Method and system for accelerating construction of constellation model and storage medium

Abstract

The invention discloses a method and system for accelerating construction of a constellation model and a storage medium. The method comprises the following steps: decomposing an algorithm for constructing a constellation model into an algorithm serial part and an algorithm parallel part according to a preset decomposition mode; coding modification suitable for the CPU process is carried out on the serial part of the algorithm, and parallel coding modification suitable for the AI accelerator process is carried out on the parallel part of the algorithm. And merging and reconstructing the transformed algorithm serial part and algorithm parallel part, and executing the merged and reconstructed algorithm for constructing the constellation model by using an on-board computer carrying an AI accelerator. According to the method, the AI accelerator is used for parallel computing, the constellation model construction algorithm is optimized in parallel, and the number of serial cycles of a CPU is reduced, so that the problem of low operation efficiency of the model construction algorithm is solved, and the acceleration effect is achieved.

Description

Technical Field

[0001] This application relates to the field of spaceborne intelligent computing, specifically to methods, systems, and computer-readable storage media for accelerating the construction of constellation models. Background Technology

[0002] The statements in this section are merely to provide background information in relation to this application to aid in understanding it, and such background information does not necessarily constitute prior art.

[0003] Constellation models hold an irreplaceable and crucial position in navigation and positioning systems, especially in Global Navigation Satellite Systems (GNSS), where their design and optimization directly impact positioning accuracy. The core of a navigation and positioning system lies in achieving high-precision positioning of ground targets through the collaborative work of multiple satellites, and the constellation model design determines the spatial distribution and coverage performance of these satellites. In the field of satellite communications, the construction and optimization of constellation models play a significant role in improving the performance of communication links. With the rapid development of aerospace technology and cloud computing concepts, cloud-network converged networks based on mega-constellations are gradually becoming an important component of integrated space-ground construction and 6G communications. The highly dynamic, time-varying characteristics and resource heterogeneity of constellation networks pose significant challenges to the optimization of communication links. Through rational constellation model design, communication links can be effectively optimized, communication efficiency improved, and transmission latency reduced.

[0004] However, with the continuous expansion of application scenarios, the complexity of constellation models is also rapidly increasing. Modern constellation systems are developing towards higher resolution and multi-dimensionality; for example, the rise of virtual constellations and mega-constellations requires models to handle more variables and higher-dimensional data. However, traditional algorithms often exhibit low computational efficiency and slow convergence speed when facing high-dimensional, multivariate models, making it difficult to meet real-time requirements. For example, in mega-constellation networks, due to the large number of satellites and dynamically changing topology, traditional path planning algorithms and resource allocation strategies are ill-suited to this highly complex scenario, leading to a decline in model performance. In practical applications, constellation models not only need to consider the relative positions of celestial bodies but also incorporate constraints such as time dimensions, environmental factors, and mission requirements. This multi-dimensional comprehensive modeling makes algorithm design more complex and places higher demands on computational resources. Given the current constellation model construction algorithms' low speed and limited optimization space when processing massive amounts of data, a solution to accelerate the construction of on-orbit constellation models is urgently needed. Summary of the Invention

[0005] This application decomposes the algorithm for constructing constellation models and optimizes it in parallel using an AI accelerator, specifically through the following technical solutions: According to a first aspect of this application, a method for accelerating the construction of a constellation model is provided, comprising: decomposing an algorithm for constructing the constellation model into a serial part and a parallel part according to a preset decomposition method; modifying the serial part of the algorithm for CPU processing and modifying the parallel part of the algorithm for AI accelerator processing; merging and reconstructing the modified serial and parallel parts of the algorithm, and executing the merged and reconstructed constellation model construction algorithm using an onboard computer equipped with an AI accelerator.

[0006] Preferably, decomposing the algorithm for constructing the constellation model into a serial part and a parallel part according to a preset decomposition method includes: separating the parallel part of the algorithm based on the characteristics of the parallel optimization object. The characteristics of the parallel optimization object include looping, derivation or aggregation patterns, striping or block-based patterns, and divide-and-conquer patterns.

