To address the real-world challenges in emergency rescue within dense urban core areas—exemplified by Gusu District, Suzhou—characterized by highly concentrated spatiotemporal distribution of incidents, ambiguous service boundaries between rescue centers, and real-time demand bursts, this paper proposes a periodic route optimisation method based on Two-Layer Adaptive Large Neighborhood Search (TL-ALNS). The method employs a periodic update strategy to batch-process accident locations at fixed intervals, thereby meeting real-time response requirements. The TL-ALNS utilizes a two-layer structure to learn the synergy between operator pairs and incorporates a designed data-driven predictive re-assignment destroy operator. This operator guides the re-allocation of accident demands in areas with fuzzy boundaries, facilitating the rapid generation of high-quality rescue routes. Application based on emergency incident data from Gusu District demonstrates that this method effectively adapts to the specific spatial layout and dynamic outbreak characteristics of the old city. It significantly optimizes rescue efficiency and system robustness while ensuring rapid response, providing feasible decision support for actual emergency command scheduling.