Satellite computing power routing method

Abstract

The application discloses a satellite computing power routing method, which comprises the following steps: step 1, a LEO satellite encodes computing power state information by using a table lookup method and reports the computing power state information to a GEO satellite; step 2, the GEO satellite generates a computing power state matrix and a connectivity state matrix of the LEO network according to the received computing power state information; step 3, a ground station estimates the computing power demand of a task and sends a task instruction to be executed by the LEO network and the computing power demand estimation value of the task to the GEO satellite; step 4, the GEO satellite prunes the computing power state matrix in combination with the computing power demand estimation value sent by the ground station; step 5, a scheduling strategy of the LEO network is generated by using an internal and external bidirectional distribution method and is sent to the LEO network; and step 6, the LEO satellite completes task data distribution according to the computing power scheduling strategy sent by the GEO satellite. The application provides a strategy for multi-satellite cooperative processing and dynamic resource scheduling under the condition that the single-satellite capability is limited, and provides a solution for reducing the processing time delay of on-satellite data.

Description

Technical Field

[0001] This invention belongs to the field of satellite communication technology, and specifically relates to a satellite computing power routing method. Background Technology

[0002] With the deployment of mega-constellations such as Starlink, OneWeb, and State Grid, and the planning of numerous other mega-constellations, satellite orbits have become a scarce resource that countries are vying for. To alleviate the pressure on orbital resources, multi-payload onboard integration has become a future development trend. The improvement in satellite performance, the diversification of payloads, and large-scale constellations will inevitably lead to an exponential increase in onboard data. my country's existing "space-based sensing and ground-based computing" operation mode is no longer sufficient to meet the demand for massive, low-latency on-orbit information transmission.

[0003] On the other hand, the dramatic increase in the number of satellites has also created enormous computing power resources for numerous onboard processors. If onboard computing resources can be integrated and critical information can be extracted through multi-satellite collaboration, it would reduce the amount of data transmitted downlink, trading computation for bandwidth, and simultaneously improve service timeliness. This is also an effective means of solving high latency and downlink bottlenecks. Furthermore, in local hotspot areas, there may be situations where multiple satellites simultaneously lack sufficient computing power. In such cases, satellite nodes located at the center of the hotspot area cannot allocate computing power to surrounding nodes due to insufficient computing power, inevitably further increasing task response latency. Summary of the Invention

[0004] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a satellite computing power scheduling method, which provides a solution to the problems of limited single-satellite resources and untimely service response.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A satellite computing power routing method specifically includes the following steps:

[0007] Step 1: The LEO satellite uses a lookup table to encode the computing power status information and reports the computing power status information to the GEO satellite;

[0008] Step 2: Based on the received computing power status information, the GEO satellite generates the computing power status matrix and connectivity status matrix of the LEO network.

[0009] Step 3: The ground station estimates the computing power requirements of the mission and sends the LEO network instructions to be executed and the estimated computing power requirements of the mission to the GEO satellite.

[0010] Step 4: The GEO satellite trims the computing power state matrix based on the computing power demand estimate sent by the ground station;

[0011] Step 5: Based on the pruned computing power state matrix and connectivity state matrix, the GEO satellite generates the scheduling strategy for the LEO network using a bidirectional internal and external distribution method, and then distributes it to the LEO network.

[0012] Step 6: The LEO satellite distributes the task data according to the computing power scheduling strategy issued by the GEO satellite.

[0013] Furthermore, step 1 includes the following sub-steps:

[0014] Step 11, the LEO node learns its own computing power status information; the computing power status information includes: logical computing power, parallel computing power, intelligent computing power, storage capacity, per-hop link identification communication capacity, and the current utilization rate of the above capabilities;

[0015] Step 12: Normalize the logical computing power, parallel computing power, intelligent computing power, storage capacity, and per-hop link identifier communication capability respectively;

[0016] Step 13: The normalized capability values ​​from Step 12 and the current utilization rate of each capability obtained in Step 11 are encoded by looking up a table.

[0017] Step 14: The LEO satellite reports its computing power status information to the GEO satellite.

