Distributed Fusion Filtering for Multi-sensor Networked Uncertain Systems With Unknown Communication Disturbances and Compensations of Packet Dropouts
-
摘要: 研究了带有未知通信干扰、观测丢失和乘性噪声不确定性的多传感器网络化系统的状态估计问题.通过白色乘性噪声描述系统状态和观测中的随机不确定性，采用一组服从Bernoulli分布的随机变量描述网络传输过程中存在的观测丢失现象，且数据传输中存在未知的网络通信干扰.当发生丢包时，以当前丢失观测的预报值进行补偿.对每个单传感器子系统，应用线性无偏最小方差估计准则设计了不依赖于未知通信干扰的最优线性滤波器.推导了任两个局部滤波误差之间的互协方差阵.进而，应用矩阵加权融合估计算法给出了分布式融合状态滤波器.仿真例子验证了算法的有效性.Abstract: This paper is concerned with the state estimation problem for multi-sensor networked systems with unknown communication disturbances, measurement losses and multiplicative noise uncertainties. The random uncertainties of the state and measurements of systems are described by white multiplicative noises. The phenomena of measurement losses during data transmissions through networks are described by a group of Bernoulli distributed random variables. Unknown communication disturbances exist in data transmissions. The predictors of the lost measurements are used as the compensation in the presence of packet losses. By applying the linear unbiased minimum variance estimation criterion, an optimal linear filter independent of unknown communication disturbances is designed for every single sensor subsystem. Filtering error cross-covariance matrices between any two local filters are derived. Further, a distributed fusion state filter is presented by using the matrix-weighted fusion estimation algorithm. A simulation example is given to verify the effectiveness of the proposed algorithms.
-
近年来, 网络化控制系统的广泛应用使得有关网络化系统的控制与估计问题成为广大学者研究的热点[1-4].与传统的点对点控制系统相比较, 网络化控制系统具有信息交互速度快、控制范围广等优点.然而, 网络化系统也面临着在数据网络传输过程中的数据包丢失、随机滞后和未知干扰输入等问题.这些随机不确定性因素极大地影响了系统的性能, 甚至破坏系统稳定性.因此对带有未知干扰、丢失观测和乘性噪声不确定性的网络化系统进行滤波器的设计具有重要的实际意义.
目前, 针对网络化控制系统中涉及的丢包、滞后、未知输入、乘性噪声不确定性问题已有许多研究[5-15], 但综合考虑这些问题的研究文献还鲜见.文献[5-6]研究了带未知输入系统的观测器设计问题.文献[7]给出了线性离散随机系统未知输入和状态的统一形式滤波器.文献[8-10]研究了带有传输滞后、丢包或乘性噪声网络化系统的最优滤波问题.文献[11]对具有多数据包丢失线性离散随机系统设计了故障检测滤波器.然而, 文献[8-11]没有考虑多传感器融合估计问题.考虑到多传感器系统, 文献[12-15]对带有丢包和滞后的网络化多传感器系统研究了融合估计问题.然而, 文献[8-15]在数据包丢失时, 均采用前一时刻的观测近似代替丢失观测, 是一种简单的补偿.文献[16]对带有未知观测干扰和观测丢失的随机不确定多传感器系统给出了线性无偏最小方差最优融合预报器.然而, 对丢失观测没有补偿.文献[17]采用丢失观测的预报器作为补偿设计了稳态滤波器.采用相同的补偿方法, 文献[18]对带有未知通信干扰和丢包多传感器系统设计了融合预报器.由于使用了当前时刻之前的所有观测信息, 所以带预报补偿的估计比没有补偿和利用前一时刻观测补偿的估计具有更高精度.
由于模型误差、传感器老化、外部干扰和网络通信不完全可靠等问题, 网络控制系统中未知通信干扰、丢包和乘性噪声不确定性现象不可避免地存在.本文针对带未知通信干扰、观测丢失和状态与观测中均有乘性噪声不确定性的网络化多传感器系统, 采用文献[17]的方法以丢失观测的一步预报估值作为丢包补偿, 应用线性无偏最小方差估计准则[19], 设计了基于单传感器子系统的递推状态滤波器和基于多传感器系统的分布式融合滤波器.推导了任意两传感器子系统局部滤波器之间的滤波误差互协方差阵.最后, 应用矩阵加权融合估计算法给出了分布式融合滤波器.
1. 问题阐述
考虑带未知通信干扰、观测丢失和乘性噪声不确定的多传感器离散随机系统(图 1):
$ \begin{equation} {\pmb x}(t + 1) = [{\Phi_0}(t) + \xi (t){\Phi_1}(t)]{\pmb x}(t) + \Gamma (t){\pmb w}(t) \end{equation} $
(1) $ \begin{align} {\pmb y_i}(t) = &[{H_{0i}}(t) + {\lambda _i}(t){H_{1i}}(t)]{\pmb x}(t)+ {\pmb v_i}(t), \\ & i = 1, 2, \cdots , L \end{align} $
(2) $ \begin{equation} {\pmb z_i}(t) = {\gamma _i}(t){\pmb y_i}(t) + {D_i}(t){\pmb \theta _i}(t), i = 1, 2, \cdots , L \end{equation} $
(3) 其中, ${\pmb x}(t) \in {{\bf R}^n}$是系统的状态向量, 为传感器端观测输出, 它将经由网络传输给局部处理器(局部滤波器), ${\pmb z_i}(t) \in {{\bf R}^{{m_i}}}$是局部滤波器端收到的观测, 系统噪声和观测噪声是零均值、方差分别为${Q_{\pmb w}}(t)$和的不相关白噪声, 为未知的通信干扰. $\xi (t)$和是互不相关且均与其他变量不相关的零均值、方差分别为和${Q_{{\lambda _i}}}(t)$的标量白噪声. 是Bernoulli分布的随机变量序列, 其概率分布为, , , 且不相关于其他变量. ${\Phi _0}(t)$, ${\Phi _1}(t)$, , ${H_{0i}}(t)$, ${H_{1i}}(t)$, ${D_i}(t)$分别为适当维数的矩阵, 下标$i$表示第$i$个传感器, $L$表示传感器的个数.
