A federated user prediction method for 6G edge network
By employing a federated learning-based user prediction method in 6G edge networks, utilizing patch feature extraction and Transformer Encoder combined with KD-trees, the accuracy and privacy issues of traditional prediction mechanisms are resolved, achieving efficient and secure user prediction and resource optimization, suitable for diverse 6G application scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2026-05-26
- Publication Date
- 2026-07-10
AI Technical Summary
Traditional user prediction mechanisms in 6G edge networks suffer from limited prediction accuracy and insufficient generalization performance, making them unsuitable for diverse mobile scenarios. Furthermore, centralized training leads to risks of data privacy leaks and network congestion, failing to meet the needs of real-time connection and resource scheduling for massive numbers of devices.
A user prediction method based on federated learning is adopted. The user prediction model is constructed by dividing the model into a patch feature extraction module and a prediction result alignment module. Combined with Transformer Encoder and KD tree, the model can efficiently capture user behavior features. The model is also trained collaboratively between the edge core network and the central core network to avoid privacy leakage during data transmission.
It achieves efficient and accurate user prediction, reduces computational complexity and latency, optimizes resource allocation, ensures data privacy and security, adapts to multiple business scenarios, and supports 6G applications such as intelligent transportation and industrial IoT.
Smart Images

Figure CN122372445A_ABST