower latitude, we believe that the satellite is in the descending orbit plane. Due to the relative motion speed of satellites and the constraints of intersatellite link tracking, calibration and other performance, satellites in different orbital planes cannot maintain communication with each other, that is, satellites in the ascending orbit plane cannot communicate with satellites in the descending orbit plane.

[0080] 4. Satellite networks have been divided into regions according to rules. Since satellite networks have global coverage and huge network areas, a regional division method is needed to simplify network management. For example, a regional division method based on logical division is proposed to optimize the communication efficiency and stability of satellite networks in dynamic environments. This method allows multiple satellites to cover the same area at the same time by reasonably dividing the service area, and also enables a single satellite to cover multiple areas, thereby effectively reducing communication interruptions and management complexity caused by satellite switching.

[0081] 5. The destination area of ​​the data message is known. After receiving the data message, the satellite node can obtain the target area from the message (without obtaining accurate location information or knowing the specific satellite to which the target terminal is connected). The area information is converted by the terminal or ground gateway according to the destination address in the data message.

[0082] In a feasible implementation, the step of determining a satellite node data set based on simulation analysis results of a satellite network includes: extending the simulation function on an NS3 simulator to obtain position prediction information of all satellite nodes in the satellite network at different times; determining the connection angle between each satellite node and a neighboring node based on the position prediction information; determining the link status of each satellite node and a neighboring node based on the connection angle; and obtaining a routing decision result of each satellite node for a destination node after synchronizing the link status of the entire network based on the link status; generating a satellite node data set based on the geographic location information of each satellite node, the geographic location information of the destination node, and the routing decision result.

[0083] It should be noted that in the satellite network, since the relative position of the satellites is dynamically changing, it is necessary to simulate and run the corresponding routing protocols based on the accurate simulation of the dynamic topology to simulate the actual operation of the satellite network. By expanding the NS3 simulator, we introduced the SGP4 model to accurately predict the position of the satellite constellation at a specific time. The position information is also used for subsequent routing algorithms. In addition, the distance and connection angle between the satellite and the adjacent satellite can be obtained through simple calculations. This angle can be mainly used to simulate the dynamic connection and disconnection of satellite links between neighboring satellites in adjacent orbits: because the larger the angle between satellites in adjacent orbits, the faster their relative movement speed in the latitude direction. Due to the docking adjustment of laser communication, the speed limit and the possible line of sight obstruction lead to the disconnection of neighbor links. Therefore, based on this angle, we can determine whether the connection between the satellite and the satellite in the neighboring orbit needs to be interrupted at the corresponding moment. After determining the satellite position and the connectivity of neighboring nodes, the routing algorithm module loaded by NS3 can send relevant link status notification messages and run the routing algorithm after the link status of the entire network is synchronized to obtain the routing decision results of each node for all other destination satellites. The satellite's own geographic location information and the geographic location information of the destination satellite are used as inputs for machine learning, and the routing decision results are used as outputs, thereby obtaining a simulation data set (i.e., a satellite node data set).

[0084] In the specific implementation, Figure 2 As shown in the figure, the process of obtaining the satellite node data set is described in detail, as follows:

[0085] Step S101: Initialize the orbital parameters of the satellite constellation. The parameters are determined during constellation planning and design. Based on the orbital parameters, generate a number of nodes of the constellation.

[0086] Step S102: Load the OSPF routing protocol on each node.

[0087] Step S103: Initialize the specific position of each satellite in the constellation according to the orbital parameters and the time when the simulation starts, and confirm the four neighbors of the satellite, two of which are the upper and lower satellites in the same orbit, and the other two neighbors are the neighboring nodes of the two adjacent left and right orbits.

[0088] Step S104: Each node updates its specific position and calculates the angle between itself and its adjacent track neighbors.

[0089] Step S105: Determine whether the angle is greater than a threshold, if so, jump to step S106; otherwise, jump to step S107.

[0090] Step S106: Disconnect the corresponding connection between the self and the neighbor, and jump to S108.

[0091] Step S107: If the connection with the corresponding neighbor has been disconnected, re-establishing the link.

[0092] Step S108: Each node determines whether the connection status with its neighbor has changed, and if so, notifies the neighbor.