模型(1) $\sim$ (3)描述了网络化系统中存在的未知通信干扰、乘性噪声不确定性和可能的观测丢失现象.当${\gamma _i}(t) =1$时, 观测数据没有丢失, 传感器观测经由网络按时到达局部滤波器端; 当${\gamma_i}(t) =0$时, 传感器观测数据丢失.为了改善局部滤波器估计精度, 我们采用丢失观测的预报值作为补偿, 此时, 用于局部滤波器设计的观测数据满足如下方程:
$ \begin{equation} {{\bar {\pmb z}}_i}(t) = {\gamma _i}(t){\pmb y_i}(t) + {D_i}(t){\pmb \theta _i}(t) + (1-{\gamma _i}(t)){{\hat {\pmb y}}_i}(t|t-1) \end{equation} $
(4) 其中, 丢失观测的预报器, 为状态预报值, 式中${{\hat {\pmb x}}_i}(t-1)$为$t-1$时刻状态的滤波估值.
假设1. 初始状态${\pmb x}(0)$与${\pmb w}(t)$, ${\pmb v_i}(t)$均不相关, 且满足:
$ \begin{equation} {\rm {E\{ }}{\pmb x}(0){\rm{\} }} = {\pmb \mu _0}, {\rm{E\{ [}}{\pmb x}(0) - {\pmb \mu _0}]{[{\pmb x}(0) - {\pmb \mu _0}]^{\rm{T}}}{\rm{\} }} = {P_0} \end{equation} $
(5) 其中, ${\rm E}$为期望, ${\rm T}$为转置号.
假设2 ${\rm{rank}}(D_i(t))=p_i$, $m_i>p_i$, . ${\rm rank}( * )$表示矩阵$*$的秩.
问题是基于补偿后的观测, 利用线性无偏最小方差估计准则[19]设计局部滤波器, 进而基于局部估计和按矩阵加权融合估计算法[20], 设计分布式融合递推状态滤波器.
2. 分布式融合滤波器
分布式融合滤波由于具有并行结构, 使其具有容错性好、可靠性高且易于故障诊断等优点.我们首先, 给出基于单传感器的线性无偏最小方差估计; 然后, 推导任两个局部估计误差间的互协方差阵; 最后, 应用按矩阵加权融合算法[20]给出分布式融合滤波器.
2.1 局部单传感器的滤波器
对系统(1) $\sim$ (4), 我们设计具有如下Kalman形式的局部递推状态滤波器
$ \begin{equation} {{\hat {\pmb x}}_i}(t + 1) = {F_i}(t){{\hat {\pmb x}}_i}(t) + {L_i}(t + 1){{\bar {\pmb z}}_i}(t + 1) \end{equation} $
(6) 其中增益矩阵${F_i}(t)$和${L_i}(t+1)$由如下定理1计算.
定理1. 在假设1和2下, 多传感器系统(1) $\sim$ (4)中局部单传感器子系统的递推状态滤波器(6)的增益阵${F_i}(t)$和${L_i}(t+1)$可计算如下:
$ \begin{equation} {F_i}(t) = {\Phi _0}(t) - {L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t) \end{equation} $
(7) $ \begin{equation} {L_i}(t + 1) = [G_i^{\rm{T}}(t + 1) - {\Lambda _i}(t + 1)D_i^{\rm{T}}(t + 1)]C_i^{ - 1}(t + 1) \end{equation} $
(8) 其中
$ \begin{align} {G_i}&(t + 1)= {\alpha _i}{H_{0i}}(t + 1)[{\Phi _0}(t){P_i}(t)\Phi _0^{\rm{T}}(t) + \\ & {Q_\xi }(t){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t) + \Gamma (t){Q _{\pmb w}}(t){\Gamma ^{\rm{T}}}(t)] \end{align} $
(9) $ \begin{align} {\Lambda _i}&(t + 1)= G_i^{\rm{T}}(t + 1)C_i^{ - 1}(t + 1){D_i}(t + 1)\times \\ &{[D_i^{\rm{T}}(t + 1)C_i^{ - 1}(t + 1){D_i}(t + 1)]^{ - 1}} \end{align} $
(10) $ \begin{align} {C_i}&(t + 1) = {\alpha _i}\{ {Q_{{\lambda _i}}}(t + 1){H_{1i}}(t + 1)X(t + 1)\times \\ & H_{1i}^{\rm{T}}(t + 1)+{H_{0i}}(t + 1)[{\Phi _0}(t){P_i}(t)\Phi _0^{\rm{T}}(t) + \\ & {Q_\xi }(t){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t)+\Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t)]\times \\ & H_{0i}^{\rm{T}}(t + 1) + {Q_{{\pmb v}_i}}(t + 1)\} \end{align} $
(11) 状态二阶矩计算如下:
$ \begin{align} X&(t + 1) = {\Phi _0}(t)X(t)\Phi _0^{\rm{T}}(t) + {Q_\xi }(t)\times \\ & {\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t) + \Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t) \end{align} $
(12) 初值$X(0) = {P_0} + {\pmb \mu _0} {{\pmb \mu}^{\rm{T}} _0}$.状态滤波误差方差计算为
$ \begin{align} {P_i}&(t + 1)= {\Phi _0}(t){P_i}(t)\Phi _0^{\rm{T}}(t) + {Q_\xi }(t){\Phi _1}(t)X(t)\times \\ & \Phi _1^{\rm{T}}(t) + \Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t) + {L_i}(t + 1){C_i}(t + 1)\times \\ & L_i^{\rm{T}}(t + 1)-{L_i}(t + 1){G_i}(t + 1) - \\ & G_i^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) \end{align} $
(13) 初值${{\hat {\pmb x}}_i}(0) = {\pmb \mu _0}$和${P_i}(0) = {P_0}$.
