A distributed iterative computing method and system based on inter-satellite cooperation and master satellite rotation

Abstract

The application relates to the technical field of satellite wireless communication networks, and provides a distributed iterative calculation method and system based on intersatellite cooperation and master satellite rotation, which comprises the following steps: in response to a distributed iterative calculation task request, exchanging satellite state data through a heartbeat packet and generating a whole-network dynamic resource situation table; constructing an intersatellite link graph model and generating a communication time delay cost matrix; distributing a single round of iterative task according to a minimum indivisible task unit through the communication time delay cost matrix and the nominal computing power and the used load of each satellite, generating a joint task distribution matrix; determining the first satellite and the summary satellite of each round in a continuous multi-round iterative process, generating a first satellite dynamic rotation sequence, and outputting the task distribution result and the completion time of the multi-round iterative task; and through master satellite dynamic rotation, efficient alternate operation of the master satellite in the multi-round iterative calculation task is realized, a single master satellite long-term bearing of high-intensity calculation and communication load is avoided, and the risk of satellite overload and potential failure is reduced.

Description

Technical Field

[0001] This invention relates to the technical field of satellite wireless communication networks, specifically to a distributed iterative computing method and system based on inter-satellite collaboration and primary satellite rotation, which supports highly reliable on-orbit computing tasks with dynamic primary satellite rotation during multiple iterations. Background Technology

[0002] With the development of high-performance spaceborne processing chips, such as SoCs, FPGAs, and edge AI computing units, the processing capabilities of individual satellites have significantly improved, and satellite constellations now possess the physical foundation to support complex computing tasks. During in-orbit operation, facing sudden surges in computing demands, satellites can not only forward data but also process information on-orbit. Using inter-satellite links (ISL) to distribute tasks and coordinate computing power can effectively improve the response speed and processing efficiency of in-orbit tasks, representing a key technological means to address the real-time response challenges of time-sensitive tasks.

[0003] Existing on-orbit distributed computing methods primarily employ a "one-time distribution, single-time aggregation" task model. Tasks are typically broken down and distributed across multiple satellite nodes, with scheduling optimization considering communication latency and node load within a single task round. However, these methods usually rely on a fixed master satellite for task distribution and result aggregation, limiting their optimization capabilities within a single task round. For multi-round iterative tasks, these methods struggle to achieve continuous optimization and dependency modeling between rounds, leading to difficulties in coordinating result aggregation and parameter updates, thus limiting overall scheduling efficiency. Furthermore, existing distributed computing architectures suffer from system reliability deficiencies. A fixed master satellite bears high-intensity computation and communication tasks for extended periods during multi-round iterations, easily leading to overload, performance degradation, and heat accumulation issues, creating single-point-of-failure risks. In large-scale constellation environments, satellite states exhibit dynamic changes, such as satellite failures or load fluctuations, further increasing task scheduling complexity. Existing methods struggle to simultaneously achieve computational efficiency, load balancing, and communication path optimization, resulting in limited system scheduling performance.

[0004] Therefore, existing on-orbit computing technologies still have significant shortcomings in supporting multi-round iterative tasks, system reliability, and dynamic adaptability, making it difficult to meet the needs of complex space computing tasks. Summary of the Invention

[0005] To address the issues of low scheduling efficiency and excessive host satellite load in existing on-orbit multi-round iterative missions, this invention proposes a distributed iterative computing method and system based on inter-satellite coordination and host satellite rotation to meet node reliability requirements and shorten the overall completion cycle of the iterative mission. To achieve this objective, the invention employs the following technical solution: This application provides a distributed iterative computation method based on inter-satellite cooperation and primary star rotation, including, Step S1: In response to the distributed iterative computing task request, each satellite exchanges satellite status data, including satellite orbit position, communication forwarding delay, nominal computing power, used load and satellite protection time, through heartbeat packets, and generates a dynamic resource status table for the entire network based on the interactive information formed by the satellite status data of neighboring satellites. Step S2: Construct an inter-satellite link graph model based on the dynamic resource status table of the entire network, determine the shortest communication delay between any two satellites, and generate a communication delay cost matrix; Step S3: Using the communication delay cost matrix and the nominal computing power and used load of each satellite, allocate the single-round iteration task according to the smallest indivisible task unit to obtain the task load, expected completion time and corresponding communication path of each satellite, and generate a joint task allocation matrix. Step S4: Based on the joint task allocation matrix and the satellite protection time of each satellite, determine the first satellite and the summative satellite for each round in the continuous multi-round iteration process, generate the dynamic rotation sequence of the first satellite, and output the task allocation results and completion time of the multi-round iteration tasks.

[0006] Further, step S1 includes, In response to the distributed iterative computing task request, the system obtains satellite status data for each satellite participating in the distributed iterative computing, including satellite orbital position, communication forwarding delay, nominal computing power, used load, and satellite protection time. Among these, satellite orbital position is used to represent the on-orbit spatial position relationship of each satellite, communication forwarding delay is used to represent the communication overhead of inter-satellite links, nominal computing power is used to represent the baseline computing capability of each satellite, used load is used to represent the proportion of computing resources currently occupied by each satellite, and satellite protection time is used to represent the minimum time interval between two consecutive times that the same satellite serves as the lead satellite. Based on the obtained satellite orbital positions, communication forwarding delays, nominal computing power, used load, and satellite protection time, a dynamic resource status table for the entire network is generated.

[0007] Further, step S2 includes, The satellite orbital positions of each satellite are obtained based on the dynamic resource status table of the entire network, and the free space transmission delay between any two satellites is calculated based on the satellite orbital positions using the electromagnetic wave space propagation model. The communication forwarding delay of each satellite is obtained and superimposed on the free space transmission delay to obtain the comprehensive communication delay of the corresponding inter-satellite link; An inter-satellite link graph model is constructed based on the overall communication latency of the inter-satellite links. , where the vertex For each satellite node, edge The weights are the overall communication delays of the corresponding inter-satellite links; Based on inter-satellite link graph model The Dijkstra algorithm is used to traverse and calculate the shortest communication delay and corresponding path between any two satellite nodes, and a communication delay cost matrix is ​​generated. .

[0008] Further, step S3 includes, Step S301: Set the first star and the summing star in the current iteration round; Step S302: Based on the communication delay cost matrix, the nominal computing power and used load of each satellite, calculate the inherent link cost delay of each candidate satellite and the estimated completion time after allocating the next smallest indivisible task unit; Step S303: Allocate single-round iteration tasks sequentially according to the smallest indivisible task unit, select the candidate satellite with the shortest expected completion time as the carrier satellite, and update the task quantity and expected completion time of the carrier satellite after each allocation. At the same time, record the shortest communication path from the first satellite to the carrier satellite and from the carrier satellite to the summary satellite for each single-round iteration task. Step S304: Repeat step S303 until the total task allocation for the current iteration is completed, and generate a joint task allocation matrix based on the task amount, estimated completion time and corresponding shortest communication path of each participating satellite; the completion time of the current iteration is the maximum value among the estimated completion times of each participating satellite.

[0009] Further, step S302 includes, Obtain the shortest communication delay from the first satellite to each candidate satellite, and the shortest communication delay from the candidate satellites to the aggregate satellite, from the communication delay cost matrix; For each candidate satellite, calculate the inherent link cost delay required to receive the smallest indivisible task unit in the current iteration round and return it to the summary satellite, using the formula: , in, Due to inherent link cost latency, As the first star, For summarizing stars, For satellites participating in the computing task, the subscript is... This indicates the link transmission-related delay, and the arrow indicates the direction of data transmission. and The first satellite obtained from the communication delay cost matrix To satellite and satellites To the summary star The shortest communication latency; Based on satellite The inherent link cost latency, nominal computing power, and used load are used to calculate the estimated completion time, using the following formula: ; ; in, For satellite The computation time for processing the smallest indivisible task unit. The computational load for a single task. For satellite Current load For satellite The nominal computing power Due to inherent link cost latency, For satellite The number of satellite missions already allocated in the current iteration round. For satellite The estimated completion time after the next smallest indivisible task unit is assigned in the current iteration round.

