An approach of uav flight state estimation and prediction based on telemetry data

Flight state estimation and prediction of unmanned aerial vehicles (UAVs) are essential for safe operation, and they are primary bases of prognostics and health management (PHM). Telemetry data of UAV are the most significant resource for flight state tracking. However, telemetry data has the characters of high-dimension, non-linearity, uncertainty, and associated with noise, and it's hard to get accurate complex system model needed by classical filtering algorithms in many cases. Gaussian Process Regression (GPR) has the feature of adaptive parameter estimation and nonlinear regression, and Unscented Kalman Filter (UKF) relies on unscented transform for high tracking accuracy. In this article, a hybrid method based on Gaussian Process-Unscented Kalman Filter (GP-UKF) is proposed. The GP recursive model is constructed based on real-time telemetry data, which can be used as the state transition equation in UKF. The proposed method which combines the advantages of these two algorithms can achieve effective estimation and prediction of UAV flight state. Experiments based on real telemetry data of UAV verified the effectiveness of the method, and fast accurate UAV flight state tracking is achieved.