证明.由式(6), 多传感器系统(1) $\sim$ (4)的基于第$i$个传感器子系统的局部滤波误差方程为
$ \begin{align} {{\tilde {\pmb x}}_i}&(t + 1) = {\pmb x}(t + 1) - {{\hat {\pmb x}}_i}(t + 1)= \\ & \{ [{\Phi _0}(t) + \xi (t){\Phi _1}(t)] - {F_i}(t) - {L_i}(t + 1) \times \\ & {H_{0i}}(t + 1){\Phi _0}(t) - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1) \times \\ & \xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){\lambda _i}(t + 1){L_i}(t + 1) \times \\ & {H_{1i}}(t + 1)[{\Phi _0}(t) + \xi (t){\Phi _1}(t)] \} {\pmb x}(t)+ \\ & [{F_i}(t) + {L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t) - {\gamma _i}(t+ 1)\times \\ & {L _i}(t + 1){H _{0i}}(t + 1){\Phi _0}(t)]{{\tilde {\pmb x}}_i}(t) +{\Gamma }(t){\pmb w}(t) - \\ & {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1)\Gamma (t){\pmb w}(t) - \\ & {\gamma _i}(t + 1){\lambda _i}(t + 1){L_i}(t + 1){H_{1i}}(t + 1)\Gamma (t){\pmb w}(t) - \\ & {\gamma _i}(t + 1){L_i}(t + 1){{\pmb v}_i}(t + 1) - \\ & {L_i}(t + 1){D_i}(t + 1){{\pmb \theta} _i}(t + 1) \end{align} $
(14) 对任意的未知输入${\pmb \theta _i}(t)$, 为了使状态估计满足无偏性, 即满足, 由(14)可得:
$ \begin{equation} {\Phi _0}(t) - {F_i}(t) - {L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t)=0 \end{equation} $
(15) $ \begin{equation} {L_i}(t + 1){D_i}(t + 1) = 0 \end{equation} $
(16) 则由式(15)引出式(7)成立.因此, 式(14)可化简为
$ \begin{align} {{\tilde {\pmb x}}_i}&(t + 1) = \{ \xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){L_i}(t + 1)\times \\ & {H_{0i}}(t + 1)\xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){\lambda _i}(t + 1)\times \\ & {L_i}(t + 1){H_{1i}}(t + 1)[{\Phi _0}(t) + \xi (t){\Phi _1}(t)]\} {\pmb x}(t)+ \\ & [{\Phi _0}(t) -{\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t)] \times \\ & {{\tilde {\pmb x}}_i}(t) +[{I_n} -{\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1) - %\times \\ & {\gamma _i}(t + 1) {\lambda _i}(t + 1){L_i}(t + 1){H_{1i}}(t + 1)] \Gamma (t){\pmb w}(t) - \\ & {\gamma _i}(t + 1){L_i}(t + 1){{\pmb v}_i}(t + 1) \end{align} $
(17) $ \begin{align} {P_i}&(t + 1) = {\rm{E}}[{{\tilde {\pmb x}}_i}(t + 1){{\tilde {\pmb x}}_i}^{\rm{T}}(t + 1)] = \\ & {\rm{E}} \{ \{ \xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1)\xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){\lambda _i}(t + 1){L_i}(t + 1){H_{1i}}(t + 1) \times \\ & [{\Phi _0}(t) + \xi (t){\Phi _1}(t)] \} {\pmb x}(t){{\pmb x}^{\rm{T}}}(t)\{ \xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1)\xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){\lambda _i}(t + 1) \times \\ & {L_i}(t + 1){H_{1i}}(t + 1)[{\Phi _0}(t) + \xi (t){\Phi _1}(t)]{\} ^{\rm{T}}}\} + {\rm{E}} \{ [{\Phi _0}(t) - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t)]{{\tilde {\pmb x}}_i}(t){{{\tilde {\pmb x}}_i}^{\rm{T}}}(t) \times \\ & {[{\Phi _0}(t) - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t)]}^{\rm{T}} \} + {\rm{E}} \{ [{I_n} - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1) - {\gamma _i}(t + 1) \times \\ & {\lambda _i}(t + 1){L_i}(t + 1){H_{1i}}(t + 1)]\Gamma (t){\pmb w}(t){{\pmb w}^{\rm{T}}}(t){\Gamma ^{\rm{T}}}(t) [{I_n} - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1) - {\gamma _i}(t + 1) \times \\ & {\lambda _i}(t + 1){L_i}(t + 1){H_{1i}}(t + 1)]^{\rm{T}} \} + {\rm{E}}\{ \gamma _i^{\rm{2}}(t + 1){L_i}(t + 1){{\pmb v}_i}(t + 1){{\pmb v}_i}^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) \} \end{align} $