[0093] Step S109: After the connection status of the entire network is synchronized, the OSPF routing protocol recalculates the entire network routes of each point and updates the routing forwarding entries.

[0094] Step S110: Each node exports the internal state information of the node, including the geographical location information and track surface information of the node, the geographical location information and track surface information of the target node, and all forwarding next hop information of the routing table.

[0095] Step S111: Update the simulation time to the next time instant and jump to S104.

[0096] In the specific implementation, Figure 3 As shown in the figure, the conversion process of the original satellite node dataset is described in detail as follows:

[0097] Step S201: According to the simulation time list, obtain the first unprocessed time, obtain the geographical location information and orbital plane information of all simulation nodes at this time, and obtain the result as [src_longitude, src_latitude, src_plane].

[0098] Step S202: According to the time, determine whether all simulation node lists have been completely processed. If they have been completely processed, jump to S204; otherwise, jump to S203.

[0099] Step S203: Find the first unprocessed node as the current node. Traverse the routing table of the current node at the current time, convert the destination address to the geographical location information and orbital surface information of the node, then convert the next hop selection of the OSPF route to the outbound port number of this point, and merge the converted data into the result of the previous step. The final result entry content is: [src_longitude, src_latitude, src_plane, dst_longitude, dst_latitude, dst_plane, nexthop].

[0100] Step S204: Determine whether all simulation time lists have been processed. If so, complete all conversions and exit this process; otherwise, jump to S201 to continue execution.

[0101] Step S20, performing model training according to the satellite node data set to obtain a decision tree model;

[0102] It should be noted that the decision tree model is implemented using the CART (Classification and Regression Tree) algorithm or other methods such as random forest. During the experiment, we also conducted relevant tests on random forest, XGBoost, etc. When the amount of training data reaches a certain level, the final recognition accuracy of different decision tree models is not very different.

[0103] In a feasible implementation manner, the step of performing model training according to the satellite node data set to obtain a decision tree model includes: loading the satellite node data set; converting the geographic location information of the node data and the geographic location information of the destination node in the satellite node data set into area numbers, and converting the routing decision results of the node data into classification results to obtain a target satellite node data set; determining the current optimal feature according to information gain and the Gini coefficient; and dividing the target satellite node data set based on the comparison result of the current optimal feature with the optimal feature threshold to obtain a decision tree model.

[0104] It should be noted that for the obtained satellite node data set, the geographical location information of the forwarding node and the destination node is converted into the area number, and the routing decision result (next hop port) is converted into the classification result, which is used as machine learning training data for model training, and finally a decision tree model is formed.

[0105] It should be noted that the decision tree model adopts the CART (Classification and Regression Tree) algorithm, which uses the Gini coefficient as the feature division standard. It only generates two branches each time it divides, and the depth and splitting conditions of the final decision tree model can be controlled by parameters such as the maximum depth of the tree, thereby preventing overfitting and providing favorable conditions for the conversion of the final Match & Action items.

[0106] In the specific implementation, Figure 4 As shown in the figure, the training process of the decision tree model is as follows:

[0107] Step S301: loading a satellite node data set, and preprocessing the satellite node data set to obtain a target satellite node data set;

[0108] Step S302: construct the root node of the decision tree, which is the current node, and put all satellite node data sets into the current node;

[0109] Step S303: Select the optimal feature (such as the current optimal feature at the current moment) according to the information gain and the Gini coefficient, and select the optimal feature to divide the data set in the current node;

[0110] Step S304: Generate two child nodes for the current node according to the threshold of the optimal feature, and assign the divided data sets to the corresponding child nodes respectively, mark the node as a leaf node, and store the final predicted category.

[0111] Step S305: Determine the node purity, the depth of the tree, and whether there are still available features. If all conditions are met, jump to step S307 to exit the construction of the decision tree; otherwise jump to step S306;

[0112] Step S306: traverse the child nodes of the current node, and repeat steps S303 to S305 with the child nodes as the current node;

[0113] Step S307: Output the final decision tree model.

[0114] Step S30, compressing the decision tree model to obtain a target decision tree model;

[0115] It should be noted that model compression includes model slicing, next-hop merging, default route selection and classification merging, among which model slicing, next-hop merging and default route selection are lossless compression, and classification merging is lossy compression. The lossy compression and lossless compression here refer to whether the compressed routing decision results are consistent with the original decision tree model. If the routing decision results in any case are consistent with the original decision tree model, it is considered to be lossless compression, otherwise it is lossy compression.