[0010] Further, step S303 includes, Distribute each computation task one by one from the task pool of the current iteration round according to the smallest indivisible task unit, compare the estimated completion time of all satellites, and select the satellite with the smallest estimated completion time as the carrier satellite; The smallest indivisible task unit is assigned to the selected carrier satellite, the satellite task load and estimated completion time of the carrier satellite are updated in real time, and the shortest communication path corresponding to the shortest communication delay from the first satellite to the satellite and from the satellite to the summary satellite is recorded.

[0011] Further, step S4 includes, Step S401: Obtain the set of multi-round iterative tasks and the set of satellites participating in each iteration. And read the corresponding joint task allocation matrix, where, For the total number of rounds, A set of tasks for multiple rounds of iteration. For the first The task matrix for each iteration ; For the first The set of satellites participating in the calculation in each iteration. For the first The first round of calculations involved in the iteration One satellite, For the first The total number of satellites involved in the calculation during each iteration; Step S402: Initially determine the candidate first star for each iteration. And summary stars Constructing inter-satellite link topology And the satellite collection that will participate and task matrix Related; Step S403: Establish a multi-round iterative objective function, calculate the completion time of each round of iteration, and generate a dynamic rotation sequence of the first satellite for consecutive multi-round iterations based on the calculation results. The dynamic rotation sequence of the first satellite satisfies that the time interval between two adjacent rounds in which the same satellite is the first satellite is greater than or equal to the satellite protection time. The satellite protection time is the countdown of the corresponding satellite needing to exit high-intensity calculation and distribution due to temperature control or channel resource control. Step S404: Use a genetic algorithm to solve for the amount of computational tasks allocated and the corresponding transmission path in each iteration, so as to optimize the completion time of each round calculated by the objective function of multiple iterations, and output the amount of tasks allocated to each satellite in each iteration, the estimated completion time, and the overall optimal completion time of the multi-round iteration tasks.

[0012] Further, in step S403, a multi-round iterative objective function is established, and the completion time of each round of iteration is calculated, including: ; in, The objective function value for a multi-round iterative task. This represents the total number of rounds required in a multi-round iterative task. For the first The completion time of tasks in each iteration. For the first The set of satellites participating in the computational task in each iteration. For the first Inter-satellite link topology in round-of-round iteration For binary decision variables, when satellite The value is 1 when a satellite is allocated a task slot, and 0 otherwise. This is used to determine whether a satellite participates in this round of missions. For the first star Transmitted to satellite via inter-satellite link The shortest communication latency, For satellite Transmitted to the aggregation star via inter-satellite link The shortest communication latency, For satellite The computation time required to complete the assigned computation task. The constraints are for the objective function in multiple iterations. , The first Wheel and First The time it takes for the first satellite to complete its mission and send back the results. For the satellite in the j-th iteration The corresponding communication path identifier in the inter-satellite link topology. Satellite protection time is the minimum time interval that must be met between the completion of one round of mission distribution and result summarization as the lead satellite and the next time it participates in mission scheduling as the lead satellite.

[0013] Further, step S404 includes: Based on the joint task assignment matrix, the first star of each round and the summation star of the last round in consecutive iterations are encoded into individual strings of the genetic algorithm. , of which each For the first The first star of the round of iterations, For the final iteration's summary star, when the first After the first and final satellites of each iteration are confirmed, the task load and corresponding communication path of each satellite in this iteration are determined based on the joint task allocation matrix. Generate an initial population, wherein the initial population includes Each individual, and the first star of each individual in the first round, is confirmed by the task initiation quantity in the distributed iterative computing task request; for the second round to the third... The first star of the wheel is set to randomly initialize the number of individuals. Set the number of limited candidate initial individuals to And satisfy For numbers 1 to For an individual, the first satellite of the corresponding round is randomly selected from the currently available satellite set. For the satellite numbered... to For each individual, based on the joint task allocation matrix, the first satellite of the corresponding round is randomly selected from the top K satellites with the shortest communication delay with the first satellite of the previous round; The completion time of each individual in multiple iterations is calculated based on the objective function of multiple iterations, and the fitness value of each individual is determined in combination with the satellite protection time constraint of the first satellite in each round. Genetic algorithm iterative operations are performed on the initial population, including elite retention and roulette wheel selection based on the fitness values ​​of each individual, two-point crossover of offspring individuals to generate new individuals, and greedy replacement operation for individuals that trigger mutation. In each round of iteration, the fitness value of the individual is calculated, and the iteration is repeated until the convergence condition is met. The individual with the highest fitness value is selected as the first star dynamic rotation sequence. The convergence condition is any one of the following: reaching the preset maximum number of iterations, the change in the fitness of the best individual in consecutive iterations being less than a preset threshold, or the existence of an individual fitness value reaching a preset optimal value.

[0014] Based on the selected first satellite dynamic rotation sequence and joint task allocation matrix, the task allocation, estimated completion time and corresponding communication path of each satellite in each iteration are determined, and the overall optimal completion time of the multi-round iteration task is output.

[0015] Based on the same inventive concept, this application also provides a distributed iterative computing system based on inter-satellite cooperation and primary star rotation, employing the distributed iterative computing method based on inter-satellite cooperation and primary star rotation as described above, including: The situation awareness module is used to respond to distributed iterative computing task requests. Each satellite exchanges satellite status data, including satellite orbit position, communication forwarding delay, nominal computing power, used load and satellite protection time, through heartbeat packets. Based on the interactive information formed by the satellite status data of neighboring satellites, a dynamic resource situation table of the entire network is generated. The topology modeling module is used to construct an inter-satellite link graph model based on the dynamic resource status table of the entire network, determine the shortest communication delay between any two satellites, and generate a communication delay cost matrix. The task allocation module is used to allocate single-round iterative tasks according to the smallest indivisible task unit based on the communication delay cost matrix and the nominal computing power and used load of each satellite, to obtain the task load, expected completion time and corresponding communication path of each satellite, and generate a joint task allocation matrix. The results output module is used to determine the first and final satellites of each round in a series of iterations based on the joint task allocation matrix and the satellite protection time of each satellite, generate a dynamic rotation sequence of the first satellite, and output the task allocation results and completion time of the multi-round iteration tasks.

[0016] Furthermore, the result output module includes: The task management unit is used to obtain a set of tasks for multiple iterations. and the set of satellites participating in each iteration. And read the corresponding joint task allocation matrix, where, For the total number of rounds, A set of tasks for multiple rounds of iteration. For the first The task matrix for each iteration ; For the first The set of satellites participating in the calculation in each iteration. For the first The first round of calculations involved in the iteration One satellite, For the first The total number of satellites involved in the calculation during each iteration; The target computation unit is used to initially determine the candidate first star for each iteration. And summary stars Constructing inter-satellite link topology And the satellite collection that will participate and task matrix Correlation; Establish a multi-round iterative objective function, calculate the completion time of each round of iteration, and generate a dynamic rotation sequence of the first satellite for consecutive multi-round iterations based on the calculation results. The dynamic rotation sequence of the first satellite satisfies that the time interval between two adjacent rounds in which the same satellite is the first satellite is greater than or equal to the satellite protection time. The satellite protection time is the countdown of the corresponding satellite that needs to exit high-intensity calculation and distribution due to temperature control or channel resource control. The allocation optimization unit is used to solve the computational task allocation and corresponding transmission path in each iteration using a genetic algorithm, so as to optimize the completion time of each round calculated by the objective function of multiple iterations, and output the task allocation of each satellite in each iteration, the estimated completion time, and the overall optimal completion time of the multi-round iteration tasks.