(18) $ \begin{align} {P_i}&(t + 1) = {Q_\xi }(t){\Phi _1}(t)X(t){\Phi _1 ^{\rm{T}}}(t) + {\alpha _i}{Q_\xi }(t){L_i}(t + 1){H_{0i}}(t + 1){\Phi _1}(t)X(t)\Phi _1 ^{\rm{T}}(t)H_{0i} ^{\rm{T}}(t + 1)L_i ^{\rm{T}}(t + 1) - \\ & {\alpha _i}{Q_\xi }(t){L_i}(t + 1){H_{0i}}(t + 1){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t) - {\alpha _i}{Q_\xi }(t){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t)H_{0i}^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) + {\alpha _i} \times \\ & {Q _{\lambda _i}}(t + 1){L_i}(t + 1){H_{1i}}(t + 1){\Phi _0}(t)X(t)\Phi _0 ^{\rm{T}}(t)H_{1i} ^{\rm{T}}(t + 1) L_i^{\rm{T}}(t + 1) + {\alpha _i}{Q_{{\lambda _i}}}(t + 1){Q_\xi }(t) {L_i}(t + 1) \times \\ & {H_{1i}}(t + 1){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t)H_{1i}^{\rm{T}}(t + 1)L_i ^{\rm{T}}(t + 1) + {\Phi _0}(t){P_i}(t)\Phi _0 ^{\rm{T}}(t) - {\alpha _i}\Phi _0(t){P_i}(t) \Phi _0 ^{\rm{T}}(t)H_{0i}^{\rm{T}}(t + 1) \times \\ & L_i^{\rm{T}}(t + 1) - {\alpha _i}{L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t){P_i}(t)\Phi _0 ^{\rm{T}}(t) + {\alpha _i}{L_i}(t + 1){H_{0i}}(t + 1){\Phi _0}(t){P_i}(t)\Phi _0^{\rm{T}}(t) \times \\ & H_{0i}^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) + \Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t)- {\alpha _i}\Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t)H_{0i}^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) - {\alpha _i}{L_i}(t + 1) \times \\ & {H_{0i}}(t + 1)\Gamma (t){Q_{\pmb w}}(t) {\Gamma ^{\rm{T}}}(t) + {\alpha _i}{L_i}(t + 1){H_{0i}}(t + 1)\Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t)H_{0i}^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) +{\alpha_i}{L_i}(t + 1) \times \\ & {Q_{{{\pmb v}_i}}}(t + 1)L_i^{\rm{T}}(t + 1)+ {\alpha _i}{Q_{{\lambda _i}}}(t + 1){L_i}(t + 1){H_{1i}}(t + 1)\Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t)H_{1i}^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) \end{align} $
(19) 根据滤波误差方程(17), 有滤波误差方差阵为(见式(18) (见下页)), 经计算可得(见式(19) (见下页)).合并整理化简得:
$ \begin{align} {P_i}&(t + 1)= {\Phi _0}(t){P_i}(t)\Phi _0^{\rm{T}}(t) + {Q_\xi }(t){\Phi _1}(t)X(t)\times \\ & \Phi _1^{\rm{T}}(t) + \Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t) + {L_i}(t + 1){C_i}(t + 1)\times \\ & L_i^{\rm{T}}(t + 1)-{L_i}(t + 1){G_i}(t + 1) - \\ & G_i^{\rm{T}}(t + 1)L_i^{\rm{T}}(t + 1) \end{align} $
(20) 即式(13)成立, 其中${C_i}(t+1)$和${G_i}(t+1)$分别由式(11)和式(9)定义.
应用线性无偏最小方差估计准则[19], 并由约束条件式(16)可引出如下辅助方程:
$ \begin{align} {J_i}(t + 1) = &{\rm {tr}} \{ {P_i}(t + 1) \} + 2{\rm {tr}}\{ {\Lambda _i ^{\rm{T}}}(t + 1) \times \\ &{L_i}(t + 1){D_i}(t + 1) \} \end{align} $
(21) 为了极小化性能指标${J_i}(t + 1)$, 令, 应用矩阵迹的求导公式[21]有:
$ \begin{equation} {L_i}(t + 1){C_i}(t + 1) + {\Lambda _i}(t + 1)D_i^{\rm{T}}(t + 1) = G_i^{\rm{T}}(t + 1) \end{equation} $
(22) 将式(22)和约束条件(16)联立得矩阵方程组
$ \begin{equation} \left \{ \begin{array}{lll} {L_i}(t + 1){C_i}(t + 1) + \\\qquad\qquad {\Lambda _i}(t + 1)D_i^{\rm{T}}(t + 1) = G_i^{\rm{T}}(t + 1) \\ {L_i}(t + 1){D_i}(t + 1) = 0 \\ \end{array} \right. \end{equation} $
(23) 写为分块矩阵形式
$ \begin{align} &\left[ {\begin{array}{*{20}{c}} {{C_i}(t + 1)}&{{D_i}(t + 1)} \\ {D_i^{\rm{T}}(t + 1)}&0 \\ \end{array}} \right]\left[ {\begin{array}{*{20}{c}} {L_i^{\rm{T}}(t + 1)} \\ {\Lambda _i^{\rm{T}}(t + 1)} \\ \end{array}} \right] = \\ &\qquad\left[ {\begin{array}{*{20}{c}} {{G_i}(t + 1)} \\ 0 \\ \end{array}} \right] \end{align} $
(24) 由假设2可知方程式(24)的系数矩阵的逆存在[21], 解分块矩阵方程(24)得:
$ \begin{align} {\Lambda _i}&(t + 1) = G_i^{\rm{T}}(t + 1)C_i^{ - 1}(t + 1){D_i}(t + 1)\times \\ & {[D_i^{\rm{T}}(t + 1)C_i^{ - 1}(t + 1){D_i}(t + 1)]^{ - 1}} \end{align} $
(25) $ \begin{equation} {L_i}(t + 1) = [G_i^{\rm{T}}(t + 1) - {\Lambda _i}(t + 1)D_i^{\rm{T}}(t + 1)]C_i^{ - 1}(t + 1) \end{equation} $
(26) 即(10)与(8)成立.
