近年来, 多智能体系统已成为众多研究领域的前沿课题, 在车辆编队控制[1], 传感器网络[2], 机器人技术[3], 群集运动[4]等方面存在广泛的应用. 其中, 一致性问题[5]最为热门. 然而, 有关滤波的问题却较少提及. 传感器网络可以看作为一类特殊的多智能体系统, 而滤波问题是传感器网络最基本的课题之一. 由于传感器节点通常分布的很广, 因此, 在现实中多采用分布式滤波或估计. 在过去的几十年中, 分布式滤波问题主要采用的是Kalman滤波理论[6]. 但是, Kalman滤波需要知道外部扰动噪声, 对数学模型的精确要求非常高. 与传统的Kalman滤波相比,
众所周知, 通信网络的带宽和资源是有限的, 相比于有线网络, 无线网络对带宽的要求更是严格. 而周期触发方式把所有的采样信号都传输到滤波器, 造成不必要的网络资源的浪费. 因此, 学者们为了解决这一问题, 提出了基于事件触发机制的滤波器设计方法[9–11]. 文献[9]建立了一个结合事件触发机制和网络诱导时延的模型, 利用滤波误差系统的方法解决
受上述讨论的推动, 本文研究了具有切换拓扑的离散时间多智能体系统的事件触发
用
考虑含有
$\left\{ {\begin{array}{*{20}{c}} \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! {{x_i}\left( {k + 1} \right) = A{x_i}\left( k \right) + B{w_i}\left( k \right)} \\ \!\!{{y_i}\left( k \right) = C\displaystyle\sum\nolimits_{j \in {N_i}} {{a_{ij}}\left( {{x_i}\left( k \right) - {x_j}\left( k \right)} \right) + D{w_i}\left( k \right)} } \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!{{z_i}\left( k \right) = E{x_i}\left( k \right)} \end{array} } \right.$ | (1) |
其中,
用
$P\left\{ {r\left( {k + 1} \right) = s\left| {r\left( k \right){\rm{ = }}r} \right.} \right\} = {\pi _{rs}}$ |
其中,
转移概率矩阵可以定义如下:
$\Pi {\rm{ = }}\left[ {\begin{array}{*{20}{c}} {{\pi _{11}}}& \cdots &{{\pi _{1q}}} \\ \vdots & \ddots & \vdots \\ {{\pi _{q1}}}& \cdots &{{\pi _{qq}}} \end{array}} \right]$ |
考虑到随机切换拓扑, 系统(1)可以重新写成:
$\left\{ {\begin{array}{*{20}{c}} \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! {{x_i}\left( {k + 1} \right) = A{x_i}\left( k \right) + B{w_i}\left( k \right)} \\ \!\!{{y_i}\left( k \right) = C\displaystyle\sum\nolimits_{j \in {N_i}} {a_{_{ij}}^{r(k)}\left( {{x_i}\left( k \right) - {x_j}\left( k \right)} \right) + D{w_i}\left( k \right)} } \\ \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\! \!\!\!\!\!\!\!\!{{z_i}\left( k \right) = E{x_i}\left( k \right)} \end{array} } \right.$ | (2) |
为了便于书写, 在本文接下来的内容里, 我们令
$\left\{ {\begin{array}{*{20}{c}} \!\! {\mathop {{x_i}}\limits^ \wedge \left( {k + 1} \right) = {F_r}\mathop {{x_i}}\limits^ \wedge \left( k \right) + {G_r}\mathop {{y_i}}\limits^ \wedge \left( k \right)} \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!{\mathop {{z_i}}\limits^ \wedge \left( k \right) = E\mathop {{x_i}}\limits^ \wedge \left( k \right)} \end{array} } \right.$ | (3) |
其中,
为了节约网络资源, 我们采用了分布式事件触发传输策略和切换拓扑的结合, 通过它来判断是否需要把信号传输给滤波器. 事件触发机制设计如下:
$\chi _i^{\rm T}\left( {k_l^i + j} \right){\phi _r}{\chi _i}\left( {k_l^i + j} \right) \leqslant {\sigma _r}y_i^{\rm T}\left( {k_l^i + j} \right){\phi _r}{y_i}\left( {k_l^i + j} \right)$ | (4) |
其中,
