A hybrid approach for UAV flight data estimation and prediction based on flight mode recognition

Unmanned aerial vehicle (UAV) flight data estimation and prediction are critical for flight attitude tracking, safe operation and system health management (SHM). However, it is usually difficult to establish exact physical models for complex systems in many cases, which brings more challenges for model based filtering approaches. Moreover, due to non-linearity, uncertainty, and noise involved in flight data, using a single data-driven method is always incapable for fitting a complete flight track. To address the issues above, this paper proposes a hybrid approach of Gaussian Process-Unscented Kalman Filter (GP-UKF) based on Flight Mode Recognition (FMR) for UAV. The proposed method combines two ideas: (1) The UAV flight data are divided into different segments by FMR mechanism for improving the adaptability of estimation and prediction model. (2) According to different flight modes, GP recursive models are learned from flight data and utilized as the state transition equation in each Unscented Kalman Filter (UKF), and then GP-UKFs are constructed to realize higher estimation accuracy and uncertainty presentation. Experiments based on real UAV flight data verify the effectiveness of the proposed framework and indicate the potential for real applications.