Abstract: In the MEMS/GPS integrated navigation system, the positioning error of pure inertial navigation will quickly diverge during GPS outages. In order to suppress the error divergence, an improved radial basis neural network and an adaptive Kalman filter algorithm are proposed, and a new network training model is introduced. The adaptive quantum particle swarm optimization algorithm is used to improve the structural design and the parameters of radial basis neural network. The neural network is trained with the combined navigation data when the GPS signal is available. When the GPS outage occurs, the measurement of AKF is predicted by the improved radial basis neural network, so that the error of speed and position are provided continuously for the system. Experimental results show that when the GPS outage is up to 15 s, the horizontal positioning accuracy is improved by 62% compared with that of the original algorithm under the state of turning, which effectively suppresses the inertial navigation error.

Key words: inertial navigation system, error restraining, adaptive quantum particle swan optimization, radial basis function neural network, adaptive Kalman filtering