Model training system, method and equipment based on space-based distributed AI computing power and medium

Abstract

The invention provides a space-based distributed AI computing power-based model training system, method, device and medium, the system comprises a plurality of satellites, a computing power unit and an AI algorithm deployment unit are deployed in each satellite, and the computing power unit comprises a distributed computing power distribution module and a high-speed interconnection AI computing power module; the first AI algorithm deployment unit responds to an AI algorithm release task sent by the ground user equipment, and sends a target AI algorithm to the computing power unit of each satellite; the first computing power unit responds to the algorithm model training task, and distributes a target computing power unit for the algorithm model training task according to the algorithm model through a distributed computing power distribution module of the first computing power unit; and each target computing power unit executes the algorithm model training task based on the target AI algorithm through the high-speed interconnection AI computing power module of the target computing power unit, and sends an execution result of the algorithm model training task to the ground user equipment. By adopting the scheme, the waste of computing power resources is avoided, and the training efficiency of model training is improved.

Description

Technical Field

[0001] The present invention relates to the field of energy storage technology, and in particular to a model training system, method, device and medium based on space-based distributed AI computing power. Background Art

[0002] With the rapid development of artificial intelligence technology, the demand for computing power for model training is increasing. Traditional ground computing centers often face problems such as insufficient computing power, long training time, and low energy efficiency when facing large-scale, highly complex algorithm model training. In order to solve these problems, the industry has begun to explore the use of space resources, especially low-orbit satellite networks, to build a space-based distributed AI computing platform. This platform can use the wide-area coverage, low orbit, and low latency of satellites to achieve global computing resource sharing and optimized configuration.

[0003] When existing space-based distributed AI computing platforms perform model training tasks, computing resources are usually fixedly allocated to specific tasks or users, which results in computing resources often not being fully utilized when facing large-scale, high-complexity algorithm model training tasks. Some computing resources may be idle due to uneven task distribution or mismatch between task requirements and computing resources, resulting in a waste of resources. In addition, existing task execution usually relies on a single computing resource or a limited computing cluster to complete. Although this task execution mode may still meet the needs when processing small-scale, low-complexity algorithm model training tasks, it will significantly prolong the task execution time and reduce the task execution efficiency when facing large-scale, high-complexity tasks. In addition, due to the limitation of computing resources, it may not be able to meet the needs of real-time or high-performance computing, thus affecting the rapid development and widespread application of AI technology. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a model training system, method, device and medium based on space-based distributed AI computing power to avoid waste of computing resources and improve the training efficiency of model training.

[0005] In a first aspect, an embodiment of the present application provides a model training system based on space-based distributed AI computing power, the system comprising a number of satellites, each of which is deployed with a number of computing power units and an AI algorithm deployment unit, each computing power unit comprising a distributed computing power allocation module and a number of high-speed interconnected AI computing power modules; A first AI algorithm deployment unit, configured to respond to an AI algorithm publishing task sent by a ground user equipment and send a target AI algorithm included in the AI ​​algorithm publishing task to a computing unit of each satellite; A first computing power unit is used to respond to the algorithm model training task sent by the ground user equipment, and allocate a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module; Each target computing power unit is used to execute the algorithm model training task based on the target AI algorithm through its high-speed interconnected AI computing power module, and send the execution result of the algorithm model training task to the ground user equipment.

[0006] Optionally, each satellite also includes a multi-directional inter-satellite data transmission unit and a satellite-to-ground communication unit; The satellite-to-ground communication unit is used to realize communication and data transmission between the computing unit and the ground user equipment; The multi-directional inter-satellite data transmission unit is used to realize communication and data transmission between satellites.

[0007] Optionally, each computing power unit also includes a cascade optical communication module; The cascade optical communication module is used to realize communication and data transmission between various computing power units.

