野值存在下的BP网络自适应卡尔曼滤波  

BP Neural Network Adaptive Kalman Filtering When Existing Outliers

作　　者：李俊[1] 周良荣[1] 赵涛[1] 王小峰[1] 

机构地区：[1]西北工业大学航海学院,陕西西安710072

出　　处：《火力与指挥控制》2009年第11期61-64,67,共5页Fire Control & Command Control

基　　金：教育部新世纪人才支持基金资助项目(NECT)

摘　　要：提出了一种抗野值的BP网自适应卡尔曼滤波算法。通过BP网络对新息序列的方差和均值的速率进行在线实时监测计算,输出一组加权系数对新息做修正处理,从而有效地抑制了野值对滤波器的影响。仿真证明该算法保持了新息的统计特性,提高了滤波器的精度和稳定性。An arithmetic of BP Neural Network adaptive kalman filtering is proposed for restraining the effect of outliers. According to supervise and compute the velocity of means and covariance of innovation sequence by BP Neural Network online, a group of weighting coefficient is exported to modify the innovations. Accordingly, the innovations＇ effect is restrained effectively. The simulation results prove that the arithmetic can keep the statistic feature of innovation and increase the precision and stability of the filter.

关 键 词：BP神经网络 卡尔曼滤波 新息 残差 野值 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]