[0018] Furthermore, in step 2,

[0019] LEO's computing power state matrix in, CL i,j CP i,j CA i,j M i,j They represent LEO satellite V i,j Current logical computing capability, parallel computing capability, intelligent computing capability, and storage capability, i.e., node capability × (1 - current utilization rate), 1≤i≤m, 1≤j≤n, where m and n represent the number of orbits and the number of satellites in the orbits, respectively;

[0020] LEO's connected state matrix in, LF i,j LB i,j LL i,j LR i,j These are LEO satellites V i,j The status values ​​of the four inter-satellite links.

[0021] Furthermore, step 4 includes the following sub-steps:

[0022] Step 41: The GEO satellite divides the current mission into a series of sub-tasks and estimates the computing power and storage space X required to process the sub-tasks based on the size of the sub-task data and the size of the mission data. min ,in, CL min CP min CA min M min These represent the logical computing capability, parallel computing capability, intelligent computing capability, and storage capability required for subtask processing, respectively.

[0023] Step 42: Remove the elements X in the computing power state matrix whose computing power is less than the computing power required for subtask processing. i,j Set to 0, that is

[0024] Furthermore, step 5 includes the following sub-steps:

[0025] Step 51: The GEO satellite generates an undirected graph G based on the computing power state matrix and the connected state matrix. Each node G in the undirected graph G... i,j Corresponding satellite node V i,j Node G i,j The connectivity with surrounding nodes corresponds to L in the connectivity state matrix. i,j ,in, LF i,j LB i,j LL i,j LR i,j These are LEO satellites V i,j The status values ​​of the four inter-satellite links; Node G i,j The computing power state corresponds to the computing power state matrix in CL i,j CP i,j CA i,j M i,j They represent LEO satellite V i,j Current logical computing capabilities, parallel computing capabilities, intelligent computing capabilities, and storage capabilities;

[0026] Step 52: The GEO satellite marks the hot spot region Ω1 in the undirected graph G, and the hot spot region Ω1 contains nodes G. i,j Its corresponding LEO satellite node V i,j Satellite nodes that submit computing power requests to GEO;

[0027] Step 53: In the undirected graph G, the GEO satellite selects the unloading region Ω2 by expanding outward from the center node of the hot spot region Ω1 in the order of 1 hop, 2 hop, ..., so that the total computing power of the nodes in the unloading region Ω2 is equal to β*CT, where β≥1.

[0028] Step 54: The central node G of the hotspot region Ω1 m,n The target node is found using the first cost function shown in the following formula, with the aim of minimizing the value of the first cost function, and thus G. m,n , destination node, and G m,n Information about the intermediate nodes that need to be traversed to reach the destination node, i.e. <G m,n Intermediate node, destination node >, denoted as computing power scheduling strategy S m,n Then delete node G in Ω1. m,n ;

[0029] Cost = αT comp,l +(1-α)*(T trans,m +T comp,m )

[0030] Among them, T comp,l For the latency calculated locally, T trans,m For the latency of task transmission, T comp,m The computation delay for the destination node is calculated, assuming only one destination node is selected each time.

[0031] α = Task1 / Task m,n 1-α=Task2 / Task m,n , 0≤α≤1; Task m,n For satellite node V m,n The amount of data to be executed in Task1 is V. m,n The actual amount of task data executed locally, while Task2 represents the amount of task data executed on the destination node.

[0032] T comp,l =max{Task 1,β / CL i,j Task 1,γ / CP i,j Task 1,δ / CA i,j Task 1,β Task 1,γ ,

[0033] Task 1,δ These represent the logical computation, parallel computation, and intelligent computation parts of Task1, respectively. 1,β / CL i,j Task 1,γ / CP i,j Task 1,δ / CA i,j These represent the time required for logical computation, parallel computation, and intelligent computation, respectively.

[0034] T comp,m =max{Task 2,β / CL i,j Task 2,γ / CP i,j Task 2,δ / CA i,j Task 2,β Task 2,γ ,

[0035] Task 2,δ These represent the logical computation, parallel computation, and intelligent computation components of Task2, respectively. 2,β / CL i,j Task 2,γ / CP i,j Task 2,δ / CA i,j These represent the time required for logical computation, parallel computation, and intelligent computation, respectively.

[0036] That is, the computation delay of the destination node is equal to the sum of the link delays between all nodes when unloading data, h represents the number of hops required from the data transmission source node to the destination node, and L is the connectivity value when passing through h hop nodes.