[0116] In a feasible implementation, the decision tree model is compressed to obtain a target decision tree model, including: when the model is compressed into a model slice, the non-leaf nodes in the decision tree model that are inconsistent with the geographical location range of the slice are used as target cropping nodes; the target cropping node in the decision tree model and all subtree nodes under the target cropping node are deleted to obtain a target decision tree model.

[0117] It should be noted that the model slicing is due to the fact that each satellite only cares about its own management area when forwarding messages, and the entries corresponding to the remaining areas outside the management area cannot be hit. Therefore, in the decision tree model, all entries in the area where the geographic location information of the source node is not the current area of ​​the node and the subsequent period of time (that is, the target clipping node that is inconsistent with the sliced ​​geographic location range and all subtree nodes under the target clipping node) can be clipped to achieve a compression effect.

[0118] In the specific implementation, Figure 5 As shown in the figure, the model slicing compression process of the decision tree model is as follows:

[0119] Step S401: inputting the decision tree, the root node, and the geographical location range and the track plane of the source node slice;

[0120] Step S402: Determine whether the current node is a leaf node. If so, jump to S407 to continue accessing the subsequent node to be accessed; otherwise, jump to S403;

[0121] Step S403: Determine the feature selection of the current node. If the geographical location of the source node intersects with the input slice geographical location range and the track plane is consistent, jump to S405 to continue execution; otherwise, jump to S404;

[0122] Step S404: Delete the node and all the subtree nodes under it, and jump to S407;

[0123] Step S405: store the left child node of the node into a to-be-accessed list;

[0124] Step S406: store the right child node of the node into a to-be-accessed list;

[0125] Step S407: Determine whether the to-be-accessed list is empty, if so, jump to S409; otherwise, jump to S408;

[0126] Step S408: Take the first node in the to-be-visited list as the current node, and jump to step S402;

[0127] Step S409: Return the compressed decision tree.

[0128] In a feasible implementation manner, the decision tree model is compressed to obtain a target decision tree model, and the steps include: when the model is compressed to a next-hop merge, determining a merge mode; when the merge mode is a first mode, if it is determined that the routing decision results of all leaf nodes under the parent node in the decision tree model are consistent, all leaf nodes under the parent node are merged, and the same routing decision results of all leaf nodes are set as the routing decision results of the parent node, and then the target decision tree model is obtained; when the merge mode is a second mode, a group of non-sibling leaf nodes with the same routing decision results are merged into the same parent node of their respective parent nodes, and the same routing decision results of the non-sibling leaf nodes are set as the routing decision results of the same parent node, and then the target decision tree model is obtained.

[0129] It should be noted that there are two modes for next-hop merging, which are divided into Level 1 (first mode) and Level 2 (second mode) according to the scope of aggregation. Among them, Level 1 means that if the parent node itself is not a decision node and the two leaf nodes are both decision nodes in the decision tree model, it is completed. As long as the next hops of the two leaf nodes of the parent node are the same, the corresponding entries can be mentioned to the parent node, thereby merging two or more entries into one entry (aggregation method such as Figure 6 As shown); Level2 aggregates a group of non-sibling decision entries with the same next hop into a super parent node entry, and allows unroutable space to exist between these non-sibling decision entries (aggregation method as shown in Figure 7 This aggregation method expands the decision space of the next hop, but because the expanded range is an area that will not be traversed during actual reasoning (for example, entries that are segmented and set as invalid in the first step of model slicing), it does not affect the actual judgment result and still belongs to lossless compression.

[0130] In the specific implementation, Figure 8 As shown in the figure, the next-hop merging and compression process of the decision tree model is as follows:

[0131] Step S501: input a decision tree, and the root node is used as the current node.

[0132] Step S502: Determine whether the current node is a leaf node, if so, jump to S507; otherwise, jump to S503 to perform a merge determination.

[0133] Step S503: Determine whether the next hop selections of all leaf nodes under the current node are consistent. If yes, jump to S506 for merging; otherwise, jump to S504.

[0134] Step S504: store the left child node of the current node into a to-be-accessed list.