[0008] Optionally, the allocating a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module includes: Based on the ground cloud computing platform and container technology, the computing resources of each computing unit are virtualized into multiple virtual resource pools; The virtual resources are dynamically allocated according to the resource conditions of each virtual resource pool and the requirements of the algorithm model training task, and the computing environment is quickly deployed through container technology to achieve target computing power unit allocation.

[0009] Optionally, the high-speed interconnect AI computing power module includes: High-performance SOC processor with ARM+GPU+NPU as the core; The interface uses 15Gbps×8 high-speed differential transmission channels; Equipped with 32GB memory system, 1TB data storage, 64GB computing program storage and 256MB boot configuration storage.

[0010] Optionally, the multi-directional inter-satellite data transmission unit includes: The interface uses 15Gbps×4 high-speed differential transmission channels; Realize collaborative task allocation and scheduling among satellites based on advanced communication protocols.

[0011] Optionally, the cascade optical communication module includes: Dual-port wired fiber channels, each channel contains 4-way duplex 10Gbps×4 fiber channels; The interface uses 15Gbps×8 high-speed differential transmission channels for high-speed interconnection and data distribution between modules.

[0012] In a second aspect, an embodiment of the present application provides a model training method based on space-based distributed AI computing power, which is applied to a model training system based on space-based distributed AI computing power, wherein the system includes several satellites, each of which is deployed with several computing power units and AI algorithm deployment units, each computing power unit includes a distributed computing power allocation module and several high-speed interconnected AI computing power modules, and the method includes: The first AI algorithm deployment unit responds to the AI ​​algorithm publishing task sent by the ground user equipment, and sends the target AI algorithm included in the AI ​​algorithm publishing task to the computing power unit of each satellite; The first computing power unit responds to the algorithm model training task sent by the ground user equipment, and allocates a target computing power unit to the algorithm model training task through its distributed computing power allocation module according to the algorithm model included in the algorithm model training task; Each target computing unit executes the algorithm model training task based on the target AI algorithm through its high-speed interconnected AI computing module, and sends the execution result of the algorithm model training task to the ground user equipment.

[0013] Optionally, each satellite further includes a multi-directional inter-satellite data transmission unit and a satellite-to-ground communication unit, and the method further includes: The satellite-to-ground communication unit realizes communication and data transmission between the computing unit and the ground user equipment; The multi-directional inter-satellite data transmission unit realizes communication and data transmission among satellites.

[0014] Optionally, each computing unit further includes a cascade optical communication module, and the method further includes: The cascade optical communication module realizes communication and data transmission between various computing power units.

[0015] Optionally, the allocating a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module includes: Based on the ground cloud computing platform and container technology, the computing resources of each computing unit are virtualized into multiple virtual resource pools; The virtual resources are dynamically allocated according to the resource conditions of each virtual resource pool and the requirements of the algorithm model training task, and the computing environment is quickly deployed through container technology to achieve target computing power unit allocation.

[0016] Optionally, the high-speed interconnect AI computing power module includes: High-performance SOC processor with ARM+GPU+NPU as the core; The interface uses 15Gbps×8 high-speed differential transmission channels; Equipped with 32GB memory system, 1TB data storage, 64GB computing program storage and 256MB boot configuration storage.

[0017] Optionally, the multi-directional inter-satellite data transmission unit includes: The interface uses 15Gbps×4 high-speed differential transmission channels; Realize collaborative task allocation and scheduling among satellites based on advanced communication protocols.

[0018] Optionally, the cascade optical communication module includes: Dual-port wired fiber channels, each channel contains 4-way duplex 10Gbps×4 fiber channels; The interface uses 15Gbps×8 high-speed differential transmission channels for high-speed interconnection and data distribution between modules.

[0019] In a third aspect, an embodiment of the present application provides a computer device, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the model training method based on space-based distributed AI computing power described in any optional implementation manner of the second aspect are performed.

[0020] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the model training method based on space-based distributed AI computing power described in any optional implementation manner of the second aspect above are executed.