[0037] Step 55: Mark the outer nodes in the hotspot region Ω1 as G. pq The second cost function is used to find the target node among the peripheral nodes of the unloading region Ω2, with the aim of minimizing the value of the second cost function and thus G. pq , destination node, and G pq Information about the intermediate nodes that need to be traversed to reach the destination node, i.e. <G pq Intermediate node, destination node >, denoted as computing power scheduling strategy S pq Then delete node G in the unloading region Ω1. pq :

[0038] Cost = T trans,m +T comp,m

[0039] Among them, T trans,m For the latency of task transmission, T comp,m The computation delay at the destination node;

[0040] Step 56: Repeat steps 54 and 55 until set Ω1 is empty, then GEO broadcasts all computing power scheduling policies to LEO.

[0041] Furthermore, in step 53, β is taken as 1.2 or 1.5.

[0042] The advantages of this invention compared to the prior art are:

[0043] (1) Unlike most current research on computing power networks, which focuses on ground application scenarios, this invention presents a space-based application architecture that combines high and low orbits and designs an encoding method for LEO node computing power announcements by looking up a table, which effectively reduces the data volume of computing power announcement messages and provides technical support for the application of computing power networks in large-scale space-based constellations.

[0044] (2) In view of the current characteristics of limited computing power resources of single satellite and strong correlation between service response time and computing power, this invention designs a satellite computing power scheduling method that combines high and low orbits, realizes dynamic allocation of resources and multi-satellite collaborative processing of LEO network, improves the resource utilization of LEO network and reduces service response latency.

[0045] (3) In view of the congestion in hot spots, this invention designs a method for simultaneous internal and external bidirectional computing power scheduling, which provides technical support for the application of computing power networks in hot spots.

[0046] In summary, the method of this invention improves the resource utilization and overall efficiency of the LEO network through methods such as computing power notification, multi-satellite collaboration, and optimized task allocation, thereby reducing the processing latency of on-satellite data. Attached Figure Description

[0047] Figure 1 This is a flowchart of the present invention;

[0048] Figure 2 This is a schematic diagram of the computing power status information format;

[0049] Figure 3 This is a schematic diagram of a bidirectional distribution scheduling strategy. Detailed Implementation

[0050] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings.

[0051] like Figure 1 As shown, the satellite computing power routing method combining high and low orbits provided by this invention has the following implementation steps:

[0052] Step 1: The LEO satellite uses a lookup table to encode the computing power status information and reports the computing power status information to the GEO satellite.

[0053] Step 1 includes the following sub-steps:

[0054] Step 11, the LEO node learns its own computing power status information; the computing power status information includes: logical computing power, parallel computing power, intelligent computing power, storage capacity (i.e., storage capacity), per-hop link identification communication capability, and the current utilization rate of the above capabilities, with the current utilization rate ranging from 0% to 100%.

[0055] Step 12: Normalize the logical computing power, parallel computing power, intelligent computing power, storage capacity, and per-hop link identifier communication capability. The normalization method is shown in the following formula:

[0056] Where γ is the actual capability value, and α and β are the minimum and maximum values ​​in the pre-agreed range of values, respectively.

[0057] Step 13: The normalized values ​​of each capability after step 12, and the current utilization rate of each capability obtained in step 11, are encoded by looking up a table.

[0058] Specifically, the correspondence between capability values ​​and codes is shown in Table 1. The correspondence between current utilization rate and codes is shown in Table 2.

[0059] Table 1. Correspondence between Capabilities and Codings

[0060]

[0061]

[0062] Table 2 Current Usage Rate and Correspondence with Code

[0063] Serial Number Current usage rate coding Serial Number Current usage rate coding 1 0％～11％ 000 5 48％～59％ 100 2 12％～23％ 001 6 60％～71％ 101 3 24％～35％ 010 7 72％～83％ 110 4 36％～47％ 011 8 84％～100％ 111

[0064] Step 14: The LEO satellite reports its computing power status information to the GEO satellite, such as... Figure 2 As shown, its message format is as follows:

[0065] (1) Node ID: 16 bits. A unique node identifier across the entire network;

[0066] (2) Logical computing power: 7 bits in total. Of these, logical computing power capability is 4 bits long; current utilization rate is 3 bits long.

[0067] (3) Parallel computing power: 7 bits in total. Parallel computing power capability is 4 bits long; current utilization rate is 3 bits long.