[0007] Preferably, the algorithm for constructing the constellation model after merging and reconstructing includes: the serial part of the algorithm for constructing the constellation model after merging and reconstructing is executed by the CPU, and the parallel part of the algorithm for constructing the constellation model after merging and reconstructing is executed by the AI ​​accelerator.

[0008] Preferably, the parallel portion of the algorithm for constructing the constellation model after merging and reconstructing, executed by the AI ​​accelerator, includes: allocating storage space in the AI ​​accelerator's video memory for the parallel portion of the constellation model construction algorithm, and storing the starting address of this storage space in the AI ​​accelerator's video memory as a pointer in the CPU; storing the initial orbital data of the satellites used to accelerate the construction of the constellation model in the allocated AI accelerator storage space; allocating the number of threads, and calling the kernel function to execute the parallel portion of the computation according to the number of threads; assigning the computation result of the kernel function from the AI ​​accelerator storage space to the CPU; and releasing all the storage space allocated in the AI ​​accelerator's video memory.

[0009] Preferably, the algorithm for constructing the constellation model includes: reading and initializing constellation data, converting the constellation data into initial orbital data for each satellite in the constellation model, performing orbit prediction for each satellite based on the initial orbital data and the orbit prediction model, and summarizing the single-satellite data obtained from the orbit predictions to construct the constellation model; wherein the serial part of the algorithm includes: reading and initializing constellation data, and constructing the constellation model based on the calculated single-satellite data; the parallel part of the algorithm includes: converting the constellation data into initial orbital data for each satellite in the constellation model, and performing orbit prediction for each satellite based on the initial orbital data and the orbit prediction model.

[0010] Preferably, the algorithm for constructing the constellation model includes: for each satellite, calculating the orbit perturbation term, extrapolating the orbit, and calculating the position and velocity at future times in sequence based on the initial orbit data and the orbit prediction model; wherein, the calculation of the orbit perturbation term belongs to the serial part or parallel part of the algorithm, and the extrapolation of the orbit and the calculation of the position and velocity of the satellite at future times belong to the parallel part of the algorithm.

[0011] Preferably, the calculation of orbital perturbation terms based on initial orbital data and orbital prediction models includes: determining the type of orbital prediction model for the satellites in the constellation model; for the two-body orbital prediction model, the corresponding orbital perturbation term is zero; for the J2 orbital prediction model, the corresponding orbital perturbation term includes the Earth's second-order zonal harmonic coefficient; for the J4 orbital prediction model, the corresponding orbital perturbation term includes the Earth's second-order zonal harmonic coefficient and the fourth-order zonal harmonic coefficient; for the SGP4 orbital prediction model, the corresponding orbital perturbation term includes atmospheric drag perturbation and non-spherical orbital perturbation.

[0012] Preferably, the calculation of perturbation terms for the SGP4 orbit prediction model includes the calculation of long-period perturbation terms and the calculation of short-period perturbation terms; wherein, the calculation of long-period perturbation terms belongs to the serial part of the algorithm, and the calculation of short-period perturbation terms belongs to the parallel part of the algorithm.

[0013] According to a second aspect of this application, a system for accelerating the construction of constellation models is provided, comprising: an algorithm decomposition module, an encoding modification module, and an algorithm reconstruction and execution module; wherein, the algorithm decomposition module is used to decompose the algorithm for constructing the constellation model into a serial part and a parallel part according to a preset decomposition method; the encoding modification module is used to perform encoding modification of the serial part of the algorithm to be suitable for CPU processes, and to perform parallel encoding modification of the parallel part of the algorithm to be suitable for AI accelerator processes; the algorithm reconstruction and execution module is used to merge and reconstruct the modified serial part and parallel part of the algorithm, and to execute the merged and reconstructed algorithm for constructing the constellation model using an onboard computer equipped with an AI accelerator.

[0014] According to a third aspect of this application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any of the first aspects.

