A framework of data-driven real-time risks prediction method on UAV flight safety

Unmanned Aerial Vehicles (UAVs) are complex systems with a high accident rate during flight operations. This is mainly due to the lack of real-time human observation and affordable redundancy design. As UAV mission capabilities expand, the associated flight safety risks from various failures become more significant, leading to rising costs and damages. Although there have been discussions on using data-driven methods to predict flight safety risks using accumulated UAV operation data, there is a lack of mature approaches for real-time prediction analysis and application. With the rapid development of data transmission technology, UAVs now have high-bandwidth stable data links, allowing for real-time telemetry and command control. This advancement opens opportunities for real-time online risk prediction. In this paper, we propose a data-driven framework for realtime prediction of UAV flight safety risks. The framework integrates wavelet packet decomposition, BP neural network, and utilizes actual UAV data for model verification. Using this framework, we construct a UAV real-time risk prediction model by incorporating real-time data. The model achieves an 89% recall rate and demonstrates the ability to accurately predict UAV operational risks in a timely manner to a certain extent.