[0021] The technical solution provided by this application includes but is not limited to the following beneficial effects: The model training system based on space-based distributed AI computing power provided by this application is responsible for efficiently processing the AI ​​algorithm release instructions sent by the ground user equipment through the first AI algorithm deployment unit, quickly parsing the instruction content, and safely and accurately distributing the target AI algorithm to the computing nodes on each satellite, ensuring the consistency management of the algorithm version, and improving the accuracy and efficiency of system deployment. Through the distributed computing power scheduling algorithm built into the first computing power unit, it can intelligently match and allocate the optimal computing power resources for the algorithm model training task based on the specific needs of the algorithm model training task. Through the dynamic scheduling and optimal configuration of global computing power resources, it not only improves the utilization efficiency of computing power resources, but also realizes the elastic expansion and load balancing of computing power, ensuring the smooth completion of large-scale and high-complexity tasks. Through each target computing power unit using its high-performance AI computing power module, according to the assigned target AI algorithm, the algorithm model training task is efficiently executed, and the training results are sent back to the ground user equipment in a standardized format after the task is completed, which not only improves the overall performance of the system, but also enhances the user experience and satisfaction.

[0022] In summary, the model training system based on space-based distributed AI computing power realizes efficient management, intelligent scheduling and rapid response of computing power resources through the collaborative work of computing power units deployed in each satellite and AI algorithm deployment units, providing stable and efficient computing power support for AI model training, avoiding waste of computing power resources and improving the training efficiency of model training.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A schematic diagram of the structure of a model training system based on space-based distributed AI computing power provided in the first embodiment of the present invention is shown; Figure 2 A schematic diagram of the structure of a second model training system based on space-based distributed AI computing power provided in the first embodiment of the present invention is shown; Figure 3 A schematic diagram of the structure of a third model training system based on space-based distributed AI computing power provided in the first embodiment of the present invention is shown; Figure 4A flow chart of a model training method based on space-based distributed AI computing power provided in Embodiment 2 of the present invention is shown; Figure 5 A schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention is shown. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0027] Embodiment 1 To facilitate understanding of this application, Figure 1 The content described in the structural schematic diagram of a model training system based on space-based distributed AI computing power provided by the first embodiment of the present invention is described in detail.

[0028] See also Figure 1 As shown, Figure 1 A structural schematic diagram of a model training system based on space-based distributed AI computing power provided by an embodiment of the present invention is shown, wherein the system includes a plurality of satellites 1, each of which is deployed with a plurality of computing power units 11 and an AI algorithm deployment unit 12, and each computing power unit 11 includes a distributed computing power allocation module 111 and a plurality of high-speed interconnected AI computing power modules 112.

[0029] Specifically, the entire system is built on a constellation of multiple satellites, which act as computing nodes and work together to form an extensive computing network. Each satellite is equipped with a computing unit and an AI algorithm deployment unit, providing the necessary hardware and software support for the training of AI models.

[0030] The first AI algorithm deployment unit is used to respond to the AI ​​algorithm publishing task sent by the ground user equipment and send the target AI algorithm included in the AI ​​algorithm publishing task to the computing power unit of each satellite.

[0031] Specifically, the AI ​​algorithm deployment unit is responsible for the transmission of the algorithm. When the ground user device needs to release a new AI algorithm, it will send an AI algorithm release task to the system. After receiving the task, the AI ​​algorithm deployment unit will distribute the target AI algorithm to the computing power unit of each satellite. In this way, no matter where the user is located, as long as there is satellite coverage, they can enjoy the computing power support provided by the system.

[0032] The first computing power unit is used to respond to the algorithm model training task sent by the ground user equipment, and allocate a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module.

[0033] Specifically, the computing unit is the computing core of the system, which includes a distributed computing allocation module and multiple interconnected AI computing modules. The distributed computing allocation module rationally allocates computing resources according to the algorithm model training task requirements sent by the ground user equipment. The interconnected AI computing modules are responsible for executing these tasks and have powerful computing capabilities to process complex AI algorithms.

