The proliferation of IoT devices in smart cities challenges 6G networks with conflicting energy-latency requirements across heterogeneous slices. Existing approaches struggle with the energy-latency trade-off, particularly for massive scale deployments exceeding 50,000 devices km. This paper proposes an edge-aware CyberTwin framework integrating hybrid federated learning for energy-latency co-optimization in 6G network slicing. Our approach combines centralized Artificial Intelligence scheduling for latency-sensitive slices with distributed federated learning for non-critical slices, enhanced by compressive sensing-based digital twins and renewable energy-aware resource allocation. The hybrid scheduler leverages a three-tier architecture with Physical Unclonable Function (PUF) based security attestation achieving 99.7% attack detection accuracy. Comprehensive simulations demonstrate 52% energy reduction for non-real-time slices compared to Diffusion-Reinforcement Learning baselines while maintaining 0.9ms latency for URLLC applications with 99.1% SLA compliance. The framework scales to 50,000 devices km with CPU overhead below 25%, validated through NS-3 hybrid simulations across realistic smart city scenarios.

翻译：智慧城市中物联网设备的激增对6G网络提出了挑战，异构网络切片中存在相互冲突的能量与延迟需求。现有方法难以平衡能量与延迟间的权衡关系，特别是在设备密度超过每平方公里5万台的大规模部署场景中。本文提出一种集成混合联邦学习的边缘感知数字孪生框架，用于6G网络切片中的能量-延迟协同优化。该方法通过集中式人工智能调度处理延迟敏感型切片，结合分布式联邦学习处理非关键切片，并辅以基于压缩感知的数字孪生技术和可再生能源感知资源分配机制。混合调度器采用基于物理不可克隆函数（PUF）安全认证的三层架构，攻击检测准确率达到99.7%。综合仿真表明：与扩散强化学习基线相比，非实时切片能耗降低52%；同时为URLLC应用保持0.9ms延迟，服务等级协议（SLA）合规率达99.1%。该框架可扩展至每平方公里5万台设备规模，CPU开销低于25%，已通过NS-3混合仿真在真实智慧城市场景中得到验证。