Consideration for Space-Based Computing Infrastructure Network

3.1. Emergency Response and Disaster Monitoring

During natural disasters, such as earthquakes and floods, traditional communication and computing systems are at risk of damage, resulting in delays in the transmission of critical information. However, by utilizing satellite computing networks, emergency communication and computing nodes can be quickly deployed to process disaster imagery in real time. This allows for the creation of precise disaster maps and optimal rescue routes, providing decision support at a minute or even second level.¶

This greatly improves the efficiency of disaster warning, emergency response, and resource allocation. Additionally, in the event of terrestrial network failures, these satellite networks can seamlessly provide communication and edge computing capabilities to support emergency command, drone search-and-rescue operations, and post-disaster reconstruction data processing.¶

3.2. Environmental Monitoring and Ecological Management

nder traditional models, large amounts of raw satellite data, such as 0.3-meter high-resolution imagery, must be transmitted back to Earth for processing. However, due to limited satellite-to-ground communication bandwidth, less than one-tenth of the data can be transmitted, resulting in low efficiency.¶

To address this issue, AI models can be deployed in orbit to perform real-time target detection, classification, change monitoring, and feature extraction on remote sensing imagery. This allows only critical analysis results to be transmitted to the ground, improving efficiency. This technology can accurately identify farmland, forests, water bodies, and glaciers, making it easier to track carbon sinks, monitor water environments, and track vegetation degradation.¶

As a result, data utilization rates have increased from 10% to nearly 100%, greatly enhancing the timeliness and autonomy of national land resource surveys, environmental monitoring, agricultural assessments, and related fields.¶

3.3. Deep Space Exploration Mission Support

Deep-space probes experience significant communication delays with Earth, with delays of several minutes being common for missions to Mars. This reliance on ground control can be inefficient.However, by deploying computational nodes in deep-space orbits, these probes can perform in-orbit preprocessing, compression, and intelligent filtering of data.¶

This allows for coordination through inter-satellite communication networks, resulting in a significant reduction in the volume of raw data that needs to be transmitted back to Earth. This approach not only enhances the autonomous operation capabilities of probes, but also improves their mission response speed. It serves as a critical foundation for future long-term exploration missions to destinations such as the Moon, Mars, and beyond.¶

3.4. In-orbit Training and Inference for Large AI Models

Training AI models with hundreds of billions of parameters requires immense computational power, which can pose energy and thermal bottlenecks for ground-based data centers. However, by leveraging the distributed computing capabilities and green energy advantages of space computing networks, it is possible to distribute model training and inference.¶

This approach provides a new "zero-carbon" computing pathway for AI development.¶

4.1. Space-Based Computing Resource Monitoring

Spaceborne equipment faces significant constraints in terms of computational resources, including CPU/GPU processing power, storage capacity, and energy consumption limits. These limitations are due to the size, power consumption, and payload capacity of the equipment. Additionally, the computational configurations of different satellites can vary greatly. Some prioritize edge computing, while others focus on data relay.¶

Furthermore, the computational load of satellites can fluctuate depending on mission requirements. For example, sudden spikes in remote sensing data processing or IoT terminal access within a specific region can overload local satellites, while satellites in other areas may remain idle.¶

This highlights the need for a technical solution that can monitor the computational load, available resources, and energy consumption status of each satellite in real-time. This data would then be used to support cross-satellite resource allocation.¶

4.2. On-demand Traffic Scheduling

Satellite networks support a wide range of service types, each with unique demands for network and computing power. For example, emergency communications require low latency and high reliability, while remote sensing data processing requires significant computing power but is less sensitive to latency. IoT data transmission prioritizes high bandwidth and low power consumption.¶

However, a unified scheduling strategy may lead to issues such as "computing power mismatch" (e.g. assigning high-latency services to long-range satellites) or "resource wastage" (e.g. using high-performance computing satellites for simple data relay tasks).¶

Therefore, it is crucial to establish a matching mechanism between service requirements and resource capabilities, including network resources such as link status, in order to enable efficient on-demand scheduling.¶