[0034] Each target computing power unit is used to execute the algorithm model training task based on the target AI algorithm through its high-speed interconnected AI computing power module, and send the execution result of the algorithm model training task to the ground user equipment.

[0035] Specifically, when an algorithm model training task arrives, the computing unit allocates appropriate computing resources to it according to the task requirements through the distributed computing allocation module. These computing resources are then started to train the algorithm model based on the target AI algorithm. After the training is completed, they will transmit the execution results back to the ground user equipment through the satellite network to ensure that users can obtain the training results in a timely manner.

[0036] Furthermore, the user first prepares the algorithm model to be trained, training data, or the trained model that needs further reasoning or tuning according to his own needs and sets an expected task completion time so that the platform can better plan and schedule computing resources. Then the user uploads this information (including algorithm model, training data, completion time, etc.) to the satellite that is currently closest and in the overhead state through the satellite-to-ground communication link. After receiving the demand information uploaded by the user, the satellite forwards it to the distributed computing power allocation module of the space-based distributed AI computing power platform. The distributed computing power allocation module accurately evaluates the computing power demand based on the algorithm model, training data, completion time and current computing power resource status (including the idle state and performance of each computing power unit) provided by the user. Based on the evaluation results, the distributed computing power allocation module will decide how many AI computing power modules to start to perform the task, and consider whether to call the computing power resources of other satellites to form a situation of multi-satellite collaborative work, so as to ensure that the task can be completed on time. Once the computing power resources are reasonably allocated, the relevant AI computing power modules will start to perform the task. This may include algorithm training, model tuning or data recognition. During the execution of the task, the AI ​​computing module will monitor the progress and performance of the task in real time to ensure that the task can proceed as expected. When the AI ​​algorithm completes the training or recognizes the corresponding results, these results will be strictly verified and reviewed to ensure their accuracy and reliability. Once the results pass the verification, the platform will immediately distribute the results to the user quickly and accurately through the satellite-to-ground link. Users can obtain the required computing power services in the shortest time to meet their needs.

[0037] The model training system based on space-based distributed AI computing power provided in this application builds a global space-based distributed AI computing power platform through the layout of satellite constellations, computing power allocation and transmission technology, and algorithm deployment mechanism. It not only breaks through the limitations of ground computing resources, but also greatly improves the real-time and accuracy of AI services.

[0038] In an alternative embodiment, see Figure 2 As shown, Figure 2 A structural schematic diagram of a second model training system based on space-based distributed AI computing power provided in Embodiment 1 of the present invention is shown, wherein each satellite 1 also includes a multi-directional inter-satellite data transmission unit 13 and a satellite-to-ground communication unit 14.

[0039] The satellite-to-ground communication unit is used to realize communication and data transmission between the computing unit and the ground user equipment.

[0040] Specifically, the satellite-to-ground communication unit is a bridge connecting the space-based computing power platform and the ground user equipment, realizing communication and data transmission between the computing power unit and the ground user equipment, and ensuring that users can interact with the system in real time.

[0041] The multi-directional inter-satellite data transmission unit is used to realize communication and data transmission between satellites.

[0042] Specifically, the multi-directional intersatellite data transmission unit is an indispensable part of the space-based distributed AI computing platform, which is mainly responsible for realizing communication and data transmission between satellites. Through this unit, data within the satellite network can flow smoothly, and computing resources can be flexibly scheduled between different satellites.

[0043] In an alternative embodiment, see Figure 3 As shown, Figure 3 A structural schematic diagram of the third model training system based on space-based distributed AI computing power provided in the first embodiment of the present invention is shown, wherein each computing power unit also includes a cascaded optical communication module 113.

[0044] The cascade optical communication module is used to realize communication and data transmission between various computing power units.