[0135] Step S505: store the right child node of the current node into the to-be-accessed list.

[0136] Step S506: Merge the Value arrays of all leaf nodes of the current node, and set their common next hop selection as the selection of the current node.

[0137] Step S507: Determine whether the to-be-accessed list is empty, if so, jump to S509; otherwise, jump to S508.

[0138] Step S508: Take out the first node in the to-be-visited list, and jump to step S502 to continue to determine whether there is a possibility of merging subsequent nodes.

[0139] Step S509: Return the compressed decision tree.

[0140] In a feasible implementation manner, the step of compressing the decision tree model to obtain a target decision tree model includes: when the model is compressed to a default route selection, determining the routing decision result with the largest number in the decision tree model as the target routing decision result; deleting the node corresponding to the target routing decision result in the decision tree model to obtain a target decision tree model, and using the target routing decision result as the default route, so that when the destination address of the data message does not match the entry in the target decision tree model, message forwarding is completed based on the default route.

[0141] It should be noted that the compression method of default route selection is implemented by using the default route technology of the current mature commercial routing forwarding solution; the default route refers to a routing item pre-configured in the routing table. When the destination address of the data message does not match any existing entry in the routing table, the data message will be sent to the destination specified by the default route. Therefore, we can count the entries pointing to each next hop in the decision tree model (i.e., the statistical routing decision results), and select the route with the largest number of entries as the default route (i.e., the target routing decision result), thereby simplifying the structure of the routing table and reducing the number of matching rules (aggregation methods such as Fig. 9 shown).

[0142] In the specific implementation, Fig.10 As shown, the default routing compression process of the decision tree model is as follows:

[0143] Step S601: Input a decision tree and a root node.

[0144] Step S602: Determine whether the current node is a leaf node, if so, jump to S605; otherwise, jump to S603.

[0145] Step S603: If the current node has a left child node, the left child node is stored in a to-be-accessed list.

[0146] Step S604: If the current node has a right child node, the right child node is stored in a to-be-accessed list.

[0147] Step S605: The next hop decision result of the current leaf node is retrieved, and the statistical value corresponding to the next hop is increased by 1.

[0148] Step S606: Determine whether the to-be-accessed list is empty. If not, jump to S607; otherwise, jump to S608.

[0149] Step S607: Take out the first node in the to-be-accessed list and jump to step S602 to continue the subsequent compression process.

[0150] Step S608: Take out the next hop corresponding to the maximum value in the statistical value list and save it as the root node selection.

[0151] Step S609: re-traverse the decision tree and delete all decision nodes of the next hop.

[0152] Step S610: Return the compressed decision tree.

[0153] In a feasible implementation manner, the step of compressing the decision tree model to obtain a target decision tree model includes: when the model is compressed into classification merging, determining the Value array of all child nodes under the satellite node in the decision tree model, and determining the array maximum value and the array sum according to the Value array of all child nodes; determining the satellite node whose array maximum value is greater than a preset maximum value threshold and whose array sum is greater than a preset array threshold as the target satellite node; deleting the left and right subtrees of the target satellite node, and setting the routing decision result corresponding to the array maximum value as the routing decision result of the target satellite node to obtain the target decision tree model.

[0154] It should be noted that the preset array threshold and the preset maximum value threshold can be set in advance.

[0155] In the specific implementation, Fig.11 As shown, S701-S709 is the subsequent data preparation stage of traversing the binary tree structure, and steps S710-S714 is the pre-order traversal of the binary tree structure for actual compression processing; the classification, merging and compression process of the decision tree model is as follows:

[0156] Step S701: Input a decision tree and a root node.

[0157] Step S702: If the current node is not empty or the queue to be accessed is not empty, jump to S703; otherwise, jump to S710.

[0158] Step S703: If the current node is not empty, jump to S704; otherwise, jump to S705.

[0159] Step S704: store the current node in the queue to be accessed, then set the current node as the left child of the node, then jump to step S703 and continue to traverse the left subtree branch.

[0160] Step S705: Get the latest node in the to-be-visited list as the current node.

[0161] Step S706: Determine whether the current node has a right child node, if so, jump to S707; otherwise, jump to S708.

[0162] Step S707: Set the current node as the right child node, then jump to step S703 and continue to traverse the left subtree structure of the node.