注1. 由定理1可知, 由于通信干扰是未知的, 为了避免干扰对滤波器的影响, 我们设计了不依赖于未知干扰的滤波器式(6), 使其满足无偏性和滤波误差方差的迹最小.为了保证此类滤波器的存在性, 即矩阵方程式(24)的解存在, 要求假设2成立.
2.2 局部估计误差互协方差阵的计算
定理2. 在假设1和2下, 多传感器系统(1) $\sim$ (4)的第$i$个和第$j$个传感器子系统间的滤波误差互协方差阵可计算如式(27) $(i, j = 1, 2, \cdots , L)$, 初值${P_{ij}}(0) ={P_0}$.
$ \begin{align} {P_{ij}}&(t + 1) = {Q_\xi }(t){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t) + {\alpha _i}{\alpha _j}{L_i}(t + 1){H_{0i}}(t + 1){Q_\xi }(t){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t)H_{0j}^{\rm{T}}(t + 1)L_j^{\rm{T}}(t + 1)- \\ & {\alpha _i}{Q_\xi }(t){L_i}(t + 1){H_{0i}}(t + 1){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t) - {\alpha _j}{Q_\xi }(t){\Phi _1}(t)X(t)\Phi _1^{\rm{T}}(t)H_{0j}^{\rm{T}}(t + 1) L_j^{\rm{T}}(t + 1)+ {\Phi _0}(t) \times \\ & {P_{ij}}(t)\Phi _0^{\rm{T}}(t) - {\alpha _i}{L_i}(t + 1){H_{0i}}(t + 1) {\Phi _0}(t){P_{ij}}(t)\Phi _0^{\rm{T}}(t) - {\alpha _j}{\Phi _0}(t){P_{ij}}(t) \Phi _0^{\rm{T}}(t)H_{0j}^{\rm{T}}(t + 1)L_j^{\rm{T}}(t + 1)+ \\ & {\alpha _i}{\alpha _j}{L_i}(t + 1){H_{0i}}(t + 1) {\Phi _0}(t){P_{ij}}(t)\Phi _0^{\rm{T}}(t)H_{0j}^{\rm{T}}(t + 1)L_j^{\rm{T}}(t + 1) + \Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t) + {\alpha _i}{\alpha _j}{L_i}(t + 1) \times \\ & {H_{0i}}(t + 1)\Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t)H_{0j}^{\rm{T}}(t + 1) L_j^{\rm{T}}(t + 1) - {\alpha _i}{L_i}(t + 1){H_{0i}}(t + 1)\Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t) - \\ & {\alpha _j}\Gamma (t){Q_{\pmb w}}(t){\Gamma ^{\rm{T}}}(t) H_{0j}^{\rm{T}}(t + 1) L_j^{\rm{T}}(t + 1) \end{align} $
(27) $ \begin{align} {P_{ij}}&(t + 1) = {\rm{E}} \{ \{ \xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){L_i}(t + 1){H_{0i}}(t + 1)\xi (t){\Phi _1}(t) - {\gamma _i}(t + 1){\lambda _i}(t + 1){L_i}(t + 1) \times \\ & {H_{1i}}(t + 1)[{\Phi _0}(t) + \xi (t){\Phi _1}(t)] \} {\pmb x}(t) { {\pmb x}^{\rm{T}}}(t) \{ \xi (t){\Phi _1}(t) - {\gamma _j}(t + 1){L_j}(t + 1){H_{0j}}(t + 1)\xi (t){\Phi _1}(t) - \\ & {\gamma _j}(t + 1){\lambda _j}(t + 1){L_j}(t + 1){H_{1j}}(t + 1)[{\Phi _0}(t) + \xi (t){\Phi _1}(t)]{\} ^{\rm{T}}}\}+ {\rm{E}}\{ [{\Phi _0}(t) - {L_i}(t + 1){\gamma _i}(t + 1)\times \\ & {H_{0i}}(t + 1){\Phi _0}(t)]{{\tilde {\pmb x}}_i}(t){{\tilde {\pmb x}}_j}^{\rm{T}}(t)[{\Phi _0}(t) - {L_j}(t + 1){\gamma _j}(t + 1){H_{0j}}(t + 1){\Phi _0}(t){]^{\rm{T}}}\} + {\rm {E}} \{ [{I_n} - {\gamma _i}(t + 1) \times \\ & {L_i}(t + 1){H_{0i}}(t + 1) - {\gamma _i}(t + 1){\lambda _i}(t + 1) {L_i}(t + 1){H_{1i}}(t + 1)]\Gamma (t){\pmb w}(t){{\pmb w} ^{\rm{T}}}(t){\Gamma ^{\rm{T}}}(t) \times \\ & {[{I_n} - {\gamma _j}(t + 1){L_j}(t + 1){H_{0j}}(t + 1) - {\gamma _j}(t + 1){\lambda _j}(t + 1){L_j}(t + 1){H_{1j}}(t + 1)]^{\rm{T}} }\} \end{align} $
(28) 证明. 将式(17)代入中, 由与${\pmb v_j}(t)$, ${{\tilde {\pmb x}}_i}(t)$与, ${\pmb v_i}(t)$与${\pmb v_j}(t)$, $i \ne j$, 均不相关, 可得(见式(28))又由${\lambda _i}(t+1)$与不相关, 展开计算式(28)可得式(27).