[0045] Specifically, in each computing unit of the space-based distributed AI computing platform, the cascade optical communication module is an important component, which is mainly responsible for realizing communication and data transmission between each computing unit. Compared with traditional radio communications, optical communication has higher bandwidth, lower latency and stronger anti-interference ability, which makes it an ideal choice for high-speed and long-distance data transmission. The cascade optical communication module transmits data in free space between satellites by using laser beams as information carriers. It can not only meet the needs of efficient and real-time data transmission between computing units, but also support complex distributed computing and collaborative work scenarios. This feature is crucial to improving the computing efficiency and performance of the entire system.

[0046] In addition, the cascade optical communication module is flexible and scalable. As the number of computing units increases and the distributed computing tasks become more complex, the module can easily cope with it, ensuring that the system always remains in optimal working condition.

[0047] In an optional implementation, the allocating a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module includes: Based on the ground cloud computing platform and container technology, the computing resources of each computing unit are virtualized into multiple virtual resource pools. Virtual resources are dynamically allocated according to the resource conditions of each virtual resource pool and the requirements of the algorithm model training tasks. The computing environment is quickly deployed through container technology to achieve target computing unit allocation.

[0048] Specifically, the allocation of computing power units is completed through its built-in distributed computing power allocation module. This allocation process is highly intelligent and can achieve efficient and flexible scheduling of computing power resources based on ground cloud computing platforms and container technology.

[0049] First, the system virtualizes the computing resources of each computing unit into multiple virtual resource pools. This step uses the virtualization technology of the ground cloud computing platform to abstract physical computing resources into logical virtual resources, thereby realizing resource pooling and unified management. The establishment of virtual resource pools not only improves the utilization rate of computing resources, but also provides a basis for subsequent dynamic allocation.

[0050] Next, the distributed computing power allocation module will dynamically allocate virtual resources based on the resource conditions of each virtual resource pool and the requirements of the algorithm model training task. This process is highly automated and can perceive the remaining capacity of the resource pool, the complexity of the task, and the requirements of the computing environment in real time, so as to make the best allocation decision.

[0051] After determining the target computing unit, the system will quickly deploy the computing environment through container technology. Container technology is a lightweight virtualization technology that can quickly replicate and deploy the same computing environment on different computing nodes, thereby ensuring the consistency and stability of algorithm model training tasks. Through container technology, the system can quickly build the required computing environment on the target computing unit and prepare for the execution of the task.

[0052] In summary, the computing power allocation mechanism of the space-based distributed AI computing power platform fully utilizes the advantages of the ground cloud computing platform and container technology to achieve efficient and flexible scheduling of computing power resources. This mechanism not only improves the computing efficiency and performance of the system, but also provides users with a more convenient and efficient AI service experience.

[0053] In an optional implementation scheme, the high-speed interconnected AI computing power module includes: a high-performance SOC processor with ARM+GPU+NPU as the core; the interface adopts a 15Gbps×8 high-speed differential transmission channel; and is equipped with a 32GB memory system, 1TB data storage, 64GB computing power program storage and 256MB boot configuration storage.

[0054] Specifically, the high-speed interconnected AI computing module uses a high-performance SOC (System on Chip) processor with ARM+GPU+NPU as the core. This combination provides the module with powerful computing power, in which the ARM processor is responsible for general computing tasks, the GPU (graphics processing unit) accelerates graphics processing and parallel computing, and the NPU (neural network processing unit) is specifically used for AI computing tasks such as deep learning. This design enables the module to handle multiple types of computing tasks at the same time, improving the overall performance and flexibility of the system.

[0055] In order to achieve efficient data transmission between modules, the high-speed interconnected AI computing power module adopts a 15Gbps×8 high-speed differential transmission channel. This interface design not only provides high-bandwidth data transmission capabilities, but also ensures data stability and reliability. Through this interface, the module can achieve fast and accurate data exchange with other computing power units or ground user equipment.

