一类基于噪声协方差自适应的强跟踪滤波方法  

An Adaptive Strong Tracking Filtering Method Based on Noise Covariance

作　　者：王远亮 李云育 王亚飞 葛泉波 李宏[5] WANG Yuanliang;LI Yunyu;WANG Yafei;GE Quanbo;LI Hong(School of Logistics Engineering,Shanghai Maritime University,Shanghai 200135,China;School of Automation,Hangzhou Dianzi University,Hangzhou 310018,China;The Economic Commission of Songjiang,Shanghai 201600,China;School of Automation,Nanjing University of Information Science&Technology,Nanjing 210044,China;Chinese Flight Test Establishment,Xi’an 710089,China)

机构地区：[1]上海海事大学物流工程学院,上海200135 [2]杭州电子科技大学自动化学院,杭州310018 [3]上海松江经济委员会,上海201600 [4]南京信息工程大学自动化学院,南京210044 [5]中国飞行试验研究院,西安710089

出　　处：《指挥与控制学报》2025年第1期49-60,共12页Journal of Command and Control

基　　金：国家自然科学基金(61773147,62033010);中国航空科学基金(2019460T5001)资助。

摘　　要：强跟踪滤波是一种能有效应对状态突变和模型不精准系统的自适应估计方法。传统强跟踪滤波是利用渐消因子来调整预测估计误差协方差公式的状态转移矩阵相关部分,导致对滤波模型修正和模型参数估计的可解释性较弱,估计性能也有待于进一步改进。针对上述问题,在现有强跟踪渐消因子调整效果等价于调节过程噪声协方差的深度分析基础上,提出直接将渐消因子用来自适应动态调整预测估计误差协方差计算公式中过程噪声方差部分的强跟踪滤波思想,并给出了两种能有效保证协方差矩阵对称性的多重次优渐消因子计算方法。与传统的强跟踪滤波方法相比,新方法在现有滤波框架下直接实现了对系统模型参数的实时动态反馈修正,不仅避免因非对称而导致的滤波发散现象,同时具有非常清晰的原理和效果可解释性。实验结果表明,新方法具有跟踪估计性能上的明显改进。Strong tracking filtering is an adaptive estimation method that can effectively deal with state mutation and model inaccuracy in dynamic systems.The traditional strong tracking filtering uses the fading factors to adjust the relevant parts of the state transition matrix of the prediction and estimation error covariance formula,resulting in the weak interpretability of the filter model correction and model parameter estimation,and the estimation performance needs to be further improved.To solve the above problems,based on the existing deep analysis that the adjustment effect of strong tracking fading factors is equivalent to the adjustment of process noise covariance,a strong tracking filtering idea that directly uses fading factors to adapt to the process noise covariance part of the dynamic adjustment prediction estimation error covariance calculation formula is proposed,and two kinds of multiple suboptimal fading factor calculation methods that can effectively guarantee the symmetry of covariance matrix are given,direct matrix solution method and matrix trace solution.Compared with the traditional strong tracking filtering method,the new method directly realizes the real-time dynamic feedback correction of the system model parameters under the existing framework,which can not only avoid the filtering divergence caused by asymmetry but also has a very clear principle and effect interpretability.The experimental results show that the new method has the obvious improvement in tracking estimation performance.

关 键 词：强跟踪滤波 预测误差协方差 系统模型修正 渐消因子 矩阵迹 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]