Real-time Wind Estimation with a Quadrotor using BP Neural Network

This paper presents an approach based on BP neural network for quadrotors that estimates the wind velocity in real-time based on measurement data of its on-board inertial measurement unit (IMU) and GPS only. The proposed method is a gray box modelling method for the real-time wind estimation, avoids oversimplifications and determination of many parameters in the existing dynamic models or aerodynamic models of quadrotors. The nonlinear functional relationship between the wind velocity and the flight parameters provided by the on-board IMU and GPS is established after the training of the BP network, using the data collected from the quadrotor and an anemometer not far away from the quadrotor, and then applied to estimate the wind velocity in real time only with the outputs of the on-board IMU and GPS when the quadrotor is flying. The simulation results show that the proposed method can achieve wind estimation with a root mean square error (RMSE) less than 0.02 m/s.