[0056] In terms of memory, the high-speed interconnected AI computing power module is equipped with a 32GB memory system. This configuration provides the module with sufficient memory resources to ensure data caching and processing capabilities during the execution of algorithm model training tasks. At the same time, in order to store a large amount of data and algorithm models, the module is also equipped with 1TB of data storage space and 64GB of computing power program storage space. These storage spaces can meet the training needs of algorithm models of different scales and complexities.

[0057] In addition, in order to guide system startup and configuration, the high-speed interconnected AI computing power module also sets up a 256MB boot configuration storage space. This space stores the boot program and configuration files required for system startup, ensuring that the system can correctly load and configure the required hardware and software resources during the startup process.

[0058] In summary, the high-speed interconnected AI computing power module provides powerful computing power support for the space-based distributed AI computing power platform with its high-performance SOC processor, high-speed transmission interface, and sufficient memory and storage configuration. The configuration of this module not only meets the needs of current algorithm model training tasks, but also leaves ample room for future expansion and upgrades.

[0059] In an optional implementation, the multi-directional inter-satellite data transmission unit includes: an interface using a 15Gbps×4 high-speed differential transmission channel; and implementing inter-satellite collaborative task allocation and scheduling based on an advanced communication protocol.

[0060] Specifically, the multi-directional intersatellite data transmission unit uses a 15Gbps×4 high-speed differential transmission channel. This configuration provides an extremely high data transmission rate, ensuring that satellites can exchange data in real time and accurately. The high-speed differential transmission channel not only improves the data transmission efficiency, but also enhances the data's anti-interference ability, making the data more stable and reliable during transmission.

[0061] In order to achieve collaborative task allocation and scheduling between satellites, the multi-directional intersatellite data transmission unit adopts an advanced communication protocol. This protocol can ensure that each satellite maintains a high degree of consistency and coordination during the communication process, thereby achieving efficient task allocation and scheduling. Through this protocol, the system can perceive the status and task requirements of each satellite in real time and make the best decision to ensure the stable operation of the entire system.

[0062] In summary, the multi-directional intersatellite data transmission unit, with its high-speed differential transmission channel and advanced communication protocol configuration, provides efficient and stable data transmission support for the space-based distributed AI computing platform. The configuration of this module not only meets the current needs of inter-satellite data transmission, but also provides a solid foundation for future expansion and upgrades.

[0063] In an optional embodiment, the cascaded optical communication module includes: a dual-port wired fiber channel, each channel contains 4 10Gbps×4 fiber channels with duplex transmission and reception; the interface adopts a 15Gbps×8 high-speed differential transmission channel for high-speed interconnection and data distribution between modules.

[0064] Specifically, the cascade optical communication module is equipped with a dual-port wired fiber channel, each of which contains 4 10Gbps×4 fiber channels with duplex transmission and reception. This configuration provides extremely high data transmission bandwidth, ensuring that large amounts of data can be exchanged in real time and efficiently between computing units. The duplex transmission and reception design allows data to be transmitted in both directions simultaneously, further improving transmission efficiency.

[0065] In addition to the dual-port wired fiber channel, the cascade optical communication module also uses a 15Gbps×8 high-speed differential transmission channel as the interface between modules. This interface design not only provides a higher data transmission rate, but also enhances the stability and anti-interference ability of the data. Through this interface, the cascade optical communication module can achieve high-speed interconnection with other computing power units to ensure the accuracy and integrity of data during transmission.

[0066] In summary, the cascade optical communication module, with its dual-port wired fiber channel and high-speed differential transmission interface configuration, provides powerful and efficient data transmission support for the space-based distributed AI computing platform. The configuration of this module not only meets the current needs of data transmission between computing units, but also provides broad space for future expansion and upgrades.

