基于联邦学习的路由策略网络的训练方法和装置
By employing a federated learning-based routing policy network training method in large-scale constellation networks, the satellite network is divided into regions and autonomously managed. The routing policy network is trained using graph neural networks and multi-agent reinforcement learning algorithms, which solves the problems of high computational and storage pressure and limited resources in constellation networks, thereby improving network performance and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2025-10-15
- Publication Date
- 2026-07-17
AI Technical Summary
In large-scale constellation networks, dynamic routing optimization faces challenges such as excessive computational and storage pressure, limited resources, limited link bandwidth, and high dynamism, leading to low network performance and efficiency.
A federated learning-based routing policy network training method is adopted, which divides the satellite network into regions, manages the autonomous management through satellite management, trains the routing policy network using graph neural networks and multi-agent reinforcement learning algorithms, and fuses and updates network parameters by aggregating satellites to realize the generation and deployment of a global routing policy model.
It reduces the computational and storage pressure on individual nodes, adapts to the conditions of limited onboard resources, improves the routing performance and network efficiency of the entire constellation network, avoids local optima and policy conflicts, and achieves global consistency and collaborative optimization of dynamic routing decisions.
Smart Images
Figure CN121509301B_ABST