[0068] (4) Intelligent computing power: 7 bits in total. Among them, intelligent computing power capability is 4 bits in length; current utilization rate is 3 bits in length;

[0069] (5) Storage capacity: 7 bits in total. Of which, storage capacity is 4 bits long; current usage rate is 3 bits long.

[0070] (6) Communication capability: 28 bits in total. Each satellite has 4 inter-satellite links (two in the same orbit and two in different orbits). Each link hop identifies two parameters: communication capability and current utilization rate. The communication capability is 4 bits, in Mbps. The current utilization rate is 3 bits in length.

[0071] (7) Check bit: 8 bits;

[0072] In this embodiment, the ID number of a satellite node is 10; logical computing power is 2 TOPS, currently used at 20%; parallel computing power is 1 TFLOPS, currently used at 30%; intelligent computing power is 0 GFLOPS, currently used at 0%; storage capacity is 2 TByte, currently used at 1%; communication capabilities are: Link 1 (2000 Mbps, currently used at 15%), Link 2 (2000 Mbps, currently used at 10%), Link 3 (2000 Mbps, currently used at 70%), and Link 4 (2000 Mbps, currently used at 50%). It is assumed that the values ​​of logical computing power, parallel computing power, and intelligent computing power range from 1 TOPS to 100 TOPS, 1 TFLOPS to 100 TFLOPS, and 1 TFLOPS to 100 TFLOPS, respectively; the storage capacity ranges from 1 GBytes to 10 TBytes; and the inter-satellite link transmission rate ranges from 100 Mbps to 10 Gbps. The computing power notification message for this satellite node is shown in Table 3:

[0073] Table 3. Computing power notification messages of satellite nodes

[0074]

[0075] Step 2: Based on the received computing power status information, the GEO satellite generates the computing power status matrix and connectivity status matrix of the LEO network.

[0076] Specifically, the LEO computing power state matrix in, CL i,j CP i,j CA i,j M i,j They represent LEO satellite V i,j Current logical computing capability, parallel computing capability, intelligent computing capability, and storage capability, i.e., node capability × (1 - current utilization rate), 1≤i≤m, 1≤j≤n, where m and n represent the number of orbits and the number of satellites in the orbits, respectively;

[0077] Specifically, the connected state matrix of LEO in, LF i,j LB i,j LL i,j LR i,j These are LEO satellites V i,j The status values ​​of the four inter-satellite links.

[0078] Step 3: The ground station estimates the computing power requirements of the mission and sends the LEO network instructions to be executed and the estimated computing power requirements of the mission to the GEO satellite.

[0079] Specifically, firstly, the ground station establishes a computing power requirement sample database through extensive simulations and field measurements. This database records the computing power requirements of commonly used tasks under different task types, data types, data volumes, and algorithm complexities. Then, the ground station constructs a three-dimensional scalar based on the task type, data type, and algorithm complexity of the current task. It then uses Euclidean distance calculations within the computing power requirement sample database to find the task item that best matches the current task. Finally, based on the size of the computing power requirement in the database, the size of the current task's data volume, and the number of items in the database, the computing power requirement (CT) for the current task is estimated.

[0080] In this embodiment, the current task is to detect ships in a certain sea area with a swath width of 1km × 1km, using the YOLO algorithm. The computational power requirement estimation method for the current task is as follows:

[0081] First, the ground station uses parameters such as task type (object detection), data type (optical image), and algorithm complexity (complexity of the YOLO algorithm) to find the task with the smallest Euclidean distance between the current task and the sample database, and marks it as a Task. i Then, the computing power requirement of the current task is estimated by the ratio of the data volume of the two, i.e., CT = data volume of the current task / Task. i Data volume × Task i The computing power requirements.

[0082] Step 4: The GEO satellite trims the computing power state matrix based on the computing power demand estimate sent by the ground station.

[0083] Step 4 includes the following sub-steps:

[0084] Step 41: The GEO satellite divides the current mission into a series of sub-tasks and estimates the computing power and storage space X required to process the sub-tasks based on the size of the sub-task data and the size of the mission data. min ,in, CL min CP min CA min M minThese represent the logical computing capabilities, parallel computing capabilities, intelligent computing capabilities, and storage capabilities required for subtask processing, respectively.