[0017] Compared with the prior art, the present invention has at least one of the following beneficial effects: This application achieves efficient alternation of primary satellites in multi-round iterative computing tasks through dynamic primary satellite rotation, avoiding the long-term burden of high-intensity computing and communication loads on a single primary satellite, and reducing the risk of satellite overload and potential failure. Simultaneously, by combining shortest path computation and load balancing strategies, a joint scheduling result of computing task allocation and data transmission paths is generated within a single iteration, constructing a collaborative optimization model for transmission paths and computing resources, significantly reducing solution complexity and improving scheduling efficiency. This method is applicable to various on-orbit distributed computing scenarios such as large-scale constellation task scheduling, on-board autonomous collaborative computing, and distributed model training, especially suitable for multi-round iterative tasks such as federated learning and multi-source data fusion processing, providing an efficient and reliable collaborative computing strategy. Furthermore, this invention has good scalability and engineering practicality; its scheduling framework can be flexibly adjusted according to constellation size, satellite status, and available computing power. The design of each module is universal and replaceable, facilitating practical deployment and optimization, and possessing high engineering application value. Attached Figure Description

[0018] Figure 1 This is a flowchart of the distributed iterative computation method based on inter-satellite collaboration and primary star rotation in an embodiment of the present invention; Figure 2 This is a flowchart of single-round iterative task allocation and communication path generation in an embodiment of the present invention; Figure 3 This is a schematic diagram of a satellite node group in an embodiment of the present invention; Figure 4 This is a comparison chart of the total time consumption of the embodiments of the present invention and the prior art under different single-round task loads; Figure 5 This is a comparison chart of the total time consumption of the embodiments of the present invention and the prior art under different iteration rounds; Figure 6 This is a comparison chart of the total time consumption of the embodiments of the present invention and the prior art under different satellite nominal computing power. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0021] First Embodiment With the development of high-performance spaceborne processing chips, including SoCs, FPGAs, and edge AI computing units, the computing power of individual satellites has been significantly improved, providing a physical foundation for satellite constellations to undertake complex computing tasks. In actual on-orbit operation, satellite constellations face sudden surges in computing demands, requiring a shift from simple data forwarding to proactive information processing. Considering the potential communication congestion issues that may arise in large-scale satellite operations and control via space-to-ground links, achieving on-orbit task distribution and computing power coordination through inter-satellite links has become a key approach for real-time response to time-sensitive tasks. Against this backdrop, this application proposes a method to achieve constellation state awareness and communication scheduling optimization by constructing a dynamic resource status table and a communication latency cost matrix for the entire network.

[0022] However, existing distributed computing methods have significant limitations: traditional methods often adopt a "one-time distribution, single-time aggregation" model, breaking down tasks and distributing them across multiple satellites for execution, lacking modeling for dependencies across iterations. For multi-round iterative tasks, each round of computation requires summarizing results and updating parameters before proceeding to the next round. Existing methods struggle to coordinate communication and computation, leading to overall inefficiency. Furthermore, existing architectures typically assign a single primary satellite to handle both task distribution and result aggregation, resulting in prolonged high-load operation during multi-round iterations, which can easily lead to performance degradation, heat accumulation, and single-point failure risks. Dynamic changes in satellite status, including faults and load fluctuations, also increase the complexity of task scheduling, making it difficult for existing methods to balance solution efficiency and allocation quality.

[0023] To address the aforementioned issues, this application proposes a multi-round iterative collaborative computation method based on dynamic rotation of the primary star, such as... Figure 2 As shown, each satellite transmits its orbital position, communication forwarding delay, nominal computing power, used load, and satellite protection time (i.e., the countdown before a satellite needs to exit high-intensity computing and distribution due to temperature control or channel resource control) via heartbeat packets. Each satellite independently forms a consistent dynamic resource status table for the entire network. Considering communication forwarding delay and path transmission delay, a communication delay cost matrix for the shortest communication delay between any two satellites is derived based on graph theory. Based on the communication delay cost matrix and load balancing strategy, according to nominal computing power and used load, the computational task allocation and path corresponding to each satellite are determined when the single iteration time is minimized, forming a joint task allocation matrix. Based on the joint task allocation matrix and combined with the satellite protection time, the combinatorial optimization problem of computational task allocation and path during continuous iteration is solved, forming the allocation result and transmission path throughout the distributed iterative task process. By jointly analyzing the satellite constellation's computing resources, communication latency, and satellite load in each iteration, a joint task allocation matrix is ​​constructed to determine the task allocation amount and corresponding transmission path for each satellite. A dynamic rotation sequence for the first satellite is generated to ensure that the time interval between adjacent iterations where the same satellite is the first satellite meets the satellite protection time requirement, reducing the risk of single-satellite overload and ensuring the continuity of iterative tasks. Furthermore, this application employs a genetic algorithm to optimize the selection of the first satellite and the task allocation path for multiple iterations, minimizing the completion time of each iteration and outputting the task allocation amount for each satellite in each iteration, the estimated completion time, and the overall optimal completion time for multiple iterations. Through the combination of shortest communication path calculation and load balancing strategies, this invention can efficiently generate high-quality task allocation schemes in complex dynamic environments, improving the efficiency of constellation collaborative computing while balancing system reliability and task completion timeliness, demonstrating significant engineering application value. Specific implementation methods are as follows: like Figure 1 , 3 As shown, this application provides a distributed iterative computing method based on inter-satellite cooperation and primary star rotation, including: Step S1: In response to the distributed iterative computing task request, each satellite exchanges satellite status data, including satellite orbit position, communication forwarding delay, nominal computing power, used load and satellite protection time, through heartbeat packets. The satellite protection time is the countdown before the satellite needs to exit high-intensity computing and distribution due to temperature control or channel resource control. Based on the satellite status data of neighboring satellites, a dynamic resource status table for the entire network is generated for iterative task allocation and path optimization. Specifically, each satellite broadcasts heartbeat data units via inter-satellite links (ISL) at preset intervals. The heartbeat packet includes the satellite's current orbital position parameters, internal processing and forwarding latency, nominal computing power (floating-point arithmetic capability), real-time used computing load, and satellite protection time (i.e., a countdown before the satellite needs to exit high-intensity computing and distribution due to temperature control or channel resource control). After receiving the heartbeat packet from a neighboring satellite, each satellite uses a distributed consensus mechanism to locally construct and independently maintain a globally consistent dynamic resource status table. This table reflects the spatial topology and computing resource abundance of each satellite within the constellation in real time, providing support for resource scheduling.

[0024] Step S2: Based on the satellite orbit positions and communication forwarding delays in the dynamic resource status table of the entire network, the constellation is abstracted into a graph model, and the shortest communication delay between any two satellites is calculated using the Dijkstra algorithm, and a communication delay cost matrix is ​​generated. Step S3: Using the communication delay cost matrix and the nominal computing power and used load of each satellite, the single-round iteration tasks are allocated to the satellite with the shortest expected completion time in sequence according to the smallest indivisible task unit. During the allocation process, the task quantity and corresponding expected completion time of each satellite are updated in real time. At the same time, the shortest communication path of each task from the first satellite to the satellite and then to the summing satellite is recorded to generate a joint task allocation matrix. The joint task allocation matrix includes the task quantity of each satellite and the corresponding communication path. Step S4: Based on the joint task allocation matrix and the satellite protection time of each satellite, determine the first satellite and the summing satellite for each round in the continuous multi-round iteration process, and generate a dynamic rotation sequence of the first satellite to determine the time interval between two adjacent rounds in which the same satellite is the first satellite, so as to form the task allocation result of the entire distributed iterative task and the communication path corresponding to each task. Output the satellite task allocation, estimated completion time and overall optimal completion time of the multi-round iterative task for each satellite in each round of iteration.

[0025] Further, step S1 includes, In response to the distributed iterative computing task request, the system obtains satellite status data for each satellite participating in the distributed iterative computing, including satellite orbital position, communication forwarding delay, nominal computing power, used load, and satellite protection time. Among these, satellite orbital position is used to represent the on-orbit spatial position relationship of each satellite, communication forwarding delay is used to represent the communication overhead of inter-satellite links, nominal computing power is used to represent the baseline computing capability of each satellite, used load is used to represent the proportion of computing resources currently occupied by each satellite, and satellite protection time is used to represent the minimum time interval between two consecutive times that the same satellite serves as the lead satellite. Based on the obtained satellite orbital positions, communication forwarding delays, nominal computing power, used load, and satellite protection time, a dynamic resource status table for the entire network is generated.

[0026] Further, step S2 includes, The satellite orbit position of each satellite is obtained based on the dynamic resource status table of the entire network, and the free space transmission delay between any two satellites is calculated based on the satellite orbit position using the electromagnetic wave space propagation model. The communication forwarding delay of each satellite is obtained and superimposed on the free space transmission delay to obtain the comprehensive communication delay of each link; Abstracting star clusters into a graphical model of inter-satellite links , where the vertex For each satellite node, edge The weights are the combined communication delays between the corresponding satellites. Dijkstra's algorithm is used to traverse and calculate the shortest communication delay and corresponding path between any two satellite nodes. The shortest communication delay and corresponding path between each satellite node are summarized to generate a communication delay cost matrix. .