2.3 按矩阵加权融合滤波器
基于定理1的局部滤波器和定理2的任两个局部滤波误差互协方差阵, 应用在线性最小方差意义下的按矩阵加权融合估计算法[20]有如下分布式融合状态滤波器:
$ \begin{equation} {{\hat {\pmb x}}_o}(t) = \sum\limits_{i = 1}^L {{{\bar A}_i}} (t){{\hat {\pmb x}}_i}(t) \end{equation} $
(29) 最优加权矩阵${\bar A_i}(t)$, $i = 1, 2, \cdots , L$, 计算如下:
$ \begin{equation} [{\bar A_1}(t), {\bar A_2}(t), \cdots, {\bar A_L}(t)] = {[{e^{\rm{T}}}{\Sigma ^{ - 1}}(t)e]^{ - 1}}{e^{\rm{T}}}{\Sigma ^{ - 1}}(t) \end{equation} $
(30) 其中, $e = {[{I_n}, {I_n}, \cdots , {I_n}]^{\rm{T}}}$为的矩阵, 矩阵$\Sigma (t)$为第$(i, j)$块元素为${P_{ij}}(t)$的矩阵.分布式融合估计误差方差阵计算为
$ \begin{equation} {P_o}(t) = {[{e^{\rm{T}}}{\Sigma ^{ - 1}}(t)e]^{ - 1}} \end{equation} $
(31) 并且有关系
$ \begin{equation} {P_o}(t) \le {P_i}(t), \quad i = 1, 2, \cdots , L \end{equation} $
(32) 注2. 在图 1中, 我们假设从各局部滤波器到融合中心的通信是完美的, 即无数据丢失.如果有数据丢失, 只要存在局部滤波器到达融合中心, 就可应用上面的融合算法获得融合估计.若某时刻局部估计都丢失了, 则可用上一时刻的融合估计进行预报.
3. 仿真实例
考虑如下跟踪系统:
$ \begin{align} {\pmb x}&(t + 1) = \left( \left[ {\begin{array}{*{20}{c}} {0.95}&{{T}} \\ 0&{0.95} \\ \end{array}} \right] + \xi (t) \times \right. \\ & \left. \left[ {\begin{array}{*{20}{c}} {0.05}&{\rm{0}} \\ 0&{0.05} \\ \end{array}} \right] \right) {\pmb x}(t) + \left[ {\begin{array}{*{20}{c}} \frac{T^2}{2} \\ {{T}} \\ \end{array}} \right]{\pmb w}(t) \end{align} $
(33) $ \begin{equation} {\pmb y_i}(t) = ({H_{0i}} + {\lambda _i}(t){H_{1i}}){\pmb x}(t) + {\pmb v_i}(t), \quad i = 1, 2, 3 \end{equation} $
(34) $ \begin{equation} {\pmb z_i}(t) = {\gamma _i}(t){\pmb y_i}(t) + {D_i}{\pmb \theta _i}(t), \quad i = 1, 2, 3 \end{equation} $
(35) 取采样周期${{T}} = 1$, 观测阵, , , , , , 其中不相关白噪声${\pmb w}(t)$和${\pmb v_i}(t)$的方差分别为${Q _{\pmb w}} =2$, ${Q_{{\pmb v}_1}} = {I_2}$, , ${Q_{{\pmb v}_3}} = 0.8{I_2}$, ${I_2}$为2维单位阵; 互不相关且与其他噪声均不相关的白噪声和${\lambda _i}(t)$的方差分别为${Q_\xi } = 0.8$, , ${Q_{{\lambda _2}}} = 0.7$, .取干扰系数阵, ~, ~, 通道干扰${\pmb \theta _1}(t) = 1$, ${\pmb \theta _2}(t) = t/2$, ${\pmb \theta _3}(t) = \sin t$, 初值, ${P_0} = 0.1{I_2}$.取200个采样数据, 且不同传感器的接收率分别为${\alpha _1} = 0.5$, ${\alpha _2} = 0.8$, ${\alpha _3} = 0.4$.求多传感器分布式按矩阵加权融合递推状态滤波器.
图 2是分布式按矩阵加权融合状态滤波器的跟踪图, 由图 2可以看出本文所设计的分布式融合状态滤波器具有良好的跟踪特性.图 3是各单传感器局部滤波器与分布式融合滤波器的估计误差方差比较图, 表明了分布式融合滤波器的估计误差方差小于各局部滤波器的估计误差方差.这验证了融合估计比单传感器估计精度高, 达到了融合的目的.
图 3 局部与分布式融合状态滤波器估计误差方差比较图Fig. 3 Comparison of estimation error variances of local and distributed fusion state filters图 4和图 5分别给出了带补偿与无补偿的第3个局部单传感器子系统滤波器和分布式融合滤波器经过100次Monte-Carlo试验的MSE (Mean square error)比较, 从图中可以看出, 带有补偿的滤波精度比无补偿的滤波精度高.这验证了采用丢包补偿方法可以改善滤波器估计精度.
图 4 带补偿与无补偿的第3传感器子系统滤波器的MSE比较Fig. 4 MSE comparison of the 3rd sensor subsystem filters with compensation and no compensation
图 5 带补偿与无补偿的分布式融合滤波器的MSE比较Fig. 5 MSE comparison of distributed fusion filters with compensation and no compensation图 6给出了第3个局部单传感器子系统采用本文预报补偿的滤波算法和文献[10]采用以前收到的最新数据补偿的算法进行MSE比较图, 因为采用预报补偿用到了以前收到的所有观测数据, 且文献[10]未考虑未知干扰, 所以本文的滤波精度高于文献[10]的滤波精度.
4. 结论
针对带有未知通信干扰、丢失观测和乘性噪声不确定性的多传感器网络化系统, 考虑从不同传感器到局部滤波器的数据传输中具有不同丢失率情形, 当观测丢失时采用当前丢失观测的一步预报作为补偿.在线性无偏最小方差意义下, 提出了不依靠未知通信干扰的最优局部子系统状态滤波器.推导了任意两传感器子系统间的估计误差互协方差阵, 应用矩阵加权融合估计算法给出了分布式融合状态滤波器.下一步将开展系统噪声与观测噪声相关情形下状态滤波器的设计, 以及未知通信干扰的估计问题.