[0163] Step S708: Take out the latest node in the access queue as the current node, and delete the node from the queue.

[0164] Step S709: Take out the Value arrays of all child nodes of the node, add them up and store them in the current node.

[0165] Step S710: taking the root node as the current node.

[0166] Step S711: Get the maximum value in the Value array of the current node and calculate the array sum. If the result of the maximum value / array sum is greater than the threshold, delete the left and right subtrees of the node and set the next hop of the node to the next hop corresponding to the maximum value.

[0167] Step S712: If the node has a left child node, continue traversing the left subtree.

[0168] Step S713: If the node has a right child node, continue traversing the right subtree.

[0169] Step S714: Return.

[0170] It should be noted that classification merging refers to traversing the entire decision tree from top to bottom. Since each node in the decision tree model contains a value array, which stores the count of the corresponding category label under the node, the sum of the value arrays of all leaf nodes under a certain node can be determined. If a certain category accounts for the maximum value and the ratio of the total value array of the node is greater than the set threshold, it can be considered that all leaf nodes of the node can choose the next hop with the largest sample number, thereby achieving the purpose of compressing model entries. It should be noted that this compression method is lossy compression because it will modify a small number of decision behaviors of the model, and the threshold needs to be set to a relatively large value.

[0171] Step S40, converting the target decision tree model into matching action entries suitable for the switch, and deploying the matching action entries to the switch, so that after the data message enters the switch, the message is forwarded according to the matching action entries.

[0172] It should be noted that the compressed decision tree model (i.e., the target decision tree model) is converted into a Match&Action entry (i.e., a matching action entry) suitable for the switch and deployed to the switch. Since the nodes in the decision tree model save the feature division and threshold setting, they can be directly converted into M&A entries. After being deployed to the switch, the message immediately selects the final forwarding next hop (Action) according to the matching (Match) result of the target decision tree model when entering the switch, and completes the final message forwarding.

[0173] In the specific implementation, Fig.12 As shown, the specific description of the routing decision process for message forwarding is as follows:

[0174] Step S801: The satellite node receives a message, parses the relevant message, and obtains a satellite routing message header.

[0175] Step S802: Determine whether the current satellite is in the target area, if yes, jump to step S809; otherwise, jump to S811.

[0176] Step S803: If the orbit of the current satellite can reach the target area, jump to S804; otherwise jump to S805.

[0177] Step S804: among the neighbors on the same track, select the neighbor closest to the target area for forwarding.

[0178] Step S805: According to the location information of the current node and the target location information in the message, the inference input [src_longitude, src_latitude, src_plane, dst_longitude, dst_latitude, dst_plane] is merged to obtain the inference next hop nexthop.

[0179] Step S806: If the off-track next hop in the message is an invalid value, and the inferred next hop is an off-track jump, the inferred next hop is selected as the current next hop, and the off-track next hop in the message is updated to be the inferred next hop.

[0180] Step S807: If the off-track next hop in the message is a valid value, the off-track next hop in the message is selected.

[0181] Step S808: If the same-track next hop in the message is an invalid value, and the inferred next hop is a same-track jump, select the inferred hop as the current next hop, and update the same-track next hop in the message to be the inferred next hop.

[0182] Step S809: If the next hop on the same track in the message is a valid value, the next hop on the same track in the message is selected.

[0183] Step S810: If the link corresponding to the selected next hop has been interrupted, then select the next hop in the opposite direction to the next hop on the same track in the message as the current next hop.

[0184] Step S811: parse the packet and find the actual target address.

[0185] Step S812: Determine whether there is a corresponding route for this node. If so, jump to step S814; otherwise, jump to step S813.

[0186] Step S813: modify Dst_Plane in the satellite routing message header to the opposite orbital plane, clear AlienOrbit_Nexthop and SameOrbit_Nexthop, and jump to step S805 to continue forwarding.

[0187] Step S814: Perform normal routing search according to the target node address, find the next hop, strip the satellite routing message header, and send the message from the port corresponding to the next hop.

[0188] Step S815: The forwarding process ends.

