神经网络修正动态GPS卡尔曼滤波算法研究

摘要

GPS导航定位系统噪声具有非先验性,而卡尔曼滤波进行最优估计需建立准确的系统模型和观测模型,这导致标准卡尔曼滤波的精度不高.为提高滤波精度,提出了神经网络修正动态GPS卡尔曼滤波算法,采用两个BP神经网络分别在时间更新预测部分及测量更新部分对标准卡尔曼滤波器进行修正,这样既考虑了现实环境的动态变化对系统模型造成的随机干扰影响,又融合了神经网络的自学习性和自适应性,使其对动态环境的扰动具有了自适应能力.仿真研究表明:该算法优于标准卡尔曼滤波器.%The noise of GPS navigation position system has the character of apriority,and the optimal estimation carried out by Kalman filter need to establish accurate system model and observation model.These lead to the low accuracy of the standard Kalman filter. In order to improve filtering accuracy,arithmetic for dynamic GPS Kalman filtering modified by neural network is proposed. Two BP neural network is used to correct the standard Kalman filter in the time update predict part and measure update part. The method not only considers the random interference with the dynamic change,but also utilizes neural network's capability of self-taught and self-adapted. The simulation results show that the algorithm is better than that of standard Kalrnan filtering.