[0085] Step 42: Remove the elements X in the computing power state matrix whose computing power is less than the computing power required for subtask processing. i,j Set to 0, that is

[0086] In this embodiment, the computational and storage requirements for target detection in optical images are CT. If the image is divided into four sub-images for target detection, then the computational and storage requirements for each sub-image are approximately 1 / 4 of the total task, i.e., X. min =CT / 4; then, in the LEO network, as long as any element is less than X... min If the corresponding element in the state matrix is ​​not found, then the LEO satellite does not meet the minimum requirements for subtask execution. Therefore, the element corresponding to the LEO satellite is set to 0 in the state matrix. This satellite will no longer be considered during subsequent computing power scheduling, thus reducing the amount of computation required.

[0087] Step 5: Based on the pruned computing power state matrix and connectivity state matrix, the GEO satellite generates the LEO network's scheduling strategy using a bidirectional internal and external distribution method, and then distributes it to the LEO network, such as... Figure 3 As shown.

[0088] Step 5 includes the following sub-steps:

[0089] Step 51: The GEO satellite generates an undirected graph G based on the computing power state matrix and the connected state matrix. Each node G in the undirected graph G... i,j Corresponding satellite node V i,j Node G i,j The connectivity with surrounding nodes corresponds to L in the connectivity state matrix. i,j ,in, LF i,j LB i,j LL i,j LR i,j These are LEO satellites V i,j The status values ​​of the four inter-satellite links; Node G i,j The computing power state corresponds to the computing power state matrix in CL i,j CP i,j CA i,j M i,j They represent LEO satellite V i,j Current logical computing capabilities, parallel computing capabilities, intelligent computing capabilities, and storage capabilities.

[0090] Step 52: The GEO satellite marks the hot spot region Ω1 in the undirected graph G, and the hot spot region Ω1 contains nodes G. i,j Its corresponding LEO satellite node V i,j Satellite nodes that submit computing power requests to GEO;

[0091] Step 53: In the undirected graph G, the GEO satellite selects the offloading region Ω2 by expanding outward from the center node of the hot spot region Ω1 in the order of 1 hop, 2 hop, ..., so that the total computing power of the nodes in the offloading region Ω2 is equal to β*CT, where β≥1, and typical values ​​can be 1.2 or 1.5.

[0092] Step 54: The central node G of the hotspot region Ω1 m,n The target node is found using the first cost function shown in the following formula, with the aim of minimizing the value of the first cost function, and thus G. m,n , destination node, and G m,n Information about the intermediate nodes that need to be traversed to reach the destination node, i.e. <G m,n Intermediate node, destination node >, denoted as computing power scheduling strategy S m,n Then delete node G in Ω1. m,n ;

[0093] Cost = αT comp,l +(1-α)*(T trans,m +T comp,m )

[0094] Among them, T comp,l For the latency calculated locally, T trans,m For the latency of task transmission, T comp,m The computation delay for the destination node is calculated, assuming only one destination node is selected each time.

[0095] α = Task1 / Task m,n 1-α=Task2 / Task m,n , 0≤α≤1; Task m,n For satellite node V m,n The amount of data to be executed in Task1 is V. m,n The actual amount of task data executed locally, while Task2 represents the amount of task data executed on the destination node.

[0096] T comp,l =max{Task 1,β / CL i,j Task 1,γ / CP i,j Task 1,δ / CA i,j Task 1,β Task1,γ ,

[0097] Task 1,δ These represent the logical computation, parallel computation, and intelligent computation parts of Task1, respectively. 1,β / CL i,j Task 1,γ / CP i,j Task 1,δ / CA i,j These represent the time required for logical computation, parallel computation, and intelligent computation, respectively.

[0098] T comp,m =max{Task 2,β / CL i,j Task 2,γ / CP i,j Task 2,δ / CA i,j Task 2,β Task 2,γ ,

[0099] Task 2,δ These represent the logical computation, parallel computation, and intelligent computation components of Task2, respectively. 2,β / CL i,j Task 2,γ / CP i,j Task 2,δ / CA i,j These represent the time required for logical computation, parallel computation, and intelligent computation, respectively.