[0189] In the specific implementation, Fig.13 The location-oriented intelligent satellite network routing system description diagram shown in the figure can obtain the optimal next-hop routing result of the traditional routing algorithm through offline simulation based on the regularity and regularity of the satellite network. The data is combined with the geographical location information of the node at the corresponding time to form a training data set; the data set is studied by the machine learning algorithm to extract the routing forwarding model based on the geographical location information to provide guidance for the next-hop routing decision of the message; finally, the routing forwarding model will be converted into a Match-Action entry suitable for the router forwarding table, and finally deployed on the switch to realize the forwarding of data messages.

[0190] This embodiment can simulate the satellite network for a long time on the simulation platform to obtain simulation analysis results, and then train the decision tree model based on the simulation analysis results. Through the decision tree model, a huge global network routing table is no longer needed, and the routing problem on the satellite network is converted into a routing problem of nodes within the constellation, avoiding the real-time update of synchronous link status information of the entire network, thereby greatly reducing the occupancy of network bandwidth and saving satellite network bandwidth resources. In addition, this embodiment also achieves the streamlining of matching entry resources by compressing the decision tree model.

[0191] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Fig.14, after step S40, steps S50 to S70 are also included:

[0192] Step S50, defining a terminal information extension area, wherein the terminal information extension area is centered on the ground terminal and is a circular area;

[0193] It should be noted that the difficulties in deploying the algorithm are as follows:

[0194] 1. After receiving the data message from the ground terminal, the satellite node in the LEO satellite network can only obtain the target location and the target management area from the message, but does not know the specific satellite to which the terminal is connected. Therefore, it is necessary to solve the problem of forwarding the message to the final destination satellite, which we call the last hop problem.

[0195] 2. The current satellite does not know whether the current orbital plane of the terminal access satellite is in the ascending orbit plane or the descending orbit plane. This will cause the satellite receiving the ground terminal data message to not know to which target area to forward the message, and therefore cannot make a routing decision (for the ascending orbit plane or the descending orbit plane) (that is, the message forwarding problem).

[0196] Step S60, when the data message enters the last hop problem range, determining whether the current satellite belongs to the terminal information extension area;

[0197] It should be noted that in order to solve the problem of the last hop, we define a terminal information diffusion area: it is a circular area with the terminal as the center and a predefined threshold as the radius. After receiving the access request from the terminal, the access satellite needs to diffuse the terminal access information to all neighboring satellites. The diffusion domain is the terminal information diffusion area. After receiving the broadcast terminal access information, the neighboring satellite first saves the corresponding terminal information and the corresponding access satellite, and then determines whether the plane distance between itself and the terminal is less than the predefined threshold. If so, it needs to continue to diffuse to the neighbors.

[0198] Step S70: If the current satellite belongs to the terminal information diffusion area and the satellite does not have the information of the terminal accessing the satellite, the destination terminal orbital plane in the data message is modified to the opposite orbital plane and forwarded.

[0199] It should be noted that in the data plane message forwarding process, after the message is forwarded to the management area (or adjacent management area), it enters the terminal information diffusion area. At this time, the satellite performs deterministic path forwarding according to the previously saved access satellite, and the access satellite of the destination terminal finally forwards the message to the destination terminal. When the source terminal sends the data message to its access satellite, the access satellite cannot obtain the orbital plane where the destination terminal is located. At present, we assume that the orbital plane where the destination terminal is located is the same as the current access satellite through random selection, and forward it accordingly. When the data message enters the scope of the last hop problem, if the current satellite belongs to the terminal information diffusion area (that is, its distance from the terminal location is less than the predefined threshold), but the satellite does not have the information of the terminal access satellite, it is judged that the actual terminal is selected in the opposite orbital plane. At this time, the satellite needs to modify the data message, change the orbital plane of the destination terminal to the opposite orbital plane, and continue to forward. Obviously, if the access satellite of the destination terminal is not in the same orbital plane as the access satellite of the source terminal, the entire forwarding process will have one more forwarding process than the normal one, and the end-to-end distance will not be optimal. In order to optimize the waste of transmission paths caused by the dilemma, after receiving the message, the destination terminal needs to encode the orbital plane selected by itself in the message and respond to the source terminal, so that subsequent messages can be forwarded according to the correct orbital plane.

[0200] In a feasible implementation, the method further includes: after the data message enters the head node of the satellite network, adding a satellite routing message header to the data message, wherein the satellite routing message header includes the destination terminal orbital plane, the destination terminal longitude, the destination terminal latitude, the next hop direction in a different orbit, and the next hop direction in the same orbit.

