An UKF and PSO-based neural network hybrid algorithm for attitude determination

Abstract

In order to restrain the influence of random disturbance on the attitude determination of in-motion rolling projectile, a new hybrid filtering algorithm, which combines unscented Kalman filter (UKF) with improved adaptive BP neural network based on particle swarm optimization (PSO), is proposed. When the attitude determination of rolling projectile is influenced by random disturbance, the output of neural network will replace that of UKF. The validity of hybrid algorithm is verified through the experiment, in which three low-cost micro electro-mechanical system (MEMS) accelerometers are used as strapdown inertial measurement units (IMUs) to determine rolling projectile attitude. Experiment results show that the proposed hybrid filtering algorithm is effective and robust, and it can effectively enhance the precision of state estimation and restrain the influence of dynamic random disturbance.