[0067] Embodiment 2 See also Figure 4 As shown, Figure 4 A flow chart of a model training method based on space-based distributed AI computing power provided in the second embodiment of the present invention is shown, wherein the method is applied to a model training system based on space-based distributed AI computing power, the system includes several satellites, each satellite is deployed with several computing power units and AI algorithm deployment units, each computing power unit includes a distributed computing power allocation module and several high-speed interconnected AI computing power modules, and the method includes steps S401 to S403: S401: In response to the AI ​​algorithm publishing task sent by the ground user equipment, the first AI algorithm deployment unit sends the target AI algorithm included in the AI ​​algorithm publishing task to the computing power unit of each satellite; S402: In response to the algorithm model training task sent by the ground user equipment, the first computing power unit allocates a target computing power unit to the algorithm model training task through its distributed computing power allocation module according to the algorithm model included in the algorithm model training task; S403: Each target computing unit executes the algorithm model training task based on the target AI algorithm through its high-speed interconnected AI computing module, and sends the execution result of the algorithm model training task to the ground user equipment.

[0068] In an optional embodiment, each satellite further includes a multi-directional inter-satellite data transmission unit and a satellite-to-ground communication unit, and the method further includes: The satellite-to-ground communication unit realizes communication and data transmission between the computing unit and the ground user equipment; The multi-directional inter-satellite data transmission unit realizes communication and data transmission among satellites.

[0069] In an optional embodiment, each computing unit further includes a cascade optical communication module, and the method further includes: The cascade optical communication module realizes communication and data transmission between various computing power units.

[0070] In an optional implementation, the allocating a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module includes: Based on the ground cloud computing platform and container technology, the computing resources of each computing unit are virtualized into multiple virtual resource pools; The virtual resources are dynamically allocated according to the resource conditions of each virtual resource pool and the requirements of the algorithm model training task, and the computing environment is quickly deployed through container technology to achieve target computing power unit allocation.

[0071] In an optional implementation scheme, the high-speed interconnect AI computing power module includes: High-performance SOC processor with ARM+GPU+NPU as the core; The interface uses 15Gbps×8 high-speed differential transmission channels; Equipped with 32GB memory system, 1TB data storage, 64GB computing program storage and 256MB boot configuration storage.

[0072] In an optional implementation, the multi-directional inter-satellite data transmission unit includes: The interface uses 15Gbps×4 high-speed differential transmission channels; Realize collaborative task allocation and scheduling among satellites based on advanced communication protocols.

[0073] In an optional embodiment, the cascade optical communication module includes: Dual-port wired fiber channels, each channel contains 4-way duplex 10Gbps×4 fiber channels; The interface uses 15Gbps×8 high-speed differential transmission channels for high-speed interconnection and data distribution between modules.

[0074] Embodiment 3 Based on the same application concept, see Figure 5 As shown, Figure 5 FIG. 4 shows a schematic diagram of the structure of a computer device provided by Embodiment 3 of the present invention, wherein: Figure 5 As shown, a computer device 500 provided in Embodiment 3 of the present application includes: A processor 501, a memory 502 and a bus 503, wherein the memory 502 stores machine-readable instructions executable by the processor 501. When the computer device 500 is running, the processor 501 communicates with the memory 502 through the bus 503. When the processor 501 is running, the machine-readable instructions execute the steps of the model training method based on space-based distributed AI computing power shown in the above-mentioned embodiment 2.

[0075] Embodiment 4 Based on the same application concept, an embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the model training method based on space-based distributed AI computing power described in any of the above embodiments are executed.

[0076] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0077] The computer program product for model training based on space-based distributed AI computing power provided in an embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0078] The model training system based on space-based distributed AI computing power provided in the embodiment of the present invention can be specific hardware on the device or software or firmware installed on the device. The system provided in the embodiment of the present invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brief description, the parts not mentioned in the system embodiment can refer to the corresponding contents in the aforementioned method embodiment. Technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0079] In the embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some communication interface, device or unit, which can be electrical, mechanical or other forms.