[0015] Compared with existing technologies, the advantages of this application are: it utilizes AI accelerators for parallel computing, performs parallel optimization of the constellation model building algorithm, reduces the number of CPU serial loops, thereby solving the problem of low running efficiency of the model building algorithm and achieving the effect of acceleration. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart illustrating a method for accelerating the construction of a constellation model according to an embodiment of this application; Figure 2 This is a schematic diagram of a process for performing serial computation using a CPU according to an embodiment of this application; Figure 3 This is a schematic diagram of a process for performing parallel partial computation using an xPU according to an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided through specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application.

[0018] As mentioned in the background section, current constellation model building algorithms suffer from low processing speed and limited optimization space when handling massive amounts of data. The first step in constellation model building is to convert constellation data into initial orbital data for satellites, and then call the orbit prediction model to calculate the position and velocity of each satellite at a future time (i.e., orbit prediction). Constellation data typically refers to the six orbital parameters (semi-major axis, eccentricity, inclination, right ascension of the ascending node, argument of perigee, and mean perigee angle), orbital epoch, time step, and simulation start and end times. This data usually comes from mission planning documents, ephemeris data, or TLE (Time Limit Exceeded), and serves as the starting point for orbit prediction calculations. Although the orbital model remains consistent within the same constellation model—that is, all satellites use the same type of orbit prediction model—the amount of data processing involved in this process is still very large. After predicting the orbit of each satellite, the orbital prediction results of all satellites in the constellation need to be aggregated to form the temporal distribution pattern of the entire constellation, i.e., the constellation model. The inventors discovered the following main reasons for this problem during their research: Many constellation model construction algorithms have high time complexity, such as those employing brute-force search or nested loops, causing the computational load to increase exponentially or polynomially with the amount of data. Constellation model construction algorithms are often executed serially, failing to fully utilize modern multi-core processors or distributed computing resources, resulting in idle computing resources and low efficiency. Using inappropriate data storage and access methods can lead to frequent memory accesses and inefficient lookup operations, thus slowing down the algorithm.

[0019] To address the aforementioned issues, this invention decomposes the constellation model construction algorithm into a serial and a parallel component. It utilizes an AI accelerator for parallel computation and the CPU executes the serial component to optimize the constellation model construction algorithm in parallel. This reduces the number of CPU serial loops, thereby solving the problem of low operating efficiency and achieving acceleration.

[0020] Figure 1 This is a flowchart illustrating a method for accelerating the construction of a constellation model according to an embodiment of this application. Figure 1 As shown, according to one embodiment of this application, a method for accelerating the construction of a constellation model includes: Step S101, decomposing the algorithm for constructing the constellation model into a serial part and a parallel part according to a preset decomposition method. Step S102, performing CPU-compatible coding transformation on the serial part of the algorithm and AI accelerator-compatible parallel coding transformation on the parallel part of the algorithm. In some embodiments, a compiler is used to convert the code of the serial part of the algorithm into machine code executable by the CPU. CUDA (Compute Unified Device Architecture), i.e., a parallel computing platform and programming model, can be used to perform parallel coding transformation on the parallel part of the algorithm. Step S103, merging and reconstructing the transformed serial and parallel parts of the algorithm, and executing the merged and reconstructed constellation model construction algorithm using an onboard computer equipped with an AI accelerator. In some embodiments, the CPU executes the serial part of the merged and reconstructed constellation model construction algorithm, and the AI ​​accelerator (e.g., GPU) executes the parallel part of the merged and reconstructed constellation model construction algorithm.

