【摘要】 本文提出了采用变步长ＢＰ网络实时处理导航雷达目标跟踪滤波数据。其原理是在ＢＰ算法中加入变动量因子，用Ｎ层变步长寻优进行前馈神经网络的学习。该算法能够根据输入时间序列的特点自适应调整网络优化步长，对匀速和变加速运动目标的稳定跟踪精度分别在７．５ ｍ 和１１ ｍ 以内，而对急转弯的目标抖动２～４ 次就能及时跟踪上更多还原

【Abstract】 This paper presents a novel adaptive method to solve the problem of identifying the parameters of filter real timely for the navigation radar tracking by means of a BP network.The varying steps are employed in the BP network training in addition to the momentum factor.The steps are varied adaptively according to the characteristics of data in.The accuracy of stable tracking is below 7.5 m for a uniform motion and 11 m for a varied acceleration motion.This network can also track the target in time(passing by 2 to 4 steps)which takes a sudden turn.更多还原