基于小波分析与神经网络的组合导航的研究

Study of navigation system based on wavelet analysis and neural network

针对GPS/DR组合导航存在GPS信号被遮挡时无法对DR零点更新以及运动的高动态性造成卡尔曼滤波难以完全适应数据融合的问题,提出采用联邦卡尔曼滤波器数据融合与小波变换和正则化神经网络的DR位置预测模型相结合的方法。该方法由联邦卡尔曼滤波器得到较为精确的导航信息,与利用小波变换在不同尺度上融合所得到的误差信号输入神经网络,经过训练获得预测误差,在GPS信号失效时与导航信息相加实现精确实时定位。仿真计算结果表明,该方法可以提高导航系统的精度和速度,该模式有较好的鲁棒性,具有实用价值。

GPS/DR navigation system can＇t update the initial position for DR when the GPS signal is obstructed, and Kalman filter can＇ t be completely adapted to data fusion because of high dynamic. So the combination of the federated Kalman filter and the prediction model used neural network and wavelet transformation is proposed. It obtains more accurate navigation data by the federated Kalman filter, input to the neural network with the error signals obtained by wavelet transformation at different scales; the prediction error that is output by network after training added to Kalman filter＇ s data provides accurate real-time location. The simulation results show that the method is effective to improve the precision and calculation speed of autonomous navigation system and to provide with better robustness and practicality.