神经网络辅助的GPS/INS组合导航自适应滤波算法  

Neural Network Aided Adaptive Filtering for GPS/INS Integrated Navigation

作　　者：高为广[1] 杨元喜[2] 张婷 

机构地区：[1]信息工程大学测绘学院 [2]西安测绘研究所 [3]61081部队

出　　处：《测绘学报》2007年第1期26-30,共5页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金项目(40274002;40474001)

摘　　要：首先利用预报残差构造的最优自适应因子设计GPS/INS组合导航自适应滤波器。并针对BP神经网络存在的训练速度慢、容易陷入局部极小等问题,给出网络的改进算法。利用神经网络对自适应滤波器状态方程的预报值进行在线修正,给出神经网络辅助的GPS/INS组合导航自适应滤波算法。最后,利用实测数据进行验证。结果表明,改进的神经网络算法明显提高网络收敛速度;两种自适应滤波算法相对标准组合导航算法都能够可靠地反映载体运动轨迹;神经网络辅助的GPS/INS组合导航自适应滤波算法相对GPS/INS组合导航自适应滤波算法在精度和可靠性方面又有明显提高。Firstly, an integrated GPS/INS adaptive filter based on predicted residuals is set up. Then, in order to overcome the shortcomings of neural network, such as slow learning speed, easily arriving at local minimum, an improved network algorithm is established. Further more, an integrated GPS/INS algorithm, by using the adaptive filter and improved neural network to modify the predicted states information from kinematic model on line, is put forward. It is shown, by derivations and calculations, that the improved BP algorithm can obviously reduce the learning time. This combined algorithm can not only improve the filter estimation accuracy, but also control the influences of measurement outliers and the disturbances of the dynamical model. The new algorithm gives more actual and reliable navigation results of the maneuvering vehicles.

关 键 词：神经网络 自适应估计 自适应滤波 GPS/INS组合导航 

分 类 号：P228.41[天文地球—大地测量学与测量工程]