We introduce a momentum‑guided federated split distillation framework targeting personalized, efficient, and autonomous temporal edge intelligence. The framework consists of two key components:\ \ TeRR‑SAtt (Temporal Reservoir Student Attention) is a novel design that merges fixed reservoir representations, a lightweight temporal student network, and client‑specific output modules. This combination preserves expressive power while fitting the limited resources of edge devices.\ \ AMGF (Anticipatory Momentum‑Guided Fusion) is an anticipatory fusion mechanism that clusters clients based on learned momentum and derives specialized teacher updates for each cluster, enhancing the adaptability of global knowledge.\ \ Evaluated on real‑world smart‑building datasets, TeRR‑SAtt reduces edge training latency by 65.50%, inference latency by 44.70%, training memory usage by 18.40%, and inference CPU usage by 33.10% compared with all baselines. At the same time, AMGF improves local learning by up to 35.31% in RMSE over purely global updates.\ \ Review: By coupling momentum‑driven client clustering with a lightweight temporal student, this work delivers an efficient and scalable solution for personalized temporal prediction on edge devices. The experimental results demonstrate comprehensive gains in latency, resource consumption, and prediction accuracy.