[0027] Specifically, based on the satellite orbital position information obtained in step S1, the free-space transmission delay between satellites is calculated using an electromagnetic wave space propagation model. Simultaneously, the internal hardware forwarding delay of each satellite, fed back in the heartbeat packets, is superimposed, abstracting the entire constellation into a dynamically weighted graph. , where the vertex Representing satellite, side The weights represent the overall communication delay. An improved Dijkstra's algorithm is used to traverse and calculate the path between any two satellites in the graph, generating the shortest communication delay cost matrix. This matrix not only quantifies the minimum time cost between any two satellites, but also records the corresponding optimal physical path.

[0028] In the above steps, the calculation method for the free-space transmission delay between any two satellites using the electromagnetic wave space propagation model includes: Based on the satellite's position coordinates in the J2000.0 geocentric inertial coordinate system and Calculate the straight-line distance between the two satellites: ; The free space transmission delay is obtained by the following formula. , in The speed at which electromagnetic waves propagate in free space is approximately 3 × 10⁻⁶. 8 m / s.

[0029] The final result This refers to the free-space transmission delay between any two satellites, which can be superimposed with the internal relay delay of the satellite to form the comprehensive communication delay, and used as the weight of the edges in the graph model.

[0030] Further, step S3 includes, Step S301: Set the first star and the summing star in the current iteration round; Step S302: Based on the communication delay cost matrix, the nominal computing power and used load of each satellite, calculate the inherent link cost delay and the estimated completion time of each satellite that may participate in the computing task; Step S303: Assign tasks to the satellites with the shortest estimated completion time in order of the smallest indivisible task units, update the satellite task load and estimated completion time in real time, and record the shortest communication path for each task from the first satellite and the summarizing satellites. Step S304: Repeat step S303 until the total task allocation for this round is completed. After completion, determine the estimated completion time of each task. The estimated completion time of unallocated tasks is recorded as 0. The non-zero completion time of participating satellites is extracted, and the overall optimal completion time of the single-round iteration task is determined as the maximum value of each estimated completion time. At the same time, a joint task allocation matrix containing the satellite task amount, estimated completion time and shortest communication path of each satellite is generated.

[0031] Further, in step S302, based on the communication delay cost matrix, the nominal computing power of each satellite, and the used load, the inherent link cost delay and estimated completion time of each satellite potentially participating in the computing task are calculated, including... For each satellite in the constellation participating in the computation task, calculate the corresponding inherent link cost latency using the following formula: , in, Due to inherent link cost latency, As the first star, For summarizing stars, For satellites participating in the computing task, the subscript is... This indicates the link transmission-related delay, and the arrow indicates the direction of data transmission. and The first satellite obtained from the communication delay cost matrix To satellite and satellites To the summary star The shortest communication latency; Based on satellite The inherent link cost latency, nominal computing power, and used load are used to calculate the estimated completion time, using the following formula: ; ; in, For satellite The computation time for processing the smallest indivisible task unit, where D is the computational load of a single task. For satellite Current load For satellite The nominal computing power Due to inherent link cost latency, For satellite The number of satellite missions already allocated in the current iteration round. For satellite The estimated completion time after the next smallest indivisible task unit is assigned in the current iteration round.

[0032] Further, in step S303, tasks are assigned sequentially to the satellites with the shortest estimated completion time according to the smallest indivisible task unit, the satellite task load and estimated completion time are updated in real time, and the shortest communication path from the first satellite to the summing satellite is recorded for each task, including... Distribute each computation task one by one from the task pool of the current iteration round according to the smallest indivisible task unit, compare the estimated completion time of all satellites, and select the satellite with the smallest estimated completion time as the carrier satellite; The smallest indivisible task unit is assigned to the selected carrier satellite, the satellite task load and estimated completion time of the carrier satellite are updated in real time, and the shortest communication path corresponding to the shortest communication delay from the first satellite to the satellite and from the satellite to the summary satellite is recorded.

[0033] Specifically, this step mainly relies on the communication latency cost matrix and load balancing strategy. Based on nominal computing power and used load, it determines the computational task allocation and path for each satellite when the latency of a single iteration is minimized, forming a joint task allocation matrix. In this step, the load allocation for minimizing latency in a single iteration adopts a discrete greedy allocation strategy based on the shortest path. This method transforms a complex nonlinear optimization problem into an iterative optimization process with the minimum task unit as the step size. Its specific logical implementation process is as follows: Set the first star in the current iteration round And summary stars Based on the shortest communication delay cost matrix generated in step S2 For each potential satellite in the constellation to participate in the computing mission Calculate the inherent link cost latency when it carries the task. The inherent link cost latency From the main star To satellite Transmission delay and node To summarize satellites The sum of transmission delays constitutes, i.e. If a satellite does not undertake any computing tasks in the current round, it will not incur transmission and computing time, and this time will not be included in the overall completion time.

[0034] The greedy optimization logic of sequential task allocation takes into account the characteristic that the computational task has a minimum indivisible unit. This embodiment adopts a task allocation strategy based on the minimum unit. Let the total number of tasks in this round be... The time required for a single satellite to complete a single computing task is Where D is the computational load of a single computing task. This is the satellite's used payload. This represents the nominal computing power of the satellite, and the time set of each satellite is... The allocation process is as follows: Marginal time assessment: Before allocating each task unit, the time consumption of each satellite is calculated in real time. Based on the currently allocated workload, the estimated total completion time would be increased if additional units of this task were undertaken. This time is determined by the inherent link cost. Determined by the cumulative calculation time, the calculation formula is as follows: ,in The number of missions currently assigned to this satellite. Optimal satellite selection: Compare the estimated total completion time of all satellites. The system assigns the current mission units to the satellite with the shortest estimated total time and updates the mission allocation for that satellite in real time. and current time cost This forms a set of tasks assigned to each star. and the corresponding time cost set Iterative loop: Repeat the above process until... All task units have been assigned.

[0035] After the joint task allocation matrix is ​​generated and the allocation is completed, the allocation results for each satellite are evaluated: for satellites not assigned tasks, their time cost is recorded as 0 and they are not included in the final delay calculation. For satellites participating in the calculation, their final time cost set is extracted. Non-zero terms in the data. Optimal time determination: Select... The term with the largest lumped value is taken as the overall optimal time cost for this iteration. That is, the bottleneck time of a single iteration is determined by the longest path in the "barrel effect".

[0036] The final output is the task allocation set for each star. The time set for each star to complete calculations under the optimal path and the optimal delay value for a single round Repeat the above steps, using each star as the first and final star, and combine the above data to form a joint task allocation matrix. This serves as the basic input for multi-round continuous iterative combination optimization in step S4.

[0037] Further, step S4 includes, Obtain a set of multi-round iterative tasks and the set of satellites participating in each iteration. And read the corresponding joint task allocation matrix, where, For the total number of rounds, A set of tasks for multiple rounds of iteration. For the first The task matrix for each iteration ; For the first The set of satellites participating in the calculation in each iteration. For the first The first round of calculations involved in the iteration One satellite, For the first The total number of satellites involved in the calculation during each iteration; The initial determination of the candidate first star for each iteration And summary stars Constructing inter-satellite link topology And the satellite collection that will participate and task matrix Related; Establish a multi-round iterative objective function, calculate the completion time of each round of iteration, and generate a dynamic rotation sequence of the first satellite for consecutive multi-round iterations based on the calculation results. The dynamic rotation sequence of the first satellite satisfies that the time interval between two adjacent rounds in which the same satellite is the first satellite is greater than or equal to the satellite protection time. The satellite protection time is the countdown of the corresponding satellite that needs to exit high-intensity calculation and distribution due to temperature control or channel resource control. A genetic algorithm is used to solve for the amount of computational tasks allocated and the corresponding transmission path in each iteration, so as to optimize the completion time of each round calculated by the objective function of multiple iterations, and output the amount of tasks allocated to each satellite in each iteration, the estimated completion time, and the overall optimal completion time of the tasks in multiple iterations.

