Flight Control of Tiltrotor Uavs: Pso-Optimized Neural Network Aerodynamic Modeling With Esobased Disturbance Compensation

Tiltrotor UAVs, with their capability to operate in both fixed-wing and rotorcraft flight modes, hold significant potential for applications in complex environments. However, their nonlinear aerodynamic characteristics at low speeds are difficult to identify using traditional modeling methods, which limits control accuracy. This paper proposes an aerodynamic identification method based on neural networks and Particle Swarm Optimization (PSO), leveraging offline historical flight data processing to construct a high-precision aerodynamic model. This model is then integrated as a compensation term into the realtime control system to enhance model accuracy. To mitigate the impact of unknown disturbances during flight, extended state observers (ESOs) are further employed for real-time compensation, improving disturbance rejection and trajectory tracking accuracy. The simulation is conducted on a tiltrotor UAV using an eight-shaped flight trajectory experiment. The results indicate that this method significantly improves control performance under low-speed flight conditions, while also enhancing the system's robustness.