-
图 3 局部与分布式融合状态滤波器估计误差方差比较图
Fig. 3 Comparison of estimation error variances of local and distributed fusion state filters
图 4 带补偿与无补偿的第3传感器子系统滤波器的MSE比较
Fig. 4 MSE comparison of the 3rd sensor subsystem filters with compensation and no compensation
图 5 带补偿与无补偿的分布式融合滤波器的MSE比较
Fig. 5 MSE comparison of distributed fusion filters with compensation and no compensation
-
[1] 邢江, 关治洪.网络化控制系统的研究现状与展望.控制工程, 2006, 13(4):294-297 http://www.cnki.com.cn/Article/CJFDTOTAL-JZDF200604001.htmXing Jiang, Guan Zhi-Hong. Research progress and prospects of the networked control systems. Control Engineering of China, 2006, 13(4):294-297 http://www.cnki.com.cn/Article/CJFDTOTAL-JZDF200604001.htm [2] Zhang H S, Xie L H. Control and Estimation of Systems with Input/Output Delays. Berlin, Germany:Springer-Verlag, 2007. [3] Hespanha J P, Naghshtabrizi P, Xu Y G. A survey of recent results in networked control systems. Proceedings of the IEEE, 2007, 95(1):138-162 [4] 李洪波, 孙增圻, 孙富春.网络控制系统的发展现状及展望.控制理论与应用, 2010, 27(2):238-243 https://www.wenkuxiazai.com/doc/7727d54f767f5acfa1c7cd36.htmlLi Hong-Bo, Sun Zeng-Qi, Sun Fu-Chun. Networked control systems:an overview of state-of-the-art and the prospect in future research. Control Theory & Applications, 2010, 27(2):238-243 https://www.wenkuxiazai.com/doc/7727d54f767f5acfa1c7cd36.html [5] Lungu M, Lungu R. Full-order observer design for linear systems with unknown inputs. International Journal of Control, 2012, 85(10):1602-1615 doi: 10.1080/00207179.2012.695397 [6] Hsieh C S. On the global optimality of unbiased minimum-variance state estimation for systems with unknown inputs. Automatica, 2010, 46(4):708-715 doi: 10.1016/j.automatica.2010.01.029 [7] Yong S Z, Zhu M H, Frazzoli E. A unified filter for simultaneous input and state estimation of linear discrete-time stochastic systems. Automatica, 2016, 63:321-329 doi: 10.1016/j.automatica.2015.10.040 [8] 孙书利.具有一步随机滞后和多丢包的网络系统的最优线性估计.自动化学报, 2012, 38(3):349-356 http://www.aas.net.cn/CN/Y2012/V38/I3/349Sun Shu-Li. Optimal linear estimation for networked systems with one-step random delays and multiple packet dropouts. Acta Automatica Sinica, 2012, 38(3):349-356 http://www.aas.net.cn/CN/Y2012/V38/I3/349 [9] 李娜, 马静, 孙书利.带多丢包和滞后随机不确定系统的最优线性估计.自动化学报, 2015, 41(3):611-619 http://www.aas.net.cn/CN/Y2015/V41/I3/611Li Na, Ma Jing, Sun Shu-Li. Optimal linear estimation for stochastic uncertain systems with multiple packet dropouts and delays. Acta Automatica Sinica, 2015, 41(3):611-619 http://www.aas.net.cn/CN/Y2015/V41/I3/611 [10] Ma J, Sun S L. Optimal linear estimation for systems with multiplicative noise uncertainties and multiple packet dropouts. IET Signal Processing, 2012, 6(9):839-848 doi: 10.1049/iet-spr.2012.0065 [11] 李岳炀, 钟麦英.具有多测量数据包丢失的线性离散时变系统故障检测滤波器设计.自动化学报, 2015, 41(9):1638-1648 http://www.aas.net.cn/CN/Y2015/V41/I9/1638Li Yue-Yang, Zhong Mai-Ying. Fault detection filter design for linear discrete time-varying systems with multiple packet dropouts. Acta Automatica Sinica, 2015, 41(9):1638-1648 http://www.aas.net.cn/CN/Y2015/V41/I9/1638 [12] Xia Y Q, Shang J Z, Chen J, Liu J P. Networked data fusion with packet losses and variable delays. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 2009, 39(5):1107-1120 doi: 10.1109/TSMCB.2009.2012437 [13] 赵国荣, 韩旭, 万兵, 闫鑫.具有传感器增益退化、随机时延和丢包的分布式融合估计器.