[0038] Furthermore, a multi-round iterative objective function is established, and the completion time of each round of iteration is calculated, including: ; in, The objective function value for a multi-round iterative task. This represents the total number of rounds required in a multi-round iterative task. For the first The completion time of tasks in each iteration. For the first The set of satellites participating in the computational task in each iteration. For the first Inter-satellite link topology in round-of-round iteration For binary decision variables, when satellite The value is 1 when a satellite is allocated a task slot, and 0 otherwise. This is used to determine whether a satellite participates in this round of missions. For the first star Transmitted to satellite via inter-satellite link The shortest communication latency, For satellite Transmitted to the aggregation star via inter-satellite link The shortest communication latency, For satellite The computation time required to complete the assigned computation task. The constraints are for the objective function in multiple iterations. , The first Wheel and First The time it takes for the first satellite to complete its mission and send back the results.

[0039] Specifically, this step mainly involves solving the allocation combination optimization problem in the continuous iteration process based on the joint task allocation matrix within a single iteration, combined with the satellite protection time, to form the allocation results and transmission paths in the entire computation process.

[0040] Star Clusters Multi-round iterative computation task A total of The calculation is performed in one iteration. It includes an inter-satellite link topology First Star Summary Star , can be written as Star clusters It describes the set of all on-orbit satellites that can participate during the computational task. This refers to the number of satellites. The goal is to complete the computational tasks of all iterations as early as possible; therefore, the following objective function can be established:

[0041] in The first equation represents the time it takes for all relevant satellites to complete calculations and transmit results during a given iteration. The second equation represents the time lost when the last satellite in that iteration transmits its results, marking the end of that iteration. Specifically, the time consumed by a satellite in one iteration consists of the following three parts: The first satellite was transmitted to the satellite via multiple hops. The delay, The delay is the time it takes for the satellite to transmit the calculation results to the aggregation satellite; the two together constitute the transmission delay on the link. The time consumed in completing the assigned computing tasks for the satellite. For binary decision variables, when satellite When allocating tasks, the computation time and link overhead involved in the satellite need to be taken into account. For satellites that have not been assigned any mission load, It is important to note here that the information transmission paths, i.e., the distribution paths and aggregation paths, must satisfy the inter-satellite link topology. The constraint of this model is that the time interval for a satellite to act as the primary satellite must be approximately equal to the satellite protection time. .

[0042] Furthermore, a genetic algorithm is used to solve for the computational task allocation and corresponding transmission path in each iteration, in order to optimize the completion time of each round calculated through the multi-round iterative objective function, and output the task allocation to each satellite in each iteration, the estimated completion time, and the overall optimal completion time of the multi-round iterative tasks, including: The first and last stars of each iteration are encoded into individual strings for the genetic algorithm. , of which each For the first The first star of the round of iterations, For the final iteration, each individual represents the dynamic rotation order of the first star in multiple iterations and the corresponding task allocation scheme, and generates the initial population; The completion time of each individual in multiple iterations is calculated based on the objective function of multiple iterations, and the fitness value of each individual is determined in combination with the satellite protection time constraint of the first satellite in each round. Selection, crossover, and mutation operations are performed on the population to generate a new generation of individuals. Fitness values ​​are calculated in each iteration, and this process is repeated until convergence is met. The individual with the highest fitness is selected as the first star in the dynamic rotation sequence. To balance the algorithm's convergence speed with population diversity, a joint selection mechanism combining elite preservation and roulette wheel selection is employed. First, the best individual with the highest fitness in the parent population is directly copied into the offspring, protecting it from crossover and mutation and preserving the historically best genes. Second, the remaining offspring individuals are selected through roulette wheel selection, where the probability of an individual being selected is proportional to its fitness as a percentage of the total fitness of the current population.

[0043] To effectively achieve gene recombination, a two-point crossover strategy is employed. Two crossover points are randomly set in the coding strings of two paired parent individuals, and the gene segments between these points are interchanged. This allows for the exploration of offspring individuals with new combination characteristics while preserving some desirable sequence structures from the parent individuals.

[0044] To accelerate convergence, an adaptive mutation mechanism is introduced. First, the individual fitness is compared with the population average fitness to determine the mutation probability for different individuals. Only for individuals that trigger mutation, a local greedy optimization is performed: excluding the beginning and end of the sequence, randomly selecting internal gene loci and extracting satellites from their preceding and following gene loci; based on the task allocation matrix, the global satellite library is traversed, and the satellite with the smallest sum of delays between itself and the preceding and following gene loci is selected to replace the original gene.

[0045] Specifically, this embodiment uses a genetic algorithm to solve the dynamic rotation sequence of the first satellite in multiple consecutive iterations, and determines the task allocation and communication path in each iteration based on the joint task allocation matrix. The process is as follows: (1) Based on the joint task assignment matrix, the first star of each round and the summation star of the last round in the continuous multiple iterations are encoded into individual strings of the genetic algorithm. , of which each For the first The first star of the round of iterations, For the final iteration's summary star, when the first After the first and final satellites of each iteration are confirmed, the task load and corresponding communication path for each satellite in that iteration are determined based on the joint task allocation matrix. Specifically, when... At that time, the first The summary star of the round of iteration is the first one. The first star of the round of iteration, when At that time, the first The summary star of the round of iterations is Therefore, by solving for the first star of each round in multiple consecutive iterations and the summing star of the last round, the global task allocation result and communication path of the multi-round iterative task can be mapped.

[0046] (2) Generate an initial population, wherein the initial population includes Each individual, and the first star of each individual in the first round, is confirmed by the task initiation quantity in the distributed iterative computing task request; for the second round to the third... The first star of the wheel is set to randomly initialize the number of individuals. Set the number of limited candidate initial individuals to And satisfy For numbers 1 to For an individual, the first satellite of the corresponding round is randomly selected from the currently available satellite set. For the satellite numbered... to Individuals are randomly selected from the top K satellites with the shortest communication delay to the first satellite of the previous round, based on the joint task allocation matrix. This hybrid initialization method expands the search range for the first satellite by using randomly initialized individuals, preventing the algorithm from getting trapped in local optima too early. Furthermore, it prioritizes satellites with shorter communication delays to the first satellite of the previous round by using restricted candidate initialized individuals, improving the overall quality of the initial population and thus balancing global exploration capability with local optimization efficiency.

[0047] (3) Calculate the completion time of each individual in multiple iterations based on the objective function of multiple iterations, and determine the fitness value of each individual in combination with the satellite protection time constraint of the first satellite in each round. The steps include: taking any individual in the current population as the individual to be evaluated, which corresponds to a dynamic rotation sequence of the first satellite in a set of consecutive multiple iterations and the last round of summary satellite; determining the workload and corresponding communication path of each satellite in each iteration based on the first satellite and summary satellite in each round of the individual, and in combination with the joint task allocation matrix. For each iteration, calculate the completion time of each participating satellite in that round. The completion time includes the shortest communication delay required for the first satellite to send the current round iteration task to the participating satellite, the calculation time required for the participating satellite to complete the assigned task, and the shortest communication delay required for the participating satellite to return the calculation result to the summary satellite of that round; and determining the maximum value among the completion times of each participating satellite in that round as the completion time of that round iteration. The completion times of each iteration for the same individual are summed to obtain the total iteration time for that individual. Simultaneously, it is determined whether the dynamic rotation sequence of the first satellite in that individual satisfies the satellite protection time constraint, i.e., whether the time interval between two consecutive instances of the same satellite serving as the first satellite is greater than or equal to the corresponding satellite protection time. If the satellite protection time constraint is satisfied, the fitness value of the individual is determined based on the total iteration time. If the satellite protection time constraint is not satisfied, the fitness value of the individual is reduced, or a protection time penalty term is added to the total iteration time before calculating the fitness value. The fitness value is used to evaluate the quality of the dynamic rotation sequence of the first satellite corresponding to the individual. The smaller the total iteration time and the more it satisfies the satellite protection time constraint, the higher the fitness value of the individual; the larger the total iteration time or the greater the degree of violation of the satellite protection time constraint, the lower the fitness value of the individual. When the mission initiating satellite, as the first satellite in the first round, cannot establish a communication link with other satellites, it will be used as the sole carrier satellite for the corresponding round. The mission initiating satellite will independently complete all the computational tasks in the iteration of that round, and the completion time of the corresponding round will be determined based on the computational time of the mission initiating satellite. This ensures that each individual in the genetic algorithm can obtain an evaluable fitness value and avoids the generation of unevaluable individuals during the iteration of the genetic algorithm.

