Intelligent routing methods for low-Earth orbit satellite networks based on machine learning: A comprehensive survey

With the continuous progress of modern communication technology and the emergence of the 6G concept, people’s demand for high-quality and widely accessible data transmission is becoming increasingly intense. Low Earth Orbit (LEO) satellite networks show great attraction due to their characteristics of global coverage and low latency. Traditional terrestrial routing methods face significant challenges in adapting to LEO satellite networks due to challenges such as highly dynamic topologies, resource constraints, and insufficient multi-objective optimization capabilities. Therefore, developing routing methods suitable for LEO satellite application scenarios is crucial for further improving network transmission performance and is also one of the key technologies of future 6G. Compared with traditional algorithms, routing algorithms based on machine learning (ML) are more intelligent and begin to show obvious performance advantages, and are more suitable for 6G networks. However, in existing research work, there is a lack of comprehensive analysis content on integrating ML into LEO satellite network routing tasks. We comprehensively summarize the latest progress of intelligent routing algorithms based on ML in LEO satellite networks from four aspects: routing models, design challenges, training and deployment, and future research directions. The aim is to provide theoretical support for the design of artificial intelligence satellite communication systems and further promote the innovative development of satellite network optimization technologies.