[0201] It should be noted that there is a difficult problem when deploying the decision tree model: the current satellite does not know whether the current orbital plane of the terminal accessing the satellite is in the ascending orbit plane or the descending orbit plane. This will cause the satellite receiving the data message from the ground terminal to not know to which target area to forward the message, and therefore cannot make a routing decision (for the ascending orbit plane or the descending orbit plane).

[0202] It should be noted that in order to solve the problem of message forwarding, we need to add a layer of satellite routing message header to the data message after it enters the first node of the LEO satellite network. It mainly contains the following information: destination terminal orbital plane, destination terminal longitude, destination terminal latitude, next hop direction in different orbit, and next hop direction in the same orbit. The message structure is shown in the following table:

[0203] Dst_Long_No Dst_Lati_No Dst_Plane AlienOrbit_Nexthop SameOrbit_Nexthop Packet

[0204] The fields are explained as follows:

[0205] Dst_Long_No: 8 bits wide, the number of the target area in the longitude direction;

[0206] Dst_Lati_No: 8 bits wide, the latitude number of the target area, the combination of Dst_Long_No and Dst_Lati_No forms a globally unique identifier for the target area;

[0207] Dst_Plane: 8 bits wide, the orbital plane to which the target gateway is connected;

[0208] AlienOrbit_Nexthop: 8 bits wide, the direction of the alternate orbit selected for this message transmission is inferred by the intelligent model. During the entire satellite network transmission process, only one inference is performed and saved in the message. Subsequent forwarding relies on this result to avoid forwarding loops.

[0209] SameOrbit_Nextho: 8 bits wide, the same-orbit direction selected for this message transmission is inferred by the intelligent model. During the entire satellite network transmission process, only one inference is performed and saved in the message. Subsequent forwarding relies on this result to avoid forwarding loops;

[0210] Packet: original data message content.

[0211] This application also provides a location-oriented intelligent satellite network routing device, please refer to Fig.15 , the location-oriented intelligent satellite network routing device comprises:

[0212] A determination module 10 is used to determine a satellite node data set according to the simulation analysis result of the satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, ascending and descending track information, geographic location information of the destination node, and routing decision results;

[0213] A training module 20, configured to perform model training according to the satellite node data set to obtain a decision tree model;

[0214] A compression module 30, used for compressing the decision tree model to obtain a target decision tree model;

[0215] The deployment module 40 is used to convert the target decision tree model into matching action entries suitable for the switch, and deploy the matching action entries to the switch, so that after the data message enters the switch, the message is forwarded according to the matching action entries.

[0216] Reference below Fig.16, which shows a schematic diagram of the structure of a location-oriented intelligent satellite network routing device suitable for implementing the embodiment of the present application. The location-oriented intelligent satellite network routing device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc.

[0217] like Fig.16 As shown, the intelligent satellite network routing device based on location guidance may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the intelligent satellite network routing device based on location guidance are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 may allow the location-based intelligent satellite network routing device to communicate with other devices wirelessly or by wire to exchange data.

[0218] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the location-oriented intelligent satellite network routing method in the above-mentioned embodiment. The present application also provides a computer program product, including a computer program, which implements the steps of the location-oriented intelligent satellite network routing method as described above when executed by a processor.

[0219] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A location-oriented intelligent satellite network routing method, characterized in that: The method comprises: Determine a satellite node data set according to the simulation analysis result of the satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, ascending and descending track surface information, geographic location information of a destination node, and a routing decision result; Performing model training according to the satellite node data set to obtain a decision tree model; Performing model compression on the decision tree model to obtain a target decision tree model; The target decision tree model is converted into matching action entries suitable for the switch, and the matching action entries are deployed in the switch, so that after the data message enters the switch, the message forwarding is completed according to the matching action entries.

2. The method according to claim 1, characterized in that The step of determining the satellite node data set according to the simulation analysis result of the satellite network includes: Expanding the simulation function on the NS3 simulator to obtain the position prediction information of all satellite nodes in the satellite network at different times; Determine the angle between each satellite node and its neighboring node according to the position prediction information; Determine the link status of each satellite node and the neighboring node according to the connection angle; After synchronizing the link status of the entire network according to the link status, obtaining a routing decision result of each satellite node for the destination node; A satellite node dataset is generated based on the geographic location information of each satellite node, the geographic location information of the destination node, and the routing decision results.