[0021] In some embodiments, step S101 specifically includes separating the parallel part of the algorithm based on the characteristics of the parallel optimization object. The characteristics of the parallel optimization object include loops, derivation or aggregation patterns, striping or block patterns, and divide-and-conquer patterns. Loops are a common type of parallel optimization object, improving computational efficiency by having each thread process one or a group of data within the loop. However, it's important to consider the dependencies between parameters or algorithms within the loop, i.e., whether the loop depends on the calculation result of the previous step. Derivation or aggregation patterns refer to a section of serial code that can be processed in parallel. This parallel event is often unpredictable; at a certain point in the serial code execution, different data needs to be processed differently, and the results are then aggregated. In other words, there are no dependencies between the data, allowing for concurrent different calculations. Striping or block patterns refer to the situation where a single, particularly large set of data can be divided into blocks for parallel processing. Divide-and-conquer patterns refer to the decomposition of complex problems; that is, to solve a large, complex problem, it is broken down into several smaller problems, solved separately, and finally the solutions to the smaller problems are aggregated and processed to obtain the solution to the large, complex problem. Recursive algorithms fall into this category.

[0022] In some embodiments, the algorithm for constructing the constellation model adopts a modular design, which mainly includes reading and initializing constellation data (constellation model initialization module), converting constellation data into initial orbital data for each satellite in the constellation model (constellation data to initial orbital data conversion module), performing orbit prediction for each satellite based on the initial orbital data and orbit prediction model (including orbit prediction model initialization module and orbit prediction module), and summarizing the single-satellite data obtained from the orbit prediction to construct the constellation model (constellation model construction module). Satellite orbital data may include orbital root numbers. Single-satellite data includes the position and velocity of each satellite at future times, which involves satellite coordinate transformation and attitude calculation. During the satellite constellation's on-orbit operation, constellation model initialization, constellation data to initial orbital data conversion, and orbit prediction model initialization are prerequisites for constellation model construction. Orbit prediction depends on the orbit prediction model initialization results. From a module relationship perspective, constellation model initialization and constellation data conversion can be performed in parallel. From the perspective of the number of executions, except for constellation model initialization, the number of runs for other modules is proportional to the number of satellites, and the orbit prediction module is also affected by the number of simulation steps. Therefore, the algorithm for constructing the constellation model can be decomposed into two parts: reading and initializing constellation data, and constructing the constellation model based on single-satellite data obtained from orbit predictions are the serial parts of the algorithm. Converting the constellation data into initial orbit data for each satellite in the constellation model, and performing orbit predictions for each satellite based on the initial orbit data and the orbit prediction model, constitute the parallel parts of the algorithm. This reduces the number of CPU serial loops, meeting the high efficiency requirements of on-orbit operation. Onboard computers are typically equipped with AI accelerators such as GPUs or NPUs (collectively referred to as xPUs). This algorithm adopts a modular design, does not rely on a specific instruction set or computing framework, and can be implemented on-orbit on onboard computers equipped with various AI accelerators.

[0023] In some embodiments, orbit prediction is performed for each satellite based on initial orbit data and an orbit prediction model. This includes: for each satellite, sequentially calculating orbit perturbation terms, extrapolating the orbit, and calculating the position and velocity at future times (i.e., between the start and end times of orbit extrapolation) based on the initial orbit data and the orbit prediction model. From the perspective of module relationships, orbit parameter initialization and perturbation term calculation are prerequisites for orbit prediction. From the perspective of the number of executions, except for orbit parameter initialization and perturbation term calculation which only need to be run once, the number of runs for other modules is affected by both the number of simulation steps and the number of satellites. Therefore, in parallel computing, priority is given to the parallel processing of orbit extrapolation, coordinate transformation, and attitude calculation to reduce the number of serial loops in the CPU. Thus, the calculation of orbit perturbation terms belongs to the serial part of the algorithm, while orbit extrapolation and the calculation of the satellite's position and velocity at future times belong to the parallel part of the algorithm. In some embodiments, the calculation of orbit perturbation terms can also be placed in the parallel part of the algorithm.

[0024] In some embodiments, calculating the orbital perturbation term based on initial orbital data and the orbital prediction model includes: first, determining the type of orbital prediction model for the satellites in the constellation model. For the two-body orbital prediction model, the corresponding orbital perturbation term is zero. For the J2 orbital prediction model, the corresponding orbital perturbation term includes the Earth's second-order zonal harmonic coefficients (i.e., the J2 first-order long-term perturbation term). For the J4 orbital prediction model, the corresponding orbital perturbation term includes the Earth's second-order zonal harmonic coefficients and fourth-order zonal harmonic coefficients (i.e., the J2 first-order long-term perturbation term, the J2 second-order long-term perturbation term, and the J4 first-order long-term perturbation term). For the SGP4 orbital prediction model, the corresponding orbital perturbation term includes atmospheric drag perturbation and non-spherical orbital perturbation.