$\begin{gathered} k_{l + 1}^i = k_l^i + \mathop {\min }\limits_{{p_i}} \{ {p_i}|\chi _i^{\rm T}\left( {k_l^i + j} \right){\phi _r}{\chi _i}\left( {k_l^i + j} \right) \\ \geqslant {\sigma _r}y_i^T\left( {k_l^i + j} \right){\phi _r}{y_i}\left( {k_l^i + j} \right)\} \end{gathered} $ | (5) |
注释1. 从式(4)可以看出, 事件触发阈值由参数
注释2. 由于每个智能体都是周期采样的, 采样时间序列为
由于零阶保持器的作用是用来存储最新传送过来的数据, 滤波器输入可以表示为:
${\hat y_i}\left( k \right) = {y_i}\left( {k_l^i} \right)$ | (6) |
将式(6)带入式(3), 我们可以得到:
$\left\{ {\begin{array}{*{20}{c}} \!\!\! {\mathop {{x_i}}\limits^ \wedge \left( {k + 1} \right) = {F_r}\mathop {{x_i}}\limits^ \wedge \left( k \right) + {G_r}{y_i}\left( {k_l^i} \right)} \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! {\mathop {{z_i}}\limits^ \wedge \left( k \right) = E\mathop {{x_i}}\limits^ \wedge \left( k \right)} \end{array} } \right.$ | (7) |
其中,
运用文献[12]中的延时方法, 将系统(2)建模成等价的时间滞后系统, 我们可以得到误差向量
因此, 我们可以将事件触发条件(4)重新写成:
$e_i^{\rm T}\left( k \right){\phi _r}{e_i}\left( k \right) \leqslant {\sigma _r}y_i^{\rm T}\left( {k - \tau \left( k \right)} \right){\phi _r}{y_i}\left( {k - \tau \left( k \right)} \right)$ | (8) |
滤波器(7)可以重新写成:
$\left\{ {\begin{array}{*{20}{c}} \!\!\! {\mathop {{x_i}}\limits^ \wedge \left( {k + 1} \right) = {F_r}\mathop {{x_i}}\limits^ \wedge \left( k \right) + {G_r}{y_i}\left( {k - \tau \left( k \right)} \right) - {G_r}{e_i}\left( k \right)} \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!{\mathop {{z_i}}\limits^ \wedge \left( k \right) = E\mathop {{x_i}}\limits^ \wedge \left( k \right)} \end{array} } \right.$ | (9) |
定义以下向量:
$\xi _i^{\rm T}\left( k \right) = \left[ {x_i^{\rm T}\left( k \right),{{\hat x}_i}^{\rm T}\left( k \right)} \right]$ |
${\bar z_i}\left( k \right) = {z_i}\left( k \right) - {\hat z_i}\left( k \right)$ |
${\hat w_i}\left( k \right) = {\left[ {w_i^{\rm T}\left( k \right),w_i^{\rm T}\left( {k - \tau \left( k \right)} \right)} \right]^{\rm T}} $ |
结合式(2)和式(9), 我们可以得到以下的闭环系统:
$\left\{ {\begin{array}{*{20}{c}} \! \! \! {{\xi _i}\left( {k + 1} \right) = \overline A {\xi _i}\left( k \right) + \overline B {H_2}\displaystyle\sum\nolimits_{j \in {N_i}} {a_{ij}^r\left[ {{\xi _i}\left( {k - \tau \left( k \right)} \right)} \right.} } \\ \! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \! \! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! {\left. { - {\xi _j}\left( {k - \tau \left( k \right)} \right)} \right] + \overline C {e_i}\left( k \right) + \overline D \mathop {{w_i}}\limits^ \wedge \left( k \right)} \\ \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \! \! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \!\! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! \! {{{\overline z }_i}\left( k \right) = E{H_1}{\xi _i}\left( k \right)} \end{array} } \right.$ | (10) |