3. The method according to claim 1, characterized in that The step of performing model training according to the satellite node data set to obtain a decision tree model comprises: Load satellite node dataset; After converting the geographical location information of the node data in the satellite node data set and the geographical location information of the destination node into an area number, and converting the routing decision result of the node data into a classification result, a target satellite node data set is obtained; Determine the current optimal feature based on information gain and Gini coefficient; Based on the comparison result between the current optimal feature and the optimal feature threshold, the target satellite node data set is divided to obtain a decision tree model.

4. The method according to claim 1, characterized in that The step of compressing the decision tree model to obtain a target decision tree model includes: When the model is compressed into a model slice, a non-leaf node in the decision tree model that is inconsistent with the geographical location range of the slice is used as a target clipping node; Delete the target clipping node and all subtree nodes under the target clipping node in the decision tree model to obtain a target decision tree model.

5. The method according to claim 1, characterized in that The step of compressing the decision tree model to obtain a target decision tree model includes: When the model is compressed into a next-hop merge, determining a merge mode; When the merging mode is the first mode, if it is determined that the routing decision results of all leaf nodes under the parent node in the decision tree model are consistent, all leaf nodes under the parent node are merged, and the same routing decision results of all leaf nodes are set as the routing decision results of the parent node, and then the target decision tree model is obtained; When the merging mode is the second mode, a group of non-sibling leaf nodes with the same routing decision results are merged into the same parent node of their respective parent nodes, and the same routing decision results of the non-sibling leaf nodes are set as the routing decision results of the same parent node, thereby obtaining the target decision tree model.

6. The method according to claim 1, characterized in that The step of compressing the decision tree model to obtain a target decision tree model comprises: When the model is compressed into a default route selection, determining the route decision result with the largest number in the decision tree model as the target route decision result; Delete the node corresponding to the target routing decision result in the decision tree model to obtain the target decision tree model, and use the target routing decision result as the default route so that when the destination address of the data message does not match the entry in the target decision tree model, the message forwarding is completed based on the default route.

7. The method according to claim 1, characterized in that The step of compressing the decision tree model to obtain a target decision tree model comprises: When the model is compressed into classification merging, the Value arrays of all child nodes under the satellite node in the decision tree model are determined, and the maximum value and the sum of the arrays are determined according to the Value arrays of all child nodes; Determine a satellite node whose array maximum value is greater than a preset maximum value threshold and whose array sum is greater than a preset array threshold as a target satellite node; The left and right subtrees of the target satellite node are deleted, and the routing decision result corresponding to the maximum value of the array is set as the routing decision result of the target satellite node to obtain a target decision tree model.

8. The method according to claim 1, characterized in that The method further comprises: Define a terminal information extension area, wherein the terminal information extension area is centered on the ground terminal and is a circular area; When the data message enters the last hop problem range, determining whether the current satellite belongs to the terminal information extension area; If the current satellite belongs to the terminal information diffusion area and the satellite does not have the information of the terminal accessing the satellite, the destination terminal orbital plane in the data message is modified to the opposite orbital plane and forwarded.

9. The method according to claim 1, characterized in that The method further comprises: After the data message enters the head node of the satellite network, a satellite routing message header is added to the data message, wherein the satellite routing message header includes the destination terminal orbital plane, the destination terminal longitude, the destination terminal latitude, the next hop direction of the different orbit, and the next hop direction of the same orbit.

10. A location-oriented intelligent satellite network routing device, characterized in that: The device comprises: A determination module, used to determine a satellite node data set according to a simulation analysis result of a satellite network, wherein the satellite node data set includes node data of all satellite nodes in the satellite network, and the node data includes geographic location information, ascending and descending track information, geographic location information of a destination node, and a routing decision result; A training module, used for performing model training according to the satellite node data set to obtain a decision tree model; A compression module, used to compress the decision tree model to obtain a target decision tree model; The deployment module is used to convert the target decision tree model into a matching action entry suitable for the switch, and deploy the matching action entry to the switch, so that after the data message enters the switch, the message forwarding is completed according to the matching action entry.

Images