[0025] In some embodiments, the orbit prediction model employs a two-body model. The orbit prediction process includes: calculating two-body perturbation terms based on the six orbital elements, orbital epoch, orbital step size, and the start and end times of orbital extrapolation, along with the two-body model. Orbit extrapolation is then performed based on the two-body perturbation terms to calculate the satellite's position and velocity at future times. The calculation of the two-body perturbation terms can be considered a sequential part of the algorithm, while the extrapolation of the orbit based on the two-body perturbation terms to calculate the satellite's position and velocity at future times can be considered a parallel part.

[0026] In some embodiments, the orbit prediction model adopts the J2 orbit model. The orbit prediction process includes: calculating the J2 first-order long-term perturbation term based on the average roots, orbit epoch, orbit extrapolation start time, and the J2 orbit model; calculating the rate of change of the six orbital roots based on the J2 first-order long-term perturbation term; calculating the conversion of the average roots to instantaneous roots based on the rate of change of the six orbital roots; and calculating the future position and velocity of the satellite based on the instantaneous roots and the orbit extrapolation end time. The above process can be considered part of the parallel algorithm.

[0027] In some embodiments, the orbit prediction model adopts the J4 orbit model. The orbit prediction process includes: calculating the J2 first-order long-term perturbation term, the J2 second-order long-term perturbation term, and the J4 first-order long-term perturbation term based on the average root number, orbit epoch, orbit extrapolation start time, and the J4 orbit model; calculating the rate of change of the six orbital roots based on the J2 first-order long-term perturbation term, the J2 second-order long-term perturbation term, and the J4 first-order long-term perturbation term; calculating the conversion of the average root number to the instantaneous root number based on the rate of change of the six orbital roots; and calculating the future position and velocity of the satellite based on the instantaneous root number and the orbit extrapolation end time. The above process can be considered part of the parallel algorithm.

[0028] In some embodiments, the orbit prediction model uses the SGP4 orbit model. The orbit prediction process includes: calculating the initial average velocity and semi-major axis, atmospheric drag perturbations, and non-spherical orbit perturbations based on two-line orbit data (TLE), orbit epochs, the start time of orbit extrapolation, and the SGP4 orbit model; calculating long-period and short-period perturbation terms, i.e., calculating the periodic perturbation terms of the satellite orbital elements as a function of time, where the long-period perturbation term is the perturbation term with a period greater than the satellite orbital period, and the short-period perturbation term is the perturbation term with a period less than or equal to the satellite orbital period; and calculating the future position and velocity of the satellite based on the calculated perturbations and the end time of orbit extrapolation. The above process can be considered part of the parallel algorithm. In some embodiments, the calculation of the long-period perturbation term of the SGP4 orbit prediction model is part of the serial algorithm, while the calculation of its short-period perturbation term is part of the parallel algorithm.

[0029] like Figure 2 As shown, in some embodiments, the serial part of the algorithm for constructing the constellation model after merging and reconstructing is executed by the CPU. This includes reading in orbit prediction parameters; converting the orbit prediction parameters into initial orbit data of the satellites in orbit; initializing various variables, including declaring and assigning values ​​to parameters interacting with the AI ​​accelerator; exchanging orbit determination parameters between satellites, including mutually transmitting and sharing data used for orbit determination (e.g., satellite orbit data, clock data) to utilize this data for calculations and improve the accuracy of their own orbit data; after the AI ​​accelerator performs parallel orbit prediction calculations, the CPU reads the calculation results and then constructs the constellation model based on the single-satellite data in the results. It is evident that using the AI ​​accelerator to participate in orbit extrapolation, coordinate transformation, and attitude calculation can effectively reduce the total number of serial loops in the CPU, simplifying the CPU process.