其中,
为了简述上述系统, 定义一些新的增广变量:
$\xi \left( k \right) = {\left[ {{\xi _1}^{\rm T}\left( k \right),\xi _2^{\rm T}\left( k \right), \cdots ,\xi _N^{\rm T}\left( k \right)} \right]^{\rm T}} $ |
$e\left( k \right) = {\left[ {{e_1}^{\rm T}\left( k \right),e_2^{\rm T}\left( k \right), \cdots ,e_N^{\rm T}\left( k \right)} \right]^{\rm T}} $ |
$\hat w\left( k \right) = {\left[ {\hat w_1^{\rm T}\left( k \right),\hat w_2^{\rm T}\left( k \right), \cdots ,\hat w_N^{\rm T}\left( k \right)} \right]^{\rm T}}$ |
$\overline z \left( k \right) = {\left[ {\overline z _1^{\rm T}\left( k \right),\overline z _2^{\rm T}\left( k \right), \cdots ,\overline z _N^{\rm T}\left( k \right)} \right]^{\rm T}} $ |
因此, 可以将滤波误差系统(10)改写为:
$\left\{ {\begin{array}{*{20}{c}} \!\! \!\!{\xi \left( {k + 1} \right) = \left( {{I_N} \otimes \overline A } \right)\xi \left( k \right) + \left( {{L_r} \otimes \overline B {H_2}} \right)\xi \left( {k - \tau \left( k \right)} \right)} \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{ + \left( {{I_N} \otimes \overline C } \right)e\left( k \right) + \left( {{I_N} \otimes \overline D } \right)\hat w\left( k \right)} \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!{\overline z \left( k \right) = \left( {{I_N} \otimes E{H_1}} \right)\xi \left( k \right)} \end{array} } \right.$ | (11) |
定理1. 假设系统的系数矩阵
$\left[ {\begin{array}{*{20}{c}} {{\Xi _{11}}}&{\hat H_2^{\rm T}R}&0&0&0&{{\Xi _{16}}}&{{\Xi _{17}}}\\ * &{{\Xi _{22}}}&R&0&{{\Xi _{25}}}&{{\Xi _{26}}}&{{\Xi _{27}}}\\ * & * &{ - Q - R}&0&0&0&0\\ * & * & * &{ - {{\hat \phi }_r}}&0&{{\Xi _{46}}}&{{\Xi _{47}}}\\ * & * & * & * &{{\Xi _{55}}}&{{\Xi _{56}}}&{{\Xi _{57}}}\\ * & * & * & * & * &{ - {P_r}}&0\\ * & * & * & * & * & * &{ - R} \end{array}} \right] < 0$ | (12) |
$\sum\limits_{s = 1}^q {{\pi _{rs}}{P_s} \leqslant {P_r}} $ | (13) |
其中,
$\begin{array}{c} {\Xi _{11}}{\rm{ = }}{\left( {{I_N} \otimes E{H_1}} \right)^{\rm T}}\left( {{I_N} \otimes E{H_1}} \right) - {P_r} + \hat H_2^{\rm T}Q{{\hat H}_2} - \hat H_2^{\rm T}R{{\hat H}_2} \end{array}$ |
${\Xi _{22}}{\rm{ = }}{\sigma _r}{\left( {{L_r} \otimes C} \right)^{\rm T}}{\hat \phi _r}\left( {{L_r} \otimes C} \right) - 2R$ |
${\Xi _{25}}{\rm{ = }}{\sigma _r}{\left( {{L_r} \otimes C} \right)^{\rm T}}{\hat \phi _r}\left( {{I_N} \otimes D{H_3}} \right)$ |
$\begin{gathered} {\Xi _{55}}{\rm{ = }}{\sigma _r}{\left( {{I_N} \otimes D{H_3}} \right)^{\rm T}{\rm T}}{\hat \phi _r}\left( {{I_N} \otimes D{H_3}} \right) - {\gamma ^2}\left( {{I_N} \otimes H_2^{\rm T}{H_2}} \right) \end{gathered} $ |
