多源观测与检索增强下低空配送工程调度优化

Multi-Source Observation and RAG-Based Engineering Scheduling for Low-Altitude Delivery

摘要: 低空配送调度缺乏可核查的空域观测依据与合规约束来源。本文以OpenSky匿名ADS-B深圳终端区快照、美团无人机公开运营信息与《无人驾驶航空器飞行管理暂行条例》为依据，构建“观测–检索–调度”三层工程框架，并设置双向信息流：观测到的飞行热点转化为调度模型的规避成本，召回的法规条文参数化生成GIS禁飞多边形作为硬约束。观测录得71条状态向量，真高300 m以下无ADS-B目标，显示常规监视存在盲区。BM25检索稳定召回真高120 m管制门槛、轻型机限速与投放申请等约束。深圳5节点算例中，双机优化将完工时间由55.1 min降至43.4 min，降幅21.1%。

Abstract: Using OpenSky ADS-B snapshots over Shenzhen, Meituan’s disclosed operations and the Interim Regulation on Unmanned Aircraft Flight Management, this paper builds an observation-retrieval-scheduling engineering framework with a bidirectional flow: flight hotspots become avoidance costs, and retrieved clauses become GIS no-fly polygons as hard constraints. No recorded state vector lies below 300 m, evidencing a surveillance blind spot; a clause-level BM25 index retrieves binding provisions such as the 120 m threshold to constrain a multi-drone routing model. On five Shenzhen nodes, two-drone optimization cuts the makespan from 55.1 to 43.4 min (−21.1%).