[0100] That is, the computational delay of the destination node is equal to the sum of the link delays between all nodes when unloading data, h represents the number of hops required from the data transmission source node to the destination node, and L is the connectivity value when passing through h hop nodes (obtained from the connectivity matrix);

[0101] Step 55: Mark the outer nodes of hotspot region Ω1 as G. pq The second cost function is used to find the target node among the peripheral nodes of the unloading region Ω2, with the aim of minimizing the value of the second cost function and thus G. pq , destination node, and G pq Information about the intermediate nodes that need to be traversed to reach the destination node, i.e. <G pq Intermediate node, destination node >, denoted as computing power scheduling strategy S pq Then delete node G in the hotspot region Ω1. pq :

[0102] Cost = T trans,m +Tcomp,m

[0103] Among them, T trans,m For the latency of task transmission, T comp,m The computation delay of the destination node.

[0104] Step 56: Repeat steps 54 and 55 until set Ω1 is empty, then GEO broadcasts all computing power scheduling policies to LEO.

[0105] In step 5, due to the presence of multiple satellites with insufficient computing power in local hotspot areas, nodes within these areas cannot offload tasks to surrounding satellites because the surrounding satellites also lack sufficient computing power. They must execute the tasks themselves, and since individual satellite resources are limited, this results in long processing delays. For this reason, this invention employs a bidirectional scheduling strategy of internal and external distribution. For example, when satellite nodes 18, 24, 32, 26, and 25 (hotspot area Ω1) above a local hotspot area all have insufficient computing power, the GEO satellite first selects offloading area Ω2 (the area enclosed by the outer dashed line in the figure) based on the task computation requirement CT estimated by the ground station. This ensures that the total computing power of nodes within area Ω2 equals β*CT, where β ≥ 1, with typical values ​​of 1.2 or 1.5.

[0106] Then, the central node 25 of the hot spot area Ω1 unloads the task in the order of inside to outside, that is, in the order of 1-hop node, 2-hop node, ... After the strategy is generated, the central node 25 is deleted in the hot spot area Ω1.

[0107] For the peripheral nodes 18, 24, 26, and 32 of the hot spot area Ω1, the tasks are not executed locally. Instead, the tasks are uninstalled in the unloading area Ω2 in the order from the outside to the inside. After the strategy is generated, the central node 25 and the peripheral nodes 18, 24, 26, and 32 are deleted from the hot spot area Ω1.

[0108] This continues until set Ω1 is empty.

[0109] Step 6: The LEO satellite distributes the task data according to the computing power scheduling strategy issued by the GEO satellite.

