一种基于在线学习前向神经网络的组合导航滤波器

Integrated Navigation Filter Based on an Online Learning Feed-Forward Neural Network

首先建立了带反馈校正的组合导航数学模型 ,在此基础上提出了一种在线学习的神经网络滤波算法。这种算法不需要噪声的先验知识 ,对系统模型的依赖也较弱。仿真表明 ,卡尔曼滤波器在理想情况下有较高的估计精度 ,而神经网络滤波器在非理想情况下有较高的精度 ,对模型误差和噪声特性的变化具有良好的鲁棒性。

A mathematical model of integrated navigation system with feedback control is given, and a filtering algorithm using online learning neural network is proposed. This algorithm has the capability of making estimation without knowledge of noise statistics and depends less on system model than Kalman filtering. Simulational results indicate that the Kalman filter has better accuracy in the ideal condition, while the neural network filtering algorithm has better robustness and accuracy under conditions of model uncertainties and noise characteristic variation.