Path planning for logistics UAVs considering energy consumption and safety under wind field uncertainty

Wind disturbances at low altitudes increase energy consumption and safety risks for multirotor unmanned aerial vehicles (UAVs) in densely built urban environments. High-fidelity computational fluid dynamics (CFD) is too expensive for online path planning, while threshold methods represent hazardous regions as binary no-fly zones. This study proposes a path planning framework that combines a Fourier neural operator (FNO) surrogate with a multiplicative cost model incorporating predictive uncertainty. The FNO predicts the urban wind field and its spatial uncertainty. The predicted wind field determines physical flight energy, while uncertainty modulates this term to form an equivalent energy cost. Costs for the 26-neighborhood are precomputed, allowing 3D A ∗ search through tensor lookups. Experiments use CFD wind fields for a 3D area modeled on the Macao Peninsula, five meteorological scenarios, and 13 valid delivery routes. Mean FNO inference time is 102.013 ms. After preprocessing and cost tensor construction, mean search time per route is 5.675 s. Mean risk exposure, defined as the percentage of path nodes above a predictive uncertainty threshold, decreases from 25.76% for Optimal Energy to 7.12% for the proposed method, with scenario values from 5.68% to 11.47%. Mean physical energy increases by 2.34% relative to Optimal Energy. Relative to geometric shortest path planning, energy decreases by 4.40% in the annual average scenario and by 5.25% in winter. The framework improves the balance between uncertainty exposure and modeled energy consumption, supporting potential applications in more sustainable urban logistics planning.