自动化学报, 2016, 42(7):1053-1064 http://www.aas.net.cn/CN/Y2016/V42/I7/1053Zhao Guo-Rong, Han Xu, Wan Bing, Yan Xin. A decentralized fusion estimator with stochastic sensor gain degradation, delays and data dropouts. Acta Automatica Sinica, 2016, 42(7):1053-1064 http://www.aas.net.cn/CN/Y2016/V42/I7/1053 [14] Ma J, Sun S L. Distributed fusion filter for networked stochastic uncertain systems with transmission delays and packet dropouts. Signal Processing, 2017, 130:268-278 doi: 10.1016/j.sigpro.2016.07.004 [15] Li N, Sun S L, Ma J. Multi-sensor distributed fusion filtering for networked systems with different delay and loss rates. Digital Signal Processing, 2014, 34:29-38 doi: 10.1016/j.dsp.2014.07.016 [16] Pang C Y, Sun S L. Fusion predictors for multisensor stochastic uncertain systems with missing measurements and unknown measurement disturbances. IEEE Sensors Journal, 2015, 15(8):4346-4354 doi: 10.1109/JSEN.2015.2416511 [17] Silva E I, Solis M A. An alternative look at the constant-gain Kalman filter for state estimation over erasure channels. IEEE Transactions on Automatic Control, 2013, 58(12):3259-3265 doi: 10.1109/TAC.2013.2263647 [18] 祁波, 孙书利.带未知通信干扰和丢包补偿的多传感器系统的融合估计.系统科学与数学, 2016, 36(8):1094-1106 http://www.cnki.com.cn/Article/CJFDTOTAL-SHJT201506025.htmQi Bo, Sun Shu-Li. Fusion estimation for multi-sensor systems with unknown communication disturbances and compensation of packet losses. Journal of Systems Science and Mathematical Sciences, 2016, 36(8):1094-1106 http://www.cnki.com.cn/Article/CJFDTOTAL-SHJT201506025.htm [19] Anderson B D O, Moore J B. Optimal Filtering. Englewood Cliffs, NJ, USA: Prentice-Hall, 1979. [20] Sun S L, Deng Z L. Multi-sensor optimal information fusion Kalman filter. Automatica, 2004, 40(6):1017-1023 doi: 10.1016/j.automatica.2004.01.014 [21] 徐宁寿.随机信号估计与系统控制.北京:北京工业大学出版社, 2001. 7-13Xu Ning-Shou. Stochastic Signal Estimation and System Control. Beijing:Beijing University of Technology Press, 2001. 7-13 期刊类型引用(23)
1. Yongpeng CUI，Xiaojun SUN. Multi-Sensor Fusion Adaptive Estimation for Nonlinear Under-observed System with Multiplicative Noise. Chinese Journal of Electronics. 2024(01): 282-292 . 
必应学术2. 王江伟. 微电网V2G双向变换器运行状态分析方法. 黑龙江电力. 2024(01): 30-35+44 . 百度学术3. 崔永鹏，孙小君，张扬. 带乘性噪声的欠观测系统无迹增量Kalman融合估计. 黑龙江大学工程学报(中英俄文). 2024(02): 66-74 . 百度学术4. 韩旭，王元鑫，程显超，王小飞. 基于有限存储空间的分布式传感器融合估计器. 北京航空航天大学学报. 2023(02): 335-343 . 百度学术5. 徐迎菊，王娜，花玉. 含未知输入不确定系统的扩展递归滤波算法. 控制工程. 2023(06): 999-1005 . 百度学术6. 马静，杨晓梅，孙书利. 带时间相关乘性噪声多传感器系统的分布式融合估计. 自动化学报. 2023(08): 1745-1757 . 本站查看7. 卢建华，王元鑫，姜林君，高峰娟. 基于线性编码的随机时延和丢包下的状态估计. 计算机仿真. 2022(01): 49-55 . 百度学术8. 李笑宇，冯肖雪，潘峰，蒲宁. 网络攻击下无人机信息物理系统的自适应状态估计. 航空学报. 2022(03): 437-450 . 百度学术9. 刘淑艳，鲁小利. 基于激光传感器网络物理层随机噪声加密方法的研究. 激光杂志. 2022(06): 136-140 . 百度学术10. 姜帅，孙书利. 带相关噪声异步采样系统的分布式最优线性递推融合估计. 控制理论与应用. 2022(07): 1272-1280 . 百度学术11. 朱志英. 基于通信耦合的大面积灌溉系统设计研究. 农机化研究. 2021(06): 194-198+204 . 百度学术12. 石元博，王建辉，方晓柯，黄越洋，顾树生. 基于HJB方程的无线传感器网络系统Minimax控制器设计. 控制与决策. 2021(04): 947-952 . 百度学术13. 王鹏，胡宏彬，李勇. 大数据融合模型的智能化网络安全检测方法. 计算机测量与控制. 2021(05): 40-44 . 百度学术14. 张健. 层次化通信网络备份数据库缓存子系统设计. 现代电子技术. 2021(11): 33-36 . 百度学术15. 杨春山，经本钦，刘政，王建琦. 具有噪声方差及多种网络诱导不确定系统鲁棒Kalman估计. 控制理论与应用. 2021(10): 1607-1618 . 百度学术16. 刘晴，刘旭，汤玮，金海，袁汉云. 电网通信业务数据的融合技术分析. 信息技术. 2020(03): 153-158 . 百度学术17. 赵国荣，韩旭，王康. 具有传感器增益退化、传输时延和丢包的离线状态估计器. 自动化学报. 2020(03): 540-548 . 本站查看18. 余永龙. 具有网络攻击的多传感器系统融合滤波研究. 无线互联科技. 2020(18): 32-33 . 百度学术19. 冯肖雪，刘萌，李笑宇，潘峰. 面向高超声速飞行器双重不确定性的自适应状态估计. 宇航学报. 2020(12): 1561-1570 . 百度学术20. 余永龙，陆海霞，费绍金，王福章. 带相关噪声和不完全丢包系统的最优线性滤波. 唐山师范学院学报. 2019(03): 27-34 . 百度学术21. 吴琼. 带未知传感器干扰多速率系统的状态估计. 现代制造技术与装备. 2019(07): 140-141 . 百度学术22. 赵国荣，刘伯彦，高超. 数据丢包的参数不确定下无人机滚动时域估计. 系统工程与电子技术. 2019(12): 2849-2854 . 百度学术23. 余永龙，费绍金. 带相关噪声的多传感器系统的融合滤波器. 科技与创新. 2018(23): 26-27 . 百度学术其他类型引用(31)
-
下载:
下载:
计量
- 文章访问数: 2040
- HTML全文浏览量: 182
- PDF下载量: 628
- 被引次数: 54
