" Cross-Regional Co-Trench Localization"

"Co-trenching primary and backup fibers within the co-trench poses a severe threat to the IoT infrastructure's physical layer reliability. Existing methods exhibit insufficient generalization under cross-regional scenarios caused by varying geological conditions and environmental noise, while requiring extensive labeled data for training. We propose a zero-shot cross-regional co-trench fiber localization method for IoT networks based on LLM expert knowledge distillation and CoT prompting. Specifically, we construct 27-dimensional structured time\u0002frequency feature matrices to compel the LLM to learn invariant logical reasoning rules via knowledge distillation, enabling robust zero-shot  inference without retraining. Experiments on Baoshan, Xinxigang, and Qujing datasets show that the fine-tuned Qwen3- 8B model (trained only on 167 MB Baoshan data) achieves comprehensive index scores of 97.37%, 93.58%, and 91.21%, respectively, outperforming baseline models in cross-regional tests. Results confirm superior zero-shot cross-regional generalization capability, offering a robust and low-latency novel paradigm for smart fiber optic sensing."