${\Xi _{16}}{\rm{ = }}{\left( {{I_N} \otimes \overline A } \right)^{\rm T}}{P_r}, {\Xi _{26}}{\rm{ = }}{\left( {{L_r} \otimes \overline B } \right)^{\rm T}}{P_r}$ |
${\Xi _{46}}{\rm{ = }}{\left( {{I_N} \otimes \overline C } \right)^{\rm T}}{P_r}, {\Xi _{56}}{\rm{ = }}{\left( {{I_N} \otimes \overline D } \right)^{\rm T}}{P_r}$ |
${\Xi _{17}}{\rm{ = }}{\tau _M}{\left( {{I_N} \otimes \left( {\bar A - \tilde I} \right)} \right)^{\rm T}}\hat H_2^{\rm T}R$ |
${\Xi _{27}}{\rm{ = }}{\tau _M}{\left( {{L_r} \otimes \bar B} \right)^{\rm T}}\hat H_2^{\rm T}R,{\Xi _{47}}{\rm{ = }}{\tau _M}{\left( {{I_N} \otimes \bar C} \right)^{\rm T}}\hat H_2^{\rm T}R$ |
${\Xi _{57}}{\rm{ = }}{\tau _M}{\left( {{I_N} \otimes \bar D} \right)^{\rm T}}\hat H_2^{\rm T}R,{\hat \phi _r} = {I_N} \otimes {\phi _r},{H_3} = \left[ {\begin{array}{*{20}{c}} 0&I \end{array}} \right]$ |
证明: 我们构造如下的Lyapunov函数:
$V\left( k \right) = {V_1}\left( k \right) + {V_2}\left( k \right) + {V_3}\left( k \right)$ |
其中,
${V_1}\left( k \right) = {\xi ^{\rm T}}\left( k \right){P_r}\xi \left( k \right) $ |
${V_2}\left( k \right) = \displaystyle\sum\nolimits_{s = k - {\tau _M}}^{k - 1} {{\xi ^{\rm T}}\left( s \right)\hat H_2^{\rm T}Q{{\hat H}_2}\xi \left( s \right)} $ |
${V_3}\left( k \right) = {\tau _M}\displaystyle\sum\nolimits_{s = - {\tau _M} + 1}^0 {\displaystyle\sum\nolimits_{l = k + s - 1}^{k - 1} {{\delta ^{\rm {\rm T}}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} } $ |
$\delta \left( l \right) = \xi \left( {k + 1} \right) - \xi \left( k \right), {\hat H_2} = {I_N} \otimes {H_2}$ |
那么,
$\begin{gathered} {\rm E}\left\{ {\Delta {V_1}\left( k \right)} \right\} = {\rm E}\left\{ {{V_1}\left( {k + 1} \right) - {V_1}\left( k \right)} \right\} \\ {\rm{ = }}{\rm E}\left\{ {{\xi ^{\rm T}}\left( {k + 1} \right)} \right.\sum\nolimits_{j = 1}^N {{\pi _{rs}}{P_j}\xi \left( {k + 1} \right)} \left. { - {\xi ^{\rm T}}\left( k \right){P_r}\xi \left( k \right)} \right\} \end{gathered} $ |
$\begin{array}{l} {\rm{E}}\left\{ {\Delta {V_2}\left( k \right)} \right\} = {\rm{E}}\left\{ {{V_2}\left( {k + 1} \right) - {V_2}\left( k \right)} \right\}\\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{\rm{ = E}}\{ {\xi ^{\rm T}}\left( k \right)\hat H_2^{\rm T}Q{{\hat H}_2}\xi \left( k \right)\\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; - {\xi ^{\rm T}}\left( {k - {\tau _M}} \right)\hat H_2^{\rm T}Q{{\hat H}_2}\xi \left( {k - {\tau _M}} \right)\} \end{array}$ |
