Hazardous flight region prediction for a small UAV operated in an urban area using a deep neural network

With an increase of UAVs in logistics and transportation, the safety of UAVs operated in the urban wind environment becomes an important issue. Small UAVs are more sensitive to the wind environment because of their small size, slow cruising speed, and limited endurance. In the unmanned aircraft system traffic management (UTM), a safety risk assessment under bad weather conditions is an important component. In this study, a hazardous flight region prediction system for small UAVs operated in urban areas is developed using a deep neural network (DNN) to support a risk assessment and safe trajectory planning. A large eddy simulation (LES) is applied to reflect the terrain-driven wind environment in the urban area. The result of a weather research and forecasting (WRF) model is used as an initial and boundary condition of the LES to generate a realistic complicated wind environment in an urban area. Furthermore, an iterative nesting algorithm is applied to the LES to obtain a sufficient resolution of the wind environment, which is suitable for the small UAV scale. The deviation distance from the original flight path due to the wind environment is considered as a flight hazard criterion in this study. The proposed system is able to predict deviation distance due to the wind environment over the entire flight space over time by using the DNN model. The training data for the DNN is obtained using the multicopter flight dynamics simulator, which can take into account the influence of a specific wind environment. With the indexes considering this deviation distance and the local topography (distribution of buildings) in the urban area, the hazardous flight region is predicted. The information supplied by the proposed hazardous flight region prediction model can be used for the flight risk assessment and safe flight trajectory planning to increase the flight safety of small UAVs.