[0048] (4) Perform genetic algorithm iterative operations on the initial population, including elite retention and roulette wheel selection based on the fitness value of each individual, two-point crossover of offspring individuals to generate new individuals, and greedy replacement operation on individuals that trigger mutations. Calculate the individual fitness value in each iteration, repeat the iteration until the convergence condition is met, and select the individual with the highest fitness value as the first star dynamic rotation sequence. The convergence condition is any one of the following: reaching the preset maximum number of iterations, the change in the fitness of the best individual in continuous iterations being lower than a preset threshold, or the existence of an individual fitness value reaching a preset optimal value. The steps include: in each iteration, a combination of elite retention and roulette wheel selection is performed on the population. The individual with the highest fitness in the previous generation is directly retained, and the remaining individuals generate the next generation according to their fitness ratio through roulette wheel selection. Two crossover points are randomly set in the encoding strings of two paired parent individuals, and the positions of gene segments between the crossover points are swapped to generate a new generation of offspring individuals. At the same time, the superior sequence structure of the parent generation is preserved. The mutation probability is adaptively set according to the individual fitness and the average fitness of the population. Local greedy optimization is performed only on individuals that trigger mutation: genes at the beginning and end of the sequence are excluded, internal gene positions are randomly extracted, and the satellite with the shortest communication delay among the adjacent gene positions is selected to replace the original gene to accelerate convergence and maintain population diversity. The individual with the highest fitness is selected as the first satellite in the dynamic rotation sequence. The task load, expected completion time and corresponding communication path of each satellite in each iteration are determined according to the joint task allocation matrix, and the overall optimal completion time of the multi-round iteration task is output.

[0049] The following example illustrates step S4 in detail. Taking the constellation as an example, it includes 10 satellites in orbit (NS=10), and the total number of rounds of multi-round iteration missions is 5. =5), the minimum number of indivisible task units in each iteration is 50.

[0050] The nominal computing power of each satellite The satellite has used a payload of [5, 4.5, 4.8, 5.2, 4.9, 5.1, 4.7, 5, 4.8, 5.2] GFLOPS. The satellite protection time is [0.2, 0.3, 0.25, 0.2, 0.3, 0.15, 0.2, 0.25, 0.2, 0.15]. Set the time to 3 seconds and perform the following steps: Encode the first and last stars of each iteration into a genetic algorithm individual string: [ ], each of which Indicates the first The first star of the round of iterations, This is the final round of star aggregation. Each individual determines the dynamic rotation order of the first star in all rounds, as well as the task allocation amount and transmission path corresponding to the joint task allocation matrix. The initial population size is set to 50 individuals.

[0051] For each individual, the completion time for each round is calculated using a multi-round iterative objective function: ; And satisfy the constraints: ; in, This represents the shortest communication latency via inter-satellite links. The computation time required for the satellite to complete the task assignment. For binary decision variables (participation = 1, non-participation = 0), the fitness value is the reciprocal of the total completion time for each individual.

[0052] In the initial population, the first satellite for all individuals in the first round is assigned by the mission. In subsequent rounds, 10 individuals are randomly generated to become the first satellite for the next round; and 40 individuals are randomly selected from the top 3 satellites with the shortest latency to the first satellite of the previous round according to the mission allocation matrix. In each generation, the five individuals with the highest fitness are retained and directly enter the offspring generation; the remaining 45 individuals select their parent individuals through a roulette wheel selection method, with the selection probability being proportional to the fitness ratio, ensuring that excellent solutions are retained while maintaining diversity. Crossover employs a two-point crossover strategy: two crossover points are randomly selected, and corresponding segments between the two individuals are exchanged to generate new candidate solutions. Mutation uses an adaptive mutation strategy: individuals with fitness above the average have a mutation probability of 0.05; individuals with fitness below the average have a mutation probability of 0.2. Among the mutated individuals, the first and last ends of the individual sequence are excluded, and several internal gene loci are randomly selected. Based on the task allocation matrix, the candidate satellite with the smallest sum of satellite delays with its adjacent gene loci is selected to replace the original gene.

[0053] Fitness calculations are repeated until crossover and mutation, with a maximum of 100 generations, or the process terminates when the optimal fitness in the population no longer improves after 10 consecutive generations. Finally, the individual with the highest fitness is selected as the first satellite in the dynamic rotation sequence. Combined with the joint task allocation matrix TL, the task load, estimated completion time, and overall optimal completion time for each satellite in each iteration are output.

[0054] This specific implementation allows for dynamic rotation of the first satellite and optimization of task allocation while ensuring satellite protection time constraints. It improves the balance of communication and computing load under multi-round iterative tasks, effectively reduces the risk of single satellite overload, and ensures computing efficiency and reliability in complex space environments.

[0055] Finally, based on the selected first satellite dynamic rotation sequence and joint task allocation matrix, the task allocation and estimated completion time of each satellite in each iteration are determined, and the overall optimal completion time of the multi-round iteration task is output.

[0056] In addition, to verify the effectiveness of the solution described in this embodiment, the following experiments were conducted: The simulation experiment is as follows: Based on the configuration of large high- and low-Earth orbit satellite constellations and the capabilities of intelligent satellites, a satellite node group consisting of 34 satellites (6 geostationary orbit satellites and a 7 / 4 / 1 low-Earth orbit Walker constellation) was constructed. The satellites are numbered sequentially from 1 to 34 (high-Earth orbit satellites are numbered 1-6, and low-Earth orbit satellites are numbered 7-34). The existing payload of each satellite is randomly selected from 10%, 20.5%, and 32.5%. The multi-round iterative calculation tasks and the main parameters of the satellites are as follows.

[0057]

[0058] The results of the primary stars in each round and the final round summary stars obtained based on this method are as follows: 11 14 twenty one 31 15 twenty four 16 7 32 twenty two 9 33 12 26 10 17 27 34 25 14 11 31 twenty one 13 16 twenty four 15 7 32 twenty two 9. Its time consumption is 4.24 seconds.

[0059] This reveals two key observations: First, the primary satellite transfer primarily occurs between low-Earth orbit (LEO) satellites. If the same number of satellites need to share the load, LEO satellites can effectively reduce overhead between paths. Second, there is a periodic pattern of primary satellite switching within the primary satellite transfer path, specifically, the 7 satellites in the path... 32 twenty two Part 9 indicates that the primary star selected for the overall time-optimal objective will become a primary star node again after the protection period has passed. Both of these points demonstrate that this method can guarantee the convergence of the solution algorithm and effectively avoid high-latency resource scheduling results.

[0060] Furthermore, a comparison was made between our proposed method (LBIGA represents the genetic algorithm with adaptive mutation probability proposed in this method), the conventional genetic algorithm based on this preprocessing (LBGGA), and the conventional genetic algorithm without this preprocessing (GGA) under varying main parameters. Experimental results are as follows: Figure 4 , Figure 5 and Figure 6 As shown.

[0061] Figure 4 The experiment demonstrates the computation completion time of different solution algorithms under varying single-round computation task scales. The results show that the proposed method achieves a shorter time compared to other comparative algorithms. The time consumption of all three methods increases with the number of computation tasks. This method benefits from its shortest path-based load balancing strategy, which ensures that incremental computation tasks are effectively distributed among previously participating nodes. Furthermore, as the number of nodes increases, more nodes are introduced to participate in the computation, resulting in a near-linear increase in overall time consumption compared to the number of computation tasks. Figure 5 The computation completion time of different scheduling algorithms is shown at different iteration counts. As the number of iterations increases, the advantage of this method over other methods widens; compared to the GGA method, the time advantage of this method has increased from 2.1 seconds to 4.1 seconds. This phenomenon is mainly due to the fact that this method can form a stable periodic primary star loop; with the increase of iteration count, more periods will accumulate, resulting in a greater advantage.