$\begin{array}{l} {\rm{E}}\left\{ {\Delta {V_3}\left( k \right)} \right\}{\rm{ = E}}\left\{ {{V_3}\left( {k + 1} \right) - {V_3}\left( k \right)} \right\}\\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{\rm{ = E}}\{ \tau _M^2{\delta ^{\rm T}}\left( k \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( k \right)\\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; - {\tau _M}\displaystyle\sum\nolimits_{l = k - {\tau _M}}^{k - 1} {{\delta ^{\rm T}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} \} \end{array}$ |
其中,
$\begin{gathered} {\tau _M}\displaystyle\sum\nolimits_{l = k - {\tau _M}}^{k - 1} {{\delta ^{\rm T}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} \\ {\rm{ = }}{\tau _M}\displaystyle\sum\nolimits_{l = k - {\tau _M}}^{k - 1 - \tau \left( k \right)} {{\delta ^{\rm T}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} \\ \;\;\;{\rm{ + }}{\tau _M}\displaystyle\sum\nolimits_{l = k - \tau \left( k \right)}^{k - 1} {{\delta ^{\rm T}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} \\ \end{gathered} $ |
根据Jensen不等式, 我们可以得到:
$\begin{gathered} {\tau _M}\displaystyle\sum\nolimits_{l = k - {\tau _M}}^{k - 1 - \tau \left( k \right)} {{\delta ^{\rm T}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} \leqslant \\ - {\left[ {\displaystyle\sum\nolimits_{l = k - {\tau _M}}^{k - 1 - \tau \left( k \right)} {\delta \left( l \right)} } \right]^{\rm T}}\hat H_2^{\rm T}R{\hat H_2}\left[ {\displaystyle\sum\nolimits_{l = k - {\tau _M}}^{k - 1 - \tau \left( k \right)} {\delta \left( l \right)} } \right] \\ \end{gathered} $ |
$\begin{gathered} {\tau _M}\displaystyle\sum\nolimits_{l = k - \tau \left( k \right)}^{k - 1} {{\delta ^{\rm T}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} \leqslant \\ - {\left[ {\displaystyle\sum\nolimits_{l = k - \tau \left( k \right)}^{k - 1} {\delta \left( l \right)} } \right]^{\rm T}}\hat H_2^{\rm T}R{\hat H_2}\left[ {\displaystyle\sum\nolimits_{l = k - \tau \left( k \right)}^{k - 1} {\delta \left( l \right)} } \right] \\ \end{gathered} $ |
结合事件触发策略(8), 我们可以得到:
$\begin{gathered} {\rm E}\left\{ {\Delta V\left( k \right)} \right\} \leqslant {\rm E}\left\{ {{\xi ^{\rm T}}\left( {k + 1} \right){P_r}\xi \left( {k + 1} \right) - } \right.{\xi ^{\rm T}}\left( k \right){P_r}\xi \left( k \right) \\ \quad \quad \quad \quad \quad \quad \quad + {\xi ^{\rm T}}\left( k \right)\hat H_2^{\rm T}Q{\hat H_2}\xi \left( k \right) + \tau _M^2{\delta ^{\rm T}}\left( k \right)\hat H_2^{\rm T}R{\hat H_2}\delta \left( k \right) \\ \quad \quad - {\xi ^{\rm T}}\left( {k - {\tau _M}} \right)\hat H_2^{\rm T}Q{\hat H_2}\xi \left( {k - {\tau _M}} \right) \\ \quad \quad - {\tau _M}\displaystyle\sum\nolimits_{l = k - {\tau _M}}^{k - 1} {{\delta ^{\rm T}}\left( l \right)\hat H_2^{\rm T}R{{\hat H}_2}\delta \left( l \right)} \\ \quad \quad+ {\sigma _r}{y^{\rm T}}\left( {k - \tau \left( k \right)} \right){\hat \phi _r}y\left( {k - \tau \left( k \right)} \right) \\ - {e^{\rm T}}\left( k \right){\hat \phi _r}e\left( k \right) + {\overline z ^{\rm T}}\left( k \right)\overline z \left( k \right) \\ \quad \quad - {\gamma ^2}{\hat w^{\rm T}}\left( k \right)\hat w\left( k \right)\left. {} \right\}{\rm{ = }}{\eta ^{\rm T}}\left( k \right)\Omega \eta \left( k \right) \\ \end{gathered} $ |