[0030] In some embodiments, the parallel portion of the algorithm for constructing the constellation model after merging and reconstructing, executed by the AI ​​accelerator, includes: after the CPU initializes and assigns values ​​to relevant variables and calculates perturbation terms (two-body orbit model), allocating storage space in the AI ​​accelerator's video memory for the parallel portion of the constellation model construction algorithm, and storing the starting address of this storage space as a pointer in the CPU. The initial orbital data of the satellites used to accelerate constellation model construction is stored in the allocated AI accelerator storage space. The number of threads is allocated, and the kernel function for executing the parallel portion of the computation is called according to the thread order. The computation result of the kernel function is assigned from the AI ​​accelerator storage space to the CPU. All allocated storage space in the AI ​​accelerator's video memory is released. This embodiment improves the efficiency of memory access and lookup operations by rationalizing the storage and access of data between the CPU and the AI ​​accelerator, thereby accelerating the execution of the algorithm.

[0031] like Figure 3As shown, the specific calculation steps performed by xPU are as follows (xPU uses Nvidia GPU as an example): (1) The cudaMalloc function is used to request storage space for satellite orbit prediction in GPU memory according to the size of orbit data, and the starting address of the requested GPU memory space is stored in the CPU as a pointer.

[0032] (2) The initial satellite orbit data is stored in the requested GPU storage space using the cudaMemcpy function.

[0033] (3) Allocate the number of threads according to the data size and hardware limitations, and call the kernel function for orbit prediction calculation according to the thread order.

[0034] (4) The satellite orbit calculation results are transferred from the GPU storage space to the CPU using the cudaMemcpy function.

[0035] Based on the same inventive concept, one embodiment of this application provides a system for accelerating the construction of constellation models, comprising: an algorithm decomposition module, an encoding modification module, and an algorithm reconstruction and execution module. The algorithm decomposition module decomposes the algorithm for constructing the constellation model into a serial part and a parallel part according to a preset decomposition method. The encoding modification module performs CPU-compatible encoding modification on the serial part and AI accelerator-compatible parallel encoding modification on the parallel part. The algorithm reconstruction and execution module merges and reconstructs the modified serial and parallel parts, and executes the merged and reconstructed constellation model construction algorithm using an onboard computer equipped with an AI accelerator.

[0036] This application embodiment utilizes an AI accelerator for parallel computing to optimize the constellation model building algorithm in parallel, reducing the number of CPU serial loops and thus solving the problem of low running efficiency of the model building algorithm, thereby achieving the effect of acceleration.

[0037] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0038] The various embodiments in this application are described in a progressive manner, with each embodiment focusing on the differences from other embodiments or implementation methods. Similar or identical parts between the various embodiments of this application can be referred to mutually. The implementation principles and technical effects of the inventive concept can be mutually referenced, and will not be repeated here. Where there is no conflict, the various embodiments or implementation methods in this application can be combined with each other.

[0039] It should be noted that although the steps are described in a specific order above, it does not mean that the steps must be executed in the above specific order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required function can be achieved.

[0040] This application may be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this application.

[0041] Computer-readable storage media can be tangible devices that hold and store instructions for use by an instruction execution device. Computer-readable storage media can include, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof.

[0042] This application uses specific embodiments to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the solution and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for accelerating the construction of constellation models, wherein, include: The algorithm for constructing the constellation model is decomposed into a serial part and a parallel part according to the preset decomposition method. The serial part of the algorithm is modified to be adapted to CPU processing flow, and the parallel part of the algorithm is modified to be adapted to AI accelerator processing flow. The serial and parallel parts of the modified algorithm are merged and reconstructed, and the merged and reconstructed algorithm for building the constellation model is executed by the onboard computer equipped with an AI accelerator.

2. The method according to claim 1, wherein, The algorithm for constructing the constellation model is decomposed into a serial part and a parallel part according to the preset decomposition method, including: separating the parallel part of the algorithm according to the characteristics of the parallel optimization object of the algorithm; The characteristics of parallel optimization objects include loop, derivation or aggregation mode, strip or block mode, and divide and conquer mode.

