Urban air mobility flight hazard index prediction using WRF-LES simulations and LSTM networks

Abstract

Urban air mobility (UAM) has emerged as a potential solution to mitigate urban traffic congestion. However, severe turbulence in urban wind environments poses a significant safety issue for UAM operations. To ensure safe and reliable UAM operations, a UAM hazard prediction system is essential. This study proposes a UAM flight hazard index prediction system. To achieve this, first, UAM flight data were generated through the coupling of a UAM dynamics simulator with an actual urban wind environment. The urban wind environment was produced using the Weather Research and Forecasting-Large Eddy Simulation coupled model. By applying wingless type and lift&cruise type UAM dynamics simulators to these urban wind environments, a flight simulation database was constructed. For the assessment of the UAM flight hazard, a new hazard index, ${\overrightarrow{v}}_{\mathrm{dev}}$, was derived from wind components that induce path deviations. Analyses confirmed that ${\overrightarrow{v}}_{\mathrm{dev}}$ indicates wind-induced hazards while accounting for both wind magnitude and direction. Long Short-Term Memory networks were then trained using the flight simulation database to predict the hazard index. In particular, an initializer neural network was incorporated to enable predictions from arbitrary initial states. The resulting models demonstrated high accuracy for both types of UAM. Using these models, the hazard index in UAM corridors was evaluated. The results exhibited different trends in the hazard index under varying wind conditions. Under the headwind and tailwind conditions, the hazard index values were low for both types. In contrast, under crosswind conditions, the hazard index was high. The wind speed increasing with altitude was another factor contributing to the hazard index. Additionally, different hazard index values were observed between the two UAM types under the same wind conditions due to the different flight characteristics.