Wind Estimation for Fixed-Wing Uav Using Interacting Multiple Model Adaptive Unscented Kalman Filter

This paper proposes an Adaptive Interacting Multiple Model Unscented Kalman Filter (A-IMM-UKF) for real-time wind estimation on fixed-wing Unmanned Aerial Vehicles (UAVs). The proposed method integrates the Interacting Multiple Model (IMM) framework with ten parallel Unscented Kalman Filter (UKF) models representing diverse turbulence regimes, combined with online adaptive mechanisms for both process noise covariance (Q) and measurement noise covariance (R) estimation. The adaptive Q mechanism employs a sliding window maximum likelihood approach to track time-varying wind disturbance intensities, while the adaptive R mechanism dynamically adjusts measurement trust based on a criterion error derived from temporal and predictive residuals, improving robustness against sensor outliers. The wind measurement vector is derived from the kinematic relationship between Global Positioning System (GPS)-based ground speed and the projected airspeed vector using a six-degree-of-freedom (6 DOF) UAV model. Simulations conducted in MATLAB under a 300-second flight scenario with injected high-intensity Dryden turbulence (up to 11 m/s horizontal and 1.4 m/s vertical), resulting in total wind peaks of 11.16 m/s horizontal and 1.31 m/s vertical demonstrate that A-IMM-UKF achieves an overall Root Mean Square (RMS) error of 0.1760 m/s, representing improvements of 64.4% over IMM-UKF and 78.2% over standard A-UKF. These results confirm the complementary benefit of combining the multi-model structure with dual adaptive noise estimation for robust wind estimation. The proposed method is validated entirely through MATLAB simulation; hardware deployment is left for future work.