[0080] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0081] In addition, each functional unit in the embodiment provided by the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0082] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0083] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0084] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A model training system based on space-based distributed AI computing power, characterized in that: The system includes several satellites, each of which is deployed with several computing units and AI algorithm deployment units, and each computing unit includes a distributed computing allocation module and several high-speed interconnected AI computing modules; A first AI algorithm deployment unit, configured to respond to an AI algorithm publishing task sent by a ground user equipment and send a target AI algorithm included in the AI ​​algorithm publishing task to a computing unit of each satellite; A first computing power unit is used to respond to the algorithm model training task sent by the ground user equipment, and allocate a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module; Each target computing power unit is used to execute the algorithm model training task based on the target AI algorithm through its high-speed interconnected AI computing power module, and send the execution result of the algorithm model training task to the ground user equipment.

2. The model training system based on space-based distributed AI computing power according to claim 1, characterized in that: Each satellite also includes a multi-directional inter-satellite data transmission unit and a satellite-to-ground communication unit; The satellite-to-ground communication unit is used to realize communication and data transmission between the computing unit and the ground user equipment; The multi-directional inter-satellite data transmission unit is used to realize communication and data transmission between satellites.

3. The model training system based on space-based distributed AI computing power according to claim 1 is characterized in that: Each computing power unit also includes a cascade optical communication module; The cascade optical communication module is used to realize communication and data transmission between various computing power units.

4. The model training system based on space-based distributed AI computing power according to claim 1, characterized in that: The allocating a target computing power unit to the algorithm model training task according to the algorithm model included in the algorithm model training task through its distributed computing power allocation module includes: Based on the ground cloud computing platform and container technology, the computing resources of each computing unit are virtualized into multiple virtual resource pools; The virtual resources are dynamically allocated according to the resource conditions of each virtual resource pool and the requirements of the algorithm model training task, and the computing environment is quickly deployed through container technology to achieve target computing power unit allocation.

5. The model training system based on space-based distributed AI computing power according to claim 1, characterized in that: The high-speed interconnected AI computing power module includes: High-performance SOC processor with ARM+GPU+NPU as the core; The interface uses 15Gbps×8 high-speed differential transmission channels; Equipped with 32GB memory system, 1TB data storage, 64GB computing program storage and 256MB boot configuration storage.

6. The model training system based on space-based distributed AI computing power according to claim 2, characterized in that: The multi-directional inter-satellite data transmission unit comprises: The interface uses 15Gbps×4 high-speed differential transmission channels; Realize collaborative task allocation and scheduling among satellites based on advanced communication protocols.

7. The model training system based on space-based distributed AI computing power according to claim 3 is characterized in that: The cascade optical communication module comprises: Dual-port wired fiber channels, each channel contains 4-way duplex 10Gbps×4 fiber channels; The interface uses 15Gbps×8 high-speed differential transmission channels for high-speed interconnection and data distribution between modules.

8. A model training method based on space-based distributed AI computing power, characterized in that: Applied to a model training system based on space-based distributed AI computing power, the system includes several satellites, each satellite is deployed with several computing power units and AI algorithm deployment units, each computing power unit includes a distributed computing power allocation module and several high-speed interconnected AI computing power modules, the method includes: The first AI algorithm deployment unit responds to the AI ​​algorithm publishing task sent by the ground user equipment and sends the target AI algorithm included in the AI ​​algorithm publishing task to the computing power unit of each satellite; The first computing power unit responds to the algorithm model training task sent by the ground user equipment, and allocates a target computing power unit to the algorithm model training task through its distributed computing power allocation module according to the algorithm model included in the algorithm model training task; Each target computing unit executes the algorithm model training task based on the target AI algorithm through its high-speed interconnected AI computing module, and sends the execution result of the algorithm model training task to the ground user equipment.

9. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate through the bus. When the machine-readable instructions are executed by the processor, the steps of the model training method based on space-based distributed AI computing power are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the model training method based on space-based distributed AI computing power as described in claim 8.

Images