3. The method according to claim 1, wherein, The algorithm for constructing the constellation model after merging and refactoring includes: The serial part of the algorithm for constructing the constellation model after merging and reconstructing is executed by the CPU, while the parallel part of the algorithm for constructing the constellation model after merging and reconstructing is executed by the AI ​​accelerator.

4. The method according to claim 3, wherein, The parallel components of the algorithm for constructing the constellation model after merging and reconstructing, executed by the AI ​​accelerator, include: In the AI ​​accelerator memory, allocate storage space for the parallel part of the algorithm that builds the constellation model, and store the first address of this storage space in the AI ​​accelerator memory as a pointer in the CPU. The initial orbital data of the satellites used to accelerate the construction of the constellation model will be stored in the applied AI accelerator storage space; Allocate the number of threads and call the kernel function that performs the parallel computation according to the number of threads; The calculation results of the kernel function are assigned from the AI ​​accelerator storage space to the CPU; Release all storage space allocated in the AI ​​accelerator memory.

5. The method according to claim 1, wherein, The algorithm for constructing the constellation model includes: reading and initializing constellation data; converting the constellation data into initial orbital data for each satellite in the constellation model; performing orbit prediction for each satellite based on the initial orbital data and the orbit prediction model; and summarizing the single-satellite data obtained from the orbit predictions to construct the constellation model; and The sequential part of the algorithm includes: reading and initializing constellation data, and constructing a constellation model based on the calculated single-star data; The parallel part of the algorithm includes: converting constellation data into initial orbit data for each satellite in the constellation model, and performing orbit prediction for each satellite based on the initial orbit data and the orbit prediction model.

6. The method according to claim 1, wherein, The algorithm for constructing the constellation model includes: for each satellite, calculating the orbital perturbation term, extrapolating the orbit, and calculating the position and velocity at future times based on the initial orbital data and the orbital prediction model; The calculation of the orbital perturbation term belongs to the serial or parallel part of the algorithm, while the calculation of orbital extrapolation and the satellite's position and velocity at future moments belong to the parallel part of the algorithm.

7. The method according to claim 6, wherein, The calculation of orbital perturbation terms based on initial orbital data and orbital prediction models includes: Determine the type of satellite orbit prediction model in the constellation model; For the two-body orbit prediction model, the corresponding orbit perturbation term is zero; For the J2 orbit prediction model, the corresponding orbital perturbation term includes the Earth's second-order zonal harmonic coefficients; For the J4 orbit prediction model, the corresponding orbital perturbation terms include the Earth's second-order zonal harmonic coefficients and fourth-order zonal harmonic coefficients; For the SGP4 orbit prediction model, the corresponding orbital perturbation terms include atmospheric drag perturbation and non-spherical orbital perturbation.

8. The method according to claim 7, wherein, The calculation of perturbation terms for the SGP4 orbit prediction model includes the calculation of long-period perturbation terms and the calculation of short-period perturbation terms; The calculation of long-period perturbation terms belongs to the serial part of the algorithm, while the calculation of short-period perturbation terms belongs to the parallel part of the algorithm.

9. A system for accelerating the construction of constellation models, wherein, include: The algorithm consists of three modules: an algorithm decomposition module, an encoding modification module, and an algorithm reconstruction and execution module; among them, The algorithm decomposition module is used to decompose the algorithm for building the constellation model into a serial part and a parallel part according to a preset decomposition method. The encoding modification module is used to modify the serial part of the algorithm to be compatible with the CPU process and to modify the parallel part of the algorithm to be compatible with the AI ​​accelerator process. The algorithm reconstruction and execution module is used to merge and reconstruct the serial and parallel parts of the modified algorithm, and then use the onboard computer equipped with an AI accelerator to execute the merged and reconstructed algorithm for building the constellation model.

10. A computer-readable storage medium, wherein, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-8.

Images