[0110] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A satellite computing power routing method, characterized in that, Specifically, the steps include the following: Step 1: The LEO satellite uses a lookup table to encode the computing power status information and reports the computing power status information to the GEO satellite; Step 2: Based on the received computing power status information, the GEO satellite generates the computing power status matrix and connectivity status matrix of the LEO network. Step 3: The ground station estimates the computing power requirements of the mission and sends the LEO network instructions to be executed and the estimated computing power requirements of the mission to the GEO satellite. Step 4: The GEO satellite, in conjunction with the estimated computing power demand sent from the ground station, sets the elements in the computing power status matrix that do not meet the computing power demand to 0. Step 5: Based on the computing power state matrix and connectivity state matrix after setting elements to 0, the GEO satellite generates the scheduling strategy for the LEO network using a bidirectional internal and external distribution method, and then distributes it to the LEO network; specifically, this includes the following sub-steps: Step 51: The GEO satellite generates an undirected graph based on the computing power state matrix and the connected state matrix. G undirected graph G Each node in G i,j Corresponding satellite nodes V i,j ,node G i,j The connectivity with surrounding nodes corresponds to the connectivity state matrix. L i,j ,in, L i,j = , , , , LEO satellites V i,j The status values ​​of the four inter-satellite links; nodes G i,j The computing power state corresponds to the computing power state matrix in = , CL i,j , CP i,j CA i,j , M i,j They represent LEO satellites. V i,j Current logical computing capabilities, parallel computing capabilities, intelligent computing capabilities, and storage capabilities; Step 52: GEO satellites in undirected graphs G Mark hotspot areas Hotspot areas Nodes included G i,j Its corresponding LEO satellite node V i,j Satellite nodes that submit computing power requests to GEO; Step 53: GEO satellites in undirected graphs G In the middle, with hot spots Centered on the central node, according to 1 hop, 2 hops, The unloading area is selected by expanding outwards in sequence. This makes the unloading area The total computing power of internal nodes equals ,in β ≥1; Step 54: Hotspot Area central node G m,n The goal is to find the target node using the first cost function shown in the following formula, with the aim of minimizing the value of the first cost function. G m,n , destination node, and G m,n Information about the intermediate nodes that need to be traversed to reach the destination node, i.e. G m,n Intermediate node, destination node > , denoted as computing power scheduling strategy S m,n Then Delete node G m,n ; Cost = αT comp,l +(1- α ) ( T trans,m + T comp,m ) in, T comp,l For locally calculated latency, T trans,m For the latency of task transmission, T comp,m The computation delay for the destination node is calculated, assuming only one destination node is selected each time. α = Task 1 / Task m,n 、 1- α = Task 2 / Task m,n , 0≤ α ≤1; Task m,n For satellite nodes V m,n The amount of data that the task originally needed to be executed Task 1 is V m,n The actual amount of task data executed locally. Task 2 represents the amount of data executed by the destination node; , Task 1,β 、 Task 1,γ , Task 1,δ They represent Task Part 1 requires logical computation, parallel computation, and intelligent computation. These represent the time required for logical computation, parallel computation, and intelligent computation, respectively. , Task 2,β 、 Task 2,γ , Task 2,δ They represent Task Part 2 requires logical computation, parallel computation, and intelligent computation. These represent the time required for logical computation, parallel computation, and intelligent computation, respectively. T trans,m = That is, the computation latency of the destination node is equal to the sum of the link latency between all nodes when unloading data. h This represents the number of hops required to transmit data from the source node to the destination node. L For the process h The connectivity value when hopping nodes; Step 55: Hotspot Area The outer nodes in the middle are marked as G pq The second cost function is used in the unloading region. The goal is to find the target node among the peripheral nodes, with the aim of minimizing the value of the second cost function. G pq , destination node, and G pq Information about the intermediate nodes that need to be traversed to reach the destination node, i.e. G pq Intermediate node, destination node > , denoted as computing power scheduling strategy S pq Then in the uninstallation area Delete node G pq : Cost = T trans,m +T comp,m in, T trans,m For the latency of task transmission, T comp,m The computation delay at the destination node; Step 56: Repeat steps 54 and 55 until a set is reached. If empty, GEO broadcasts all computing power scheduling policies to LEO; Step 6: The LEO satellite distributes the task data according to the computing power scheduling strategy issued by the GEO satellite.

2. The satellite computing power routing method as described in claim 1, characterized in that, Step 1 includes the following sub-steps: Step 11, the LEO node learns its own computing power status information; the computing power status information includes: logical computing power, parallel computing power, intelligent computing power, storage capacity, per-hop link identification communication capacity, and the current utilization rate of the above capabilities; Step 12: Normalize the logical computing power, parallel computing power, intelligent computing power, storage capacity, and per-hop link identifier communication capability respectively; Step 13: The normalized capability values ​​from Step 12 and the current utilization rate of each capability obtained in Step 11 are encoded by looking up a table. Step 14: The LEO satellite reports its computing power status information to the GEO satellite.

3. The satellite computing power routing method as described in claim 2, characterized in that, In step 2, LEO's computing power state matrix X = ,in, = , CL i,j , CP i,j 、 CA i,j 、M i,j They represent LEO satellites. V i,j Current logical computing capability, parallel computing capability, intelligent computing capability, and storage capability, i.e., node capability × (1 - current utilization rate), 1 ≤ i ≤ m 1≤ j ≤ n , m , n These represent the number of orbits and the number of satellites in those orbits, respectively. LEO's connected state matrix L = ,in, L i,j = , , , , LEO satellites V i,j The status values ​​of the four inter-satellite links.

4. The satellite computing power routing method as described in claim 3, characterized in that, Step 4 includes the following sub-steps: Step 41: The GEO satellite divides the current mission into a series of sub-tasks and estimates the computing power and storage space required to process the sub-tasks based on the size of the sub-task data and the size of the mission data. X min ,in, X min = ; , , , These represent the logical computing capability, parallel computing capability, intelligent computing capability, and storage capability required for subtask processing, respectively. Step 42: Remove elements in the computing power state matrix whose computing power is less than the computing power required for subtask processing. Set to 0, that is .

5. The satellite computing power routing method as described in claim 1, characterized in that, In step 53, β Choose 1.2 or 1.5.

Images