神经网络辅助的组合导航系统信息融合方案

传统的Kalman滤波器自适应能力弱，而单纯的神经网络滤波器估计精度较差，且网络训练经验性太强。面向组合导航领域，提出BP神经网络辅助自适应联邦Kalman滤波器方案，设计并实现了SINS／GPS／TAN／SAR智能化容错组合导航系统。结合自适应滤波和神经网络两种方法共同提高系统的自适应能力，并提出新的神经网络输入量，改善了算法的实时性。系统的估计精度得到显著提高，仿真结果证明了该方案的可行性和有效性。

Neural-Network-Aided Information Fusion for Integrated Navigation Application

The classical Kalman filter has a poor adaptive capability, and the estimation accuracy of the neural-network filter is not satisfactory and its estimation effect depends on the experience for network training excessively. We propose incorporating a Back-Propagation(BP) neural-network into the adaptive federated Kalman filter configuration for the SINS/GPS/TAN/SAR(Strapdown Inertial Navigation System/Global Positioning System/Terrain Auxiliary Navigation/Synthetic Aperture Radar) integrated navigation system. The proposed algorithm combines the eatimation capability of the adaptive Kalman filter and the learning capability of the BP neural-network, thus resulting in improved adaptive capability and estimation accuracy. We put forward a new and simple input vector for the BP neural-network; it considerably improves the real-time performance of the algorithm. Compared with the estimation performance of the adaptive Kalman filter, the estimation performance of the BP-neural-network-aided adaptive Kalman filter shows considerable improvement: (1) The attitude estimating accuracy is improved 20~30 angle seconds; (2) The position estimating accuracy is improved 5~10 m; (3) The velocity estimating accuracy is improved 0.02~0.05 m/s.