广义系统降阶分布式信息融合稳态Kalman滤波器  

Reduced-order distributed information fusion steady-state Kalman filter for generalized systems

作　　者：李怀敏[1] 邓自立[2] 孙刚[1] 李恒[1] 赵正平[1] 

机构地区：[1]阜阳师范学院计算机与信息工程学院,安徽阜阳236037 [2]黑龙江大学电子工程学院,黑龙江哈尔滨150080

出　　处：《阜阳师范学院学报（自然科学版）》2015年第3期70-75,共6页Journal of Fuyang Normal University(Natural Science)

基　　金：安徽省教育厅自然科学一般项目(KJ2013B192;2015KJ012);阜阳师范学院校级一般项目(2013FSKJ08;2015FSKJ08;2013FSKJ14);阜阳师范学院科技成果孵化基金(2013KJFH05);阜阳师范学院质量工程项目(2013ZYSD05)资助

摘　　要：针对带相关噪声的多传感器广义系统,提出一种分布式分量标量加权融合稳态降阶Kalman滤波器。应用奇异值分解将原广义系统转化两个等价的降阶子系统,将广义系统状态估计问题转为正常系统的状态估计问题,并求得任两个传感器子系统之间的稳态降阶滤波误差互协方差阵。兼顾融合精度和计算负担,以线性最小方差为融合准则,得到按分量标量加权的稳态Kalman滤波器。该滤波器避免了时刻计算协方差阵和融合权重明显减小了在线计算负担,便于实时应用。Monte Carlo仿真验证方法的有效性。For generalized systems with multisensory and correlated noise, a distributed fusion steady-state reduced-order Kalman filter is presented. Applying the singular value decomposition, it is transformed into two reduced order coupled subsystems. Then the state estimation problems of generalized systems become the normal systems＇ state estimation problem. The cross-covari- anee matrix of steady-state reduced-order filtering errors between any two sensor subsystems is derived. At the same time to consider the fusion accuracy and the computational burden, it proposes the Kalman filter weighted by diagonal matrices in the linear minimum variance sense. The proposed steady-state fusion filter method avoids computing covariance matrices and fusion weights at each time step, so the computational burden can be reduced, and convenient to apply in real time. Simulation example shows the effectiveness of the proposed method.

关 键 词：广义系统 矩阵奇异值分解 KALMAN滤波器 ARMA新息模型 

分 类 号：O-211.64[理学]