[0062] Figure 6 The performance differences were demonstrated under different single-satellite computing capabilities. With increasing computing power, the time consumption of all three methods decreased, with LBIGA and LBGGA showing relatively stable trends compared to GGA. Since the total time consumption consists of communication path consumption and computation time, increasing computing power can only reduce computation time. The load balancing preprocessing proposed in this paper demonstrates optimal path selection, showing a significant improvement compared to the GGA method which lacks preprocessing.

[0063] Second Embodiment Based on the same inventive concept, this invention also provides a distributed iterative computing system based on inter-satellite cooperation and primary star rotation, employing the distributed iterative computing method based on inter-satellite cooperation and primary star rotation as described above, including: The situation awareness module is used to respond to distributed iterative computing task requests. Each satellite exchanges satellite orbital position, communication forwarding delay, nominal computing power, used load and satellite protection time through heartbeat packets, and generates a dynamic resource situation table of the whole network based on the received heartbeat packet information for iterative task allocation and path optimization. The topology modeling module is used to abstract the constellation into a graph model based on the satellite orbital positions and communication forwarding delays in the dynamic resource status table of the entire network, and to calculate the shortest communication delay between any two satellites using the Dijkstra algorithm, and generate a communication delay cost matrix. The task allocation module is used to allocate single-round iteration tasks to the satellite with the shortest expected completion time in sequence according to the smallest indivisible task unit based on the communication delay cost matrix and the nominal computing power and used load of each satellite. During the allocation process, the task quantity and corresponding expected completion time of each satellite are updated in real time. At the same time, the shortest communication path of each task from the first satellite to the satellite and then to the summing satellite is recorded to generate a joint task allocation matrix. The joint task allocation matrix includes the task quantity of each satellite and the corresponding communication path. The data processing module is used to determine the first satellite and the summative satellite in each round of continuous iteration based on the joint task allocation matrix and the satellite protection time of each satellite, and to generate a dynamic rotation sequence of the first satellite. This sequence is used to determine the time interval between two adjacent rounds in which the same satellite is the first satellite, so as to form the task allocation result and the communication path corresponding to each task in the entire process of the distributed iterative task. The results output module is used to output the satellite task allocation, estimated completion time, and overall optimal completion time for each satellite in each iteration.

[0064] Furthermore, the data processing module includes, The task management unit is used to obtain a set of tasks for multiple iterations. and the set of satellites participating in each iteration. And read the corresponding joint task allocation matrix, where, For the total number of rounds, A set of tasks for multiple rounds of iteration. For the first The task matrix for each iteration ; For the first The set of satellites participating in the calculation in each iteration. For the first The first round of calculations involved in the iteration One satellite, For the first The total number of satellites involved in the calculation during each iteration; The target computation unit is used to initially determine the candidate first star for each iteration. And summary stars Constructing inter-satellite link topology And the satellite collection that will participate and task matrix Correlation; Establish a multi-round iterative objective function, calculate the completion time of each round of iteration, and generate a dynamic rotation sequence of the first satellite for consecutive multi-round iterations based on the calculation results. The dynamic rotation sequence of the first satellite satisfies that the time interval between two adjacent rounds in which the same satellite is the first satellite is greater than or equal to the satellite protection time. The satellite protection time is the countdown of the corresponding satellite that needs to exit high-intensity calculation and distribution due to temperature control or channel resource control. The allocation optimization unit is used to solve the computational task allocation and corresponding transmission path in each iteration using a genetic algorithm, so as to optimize the completion time of each round calculated by the objective function of multiple iterations, and output the task allocation of each satellite in each iteration, the estimated completion time, and the overall optimal completion time of the multi-round iteration tasks.

[0065] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A distributed iterative computation method based on inter-satellite collaboration and primary star rotation, characterized in that, include, Step S1: In response to the distributed iterative computing task request, each satellite exchanges satellite status data, including satellite orbital position, communication forwarding delay, nominal computing power, used load and satellite protection time, through heartbeat packets, and generates a dynamic resource status table for the entire network based on the interactive information formed by the satellite status data of neighboring satellites. Step S2: Based on the dynamic resource status table of the entire network, construct an inter-satellite link graph model, determine the shortest communication delay between any two of the satellites, and generate a communication delay cost matrix; Step S3: Using the communication delay cost matrix and the nominal computing power and used load of each satellite, allocate the single-round iteration task according to the smallest indivisible task unit to obtain the task amount, expected completion time and corresponding communication path of each satellite, and generate a joint task allocation matrix. Step S4: Based on the joint task allocation matrix and the satellite protection time of each satellite, determine the first satellite and the summative satellite for each round of continuous iteration, generate the first satellite dynamic rotation sequence, and output the task allocation results and completion time of the multi-round iteration tasks.

2. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 1, characterized in that, Step S1 includes, In response to the distributed iterative computing task request, the system obtains satellite status data for each satellite participating in the distributed iterative computing, including the satellite orbital position, communication forwarding delay, nominal computing power, used load, and satellite protection time. The satellite orbital position represents the on-orbit spatial position relationship of each satellite; the communication forwarding delay represents the communication overhead of the inter-satellite link; the nominal computing power represents the baseline computing capability of each satellite; the used load represents the proportion of computing resources currently occupied by each satellite; and the satellite protection time represents the minimum time interval between two consecutive times the same satellite serves as the lead satellite. The network-wide dynamic resource status table is generated based on the obtained satellite orbital position, communication forwarding delay, nominal computing power, used load, and satellite protection time.

3. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 2, characterized in that, Step S2 includes, The satellite orbital positions of each satellite are obtained based on the dynamic resource status table of the entire network, and the free space transmission delay between any two satellites is calculated based on the satellite orbital positions using the electromagnetic wave space propagation model. The communication forwarding delay of each of the satellites is obtained and superimposed on the free space transmission delay to obtain the comprehensive communication delay of the corresponding inter-satellite link; The inter-satellite link graph model is constructed based on the comprehensive communication delay of the inter-satellite links. , where the vertex For each satellite node, edge The weights are the overall communication delays corresponding to the inter-satellite links; Based on the inter-satellite link graph model The Dijkstra algorithm is used to traverse and calculate the shortest communication delay and corresponding path between any two satellite nodes, and the communication delay cost matrix is ​​generated. .

4. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 3, characterized in that, Step S3 includes, Step S301: Set the first star and the summing star in the current iteration round; Step S302: Based on the communication delay cost matrix, the nominal computing power of each satellite and the used load, calculate the inherent link cost delay of each candidate satellite and the estimated completion time after allocating the next smallest indivisible task unit; Step S303: Allocate the single-round iteration tasks sequentially according to the smallest indivisible task unit, select the candidate satellite with the shortest expected completion time as the carrier satellite, and update the task quantity and expected completion time of the carrier satellite after each allocation. At the same time, record the shortest communication path from the first satellite to the carrier satellite and from the carrier satellite to the summarizing satellite for each single-round iteration task. Step S304: Repeat step S303 until the total task allocation for the current iteration round is completed, and generate the joint task allocation matrix based on the task amount of each participating satellite, the estimated completion time, and the corresponding shortest communication path; the completion time of the current iteration round is the maximum value among the estimated completion times of each participating satellite.

5. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 4, characterized in that, Step S302 includes, Obtain the shortest communication delay from the first satellite to each of the candidate satellites, and the shortest communication delay from the candidate satellites to the aggregate satellite from the communication delay cost matrix; For each candidate satellite, calculate the inherent link cost delay required to receive the smallest indivisible task unit in the current iteration round and return it to the aggregate satellite, using the formula: ， in, The inherent link cost latency, The first star mentioned above. For the aforementioned summary star, For the satellites participating in the computational task, the subscript... This indicates the link transmission-related delay, and the arrow indicates the direction of data transmission. and The first satellite obtained by querying the communication delay cost matrix To the satellite and the satellite To the summary star The aforementioned shortest communication delay; Based on the satellite The inherent link cost latency, the nominal computing power, and the used load are used to calculate the estimated completion time, and the formula is as follows: ； ； in, For the satellite The computation time for processing the smallest indivisible task unit. The computational cost for a single task. For the satellite The current load is already in use. For the satellite The nominal computing power mentioned above, The inherent link cost latency, For the satellite The number of satellite missions already allocated in the current iteration round, For the satellite The estimated completion time after the current iteration round is assigned to the next smallest indivisible task unit.

6. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 5, characterized in that, Step S303 includes, The computing tasks are distributed one by one from the task pool of the current iteration round according to the smallest indivisible task unit. The estimated completion times of all the satellites are compared, and the satellite with the minimum estimated completion time is selected as the carrier satellite. The minimum indivisible task unit is assigned to the selected carrier satellite, the satellite task load and the estimated completion time of the carrier satellite are updated in real time, and the shortest communication path corresponding to the shortest communication delay from the first satellite to the satellite and from the satellite to the aggregate satellite is recorded for the assigned minimum indivisible task unit.

7. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 6, characterized in that, Step S4 includes, Step S401: Obtain the set of multi-round iterative tasks and the set of satellites participating in each iteration. And read the corresponding joint task allocation matrix, where, For the total number of rounds, For the set of multi-round iterative tasks, For the first The task matrix for each iteration ; For the first The set of satellites participating in the calculation in each iteration, For the first The first round of calculations involved in the iteration The aforementioned satellites, For the first The total number of satellites involved in the calculation during each iteration; Step S402: Initially determine the candidate first star for each iteration. and the aforementioned summary star Constructing inter-satellite link topology And the satellite set that participated and the task matrix Related; Step S403: Establish a multi-round iterative objective function, calculate the completion time of each round of iteration, and generate a dynamic rotation sequence of the first satellite for consecutive multi-round iterations based on the calculation results. The dynamic rotation sequence of the first satellite satisfies that the time interval between two adjacent rounds in which the same satellite is the first satellite is greater than or equal to the satellite protection time. The satellite protection time is the countdown corresponding to the satellite needing to exit high-intensity calculation and distribution due to temperature control or channel resource control. Step S404: Use a genetic algorithm to solve for the amount of computational tasks allocated and the corresponding transmission path in each iteration, so as to optimize the completion time of each round calculated by the objective function of the multi-round iteration, and output the amount of tasks allocated to each satellite in each iteration, the expected completion time, and the overall optimal completion time of the multi-round iteration tasks.

8. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 7, characterized in that, In step S403, a multi-round iterative objective function is established, and the completion time of each round of iteration is calculated, including: ； in, The objective function value of the multi-round iterative task. The total number of rounds required in the multi-round iterative task. For the first The completion time of the task in each iteration round, For the first The set of satellites participating in the computational task during each round of iteration. For the first The inter-satellite link topology of the round iteration, As a binary decision variable, when the satellite The value is 1 when the satellite is allocated a task slot, and 0 otherwise, to determine whether the satellite participates in this round of tasks. For the first star Transmitted to the satellite via the inter-satellite link The aforementioned shortest communication latency, For the satellite Transmitted to the aggregation star via the inter-satellite link The aforementioned shortest communication latency, For the satellite The computation time required to complete the assigned computation task. The constraints are those for the objective function in the multiple iterations. , The first Wheel and First The time it takes for the first satellite to complete its mission and send back the results. For the satellite in the j-th iteration The communication path identifier corresponding to the inter-satellite link topology. The satellite protection time is the minimum time interval that the same satellite must satisfy between completing one round of task distribution and result aggregation as the lead satellite and participating in task scheduling as the lead satellite again.

9. The distributed iterative computation method based on inter-satellite coordination and primary star rotation according to claim 7, characterized in that, Step S404 includes: Based on the joint task assignment matrix, the first star of each round in consecutive iterations and the summation star of the last round are encoded into individual strings of the genetic algorithm. , of which each For the first The first star in the round of iteration, For the final iteration of the summary star, when the first After the first satellite and the summary satellite of the round of iteration are confirmed, the task volume and corresponding communication path of each satellite in the round of iteration are determined based on the joint task allocation matrix; Generate an initial population, wherein the initial population includes Each individual, the first star of each individual in the first round is confirmed by the task initiation amount in the distributed iterative computing task request; for the second round to the... The first star of the wheel is set to randomly initialize the number of individuals. Set the number of limited candidate initial individuals to And satisfy For numbers 1 to The individual, randomly selects the first satellite of the corresponding round from the currently available satellite set, for the satellite numbered... to The individual, according to the joint task allocation matrix, randomly selects the first satellite of the corresponding round from the K satellites with the shortest communication delay with the first satellite of the previous round; The completion time of each individual in the multi-round iteration is calculated based on the objective function of the multi-round iteration, and the fitness value of each individual is determined in combination with the satellite protection time constraint of the first satellite in each round. Genetic algorithm iterative operation is performed on the initial population, including elite retention and roulette wheel selection based on the fitness values ​​of each individual, two-point crossover of offspring individuals to generate new individuals, and greedy replacement operation is performed on the individuals that trigger mutation. In each iteration, the fitness value of the individual is calculated, and the iteration is repeated until the convergence condition is met. The individual with the highest fitness value is selected as the first star dynamic rotation sequence. The convergence condition is any one of the following: reaching a preset maximum number of iterations, the change in the fitness of the best individual in consecutive iterations being less than a preset threshold, or the fitness value of the individual reaching a preset optimal value. Based on the selected first satellite dynamic rotation sequence and the joint task allocation matrix, the task allocation, the estimated completion time, and the corresponding communication path for each satellite in each iteration are determined, and the overall optimal completion time for the multi-round iteration tasks is output.

10. A distributed iterative computing system based on inter-satellite collaboration and primary star rotation, employing the distributed iterative computing method based on inter-satellite collaboration and primary star rotation as described in any one of claims 1 to 9, characterized in that, include: The situation awareness module is used to respond to distributed iterative computing task requests. Each satellite exchanges satellite status data, including satellite orbit position, communication forwarding delay, nominal computing power, used load and satellite protection time, through heartbeat packets. Based on the interactive information formed by the satellite status data of neighboring satellites, a dynamic resource situation table of the entire network is generated. The topology modeling module is used to construct an inter-satellite link graph model based on the dynamic resource status table of the entire network, determine the shortest communication delay between any two of the satellites, and generate a communication delay cost matrix. The task allocation module is used to allocate single-round iterative tasks according to the smallest indivisible task unit based on the communication delay cost matrix, the nominal computing power of each satellite, and the used load, to obtain the task amount, expected completion time and corresponding communication path of each satellite, and generate a joint task allocation matrix. The result output module is used to determine the first satellite and the sum of satellites in each round of continuous iteration based on the joint task allocation matrix and the satellite protection time of each satellite, generate the first satellite dynamic rotation sequence, and output the task allocation results and completion time of the multi-round iteration tasks.

11. The distributed iterative computing system based on inter-satellite collaboration and primary star rotation according to claim 10, characterized in that, The result output module includes: The task management unit is used to obtain a set of tasks for multiple iterations. and the set of satellites participating in each iteration. And read the corresponding joint task allocation matrix, where, For the total number of rounds, For the set of multi-round iterative tasks, For the first The task matrix for each iteration ; For the first The set of satellites participating in the calculation in each iteration, For the first The first round of calculations involved in the iteration The aforementioned satellites, For the first The total number of satellites involved in the calculation during each iteration; The target computation unit is used to initially determine the candidate first star for each iteration. And Summary Star Constructing inter-satellite link topology And the satellite set that participated and the task matrix Correlation; Establish a multi-round iterative objective function, calculate the completion time of each round of iteration, and generate a dynamic rotation sequence of the first satellite for consecutive multi-round iterations based on the calculation results, wherein the dynamic rotation sequence of the first satellite satisfies that the time interval between two adjacent rounds in which the same satellite is the first satellite is greater than or equal to the satellite protection time, and the satellite protection time is the countdown corresponding to the satellite needing to exit high-intensity calculation and distribution due to temperature control or channel resource control; The allocation optimization unit is used to use a genetic algorithm to solve for the amount of computational tasks allocated and the corresponding transmission path in each iteration, so as to optimize the completion time of each round calculated by the objective function of the multi-round iteration, and output the amount of tasks allocated to each satellite in each round of iteration, the expected completion time, and the overall optimal completion time of the multi-round iteration tasks.

Images