其中,
$\begin{gathered} {\eta ^{\rm T}}\left( k \right) = \left[ {\begin{array}{*{20}{c}} {{\xi ^{\rm T}}\left( k \right)}&{{\xi ^{\rm T}}\left( {k - \tau \left( k \right)} \right)\hat H_2^{\rm T}}&{{\xi ^{\rm T}}\left( {k - {\tau _M}} \right)\hat H_2^{\rm T}} \end{array}} \right. \\ \left. {\begin{array}{*{20}{c}} {{e^{\rm T}}\left( k \right)}&{\mathop {{w^{\rm T}}}\limits^ \wedge \left( k \right)} \end{array}} \right] \\ \end{gathered} $ |
$\Omega {\rm{ = }}\Upsilon + {\Gamma ^{\rm T}}{P_r}\Gamma + \tau _M^2\Gamma _1^{\rm T}R{\Gamma _1}$ |
$\Gamma {\rm{ = }}\left[ {\begin{array}{*{20}{c}} {{I_N} \otimes \overline A }&{L \otimes \overline B }&0&{{I_N} \otimes \overline C }&{{I_N} \otimes \overline D } \end{array}} \right]$ |
${\Gamma _1} = \left[ {\begin{array}{*{20}{c}} {{I_N} \otimes \left( {\overline A - I} \right)}&{L \otimes \overline B }&0&{{I_N} \otimes \overline C } \end{array}} \right.\left. {\;\;{I_N} \otimes \overline D } \right]$ |
$\Upsilon {\rm{ = }}\left[ {\begin{array}{*{20}{c}} {{\Xi _{11}}}&{\hat H_2^{\rm T}R}&0&0&0 \\ * &{{\Xi _{22}}}&R&0&{{\Xi _{25}}} \\ * & * &{ - Q - R}&0&0 \\ * & * & * &{ - {{\hat \phi }_r}}&0 \\ * & * & * & * &{{\Xi _{55}}} \end{array}} \right]$ |
根据Schur补定理, 式(12)确保
由于以上的分析是假设滤波器参数已知的情况下, 但在实际的情况中滤波器的参数是未知的, 因此接下来我们在定理1的基础上设计
定理2. 假设系统的系数矩阵
$\left[ {\begin{array}{*{20}{c}} {{{\hat \Xi }_{11}}}&{\hat H_2^{\rm T}R}&0&0&0&{{{\hat \Xi }_{16}}}&{{\Xi _{17}}} \\ * &{{\Xi _{22}}}&R&0&{{\Xi _{25}}}&{{{\hat \Xi }_{26}}}&{{\Xi _{27}}} \\ * & * &{ - Q - R}&0&0&0&0 \\ * & * & * &{ - {{\hat \phi }_r}}&0&{{{\hat \Xi }_{46}}}&{{\Xi _{47}}} \\ * & * & * & * &{{\Xi _{55}}}&{{{\hat \Xi }_{56}}}&{{\Xi _{57}}} \\ * & * & * & * & * &{{{\hat \Xi }_{66}}}&0 \\ * & * & * & * & * & * &{ - R} \end{array}} \right] < 0$ | (14) |
$\displaystyle\sum\limits_{s = 1}^q {{\pi _{rs}}({U_{1s}} - {W_s}) \leqslant {U_{1r}} - {W_r}} $ | (15) |
其中,
${\hat \Xi _{11}}{\rm{ = }}{I_N} \otimes \left[ {\begin{array}{*{20}{c}} {{E^{\rm T}}E - {U_{1r}} + Q - R}&{ - {E^{\rm T}}E - {W_r}} \\ { - {E^{\rm T}}E - {W_r}}&{{E^{\rm T}}E - {W_r}} \end{array}} \right]$ |
$ \begin{array}{l} {\hat \Xi _{16}} = {I_N} \otimes \left[ {\begin{array}{*{20}{c}} {{A^{\rm T}}{U_{1r}}}&{{A^{\rm T}}{W_{1r}}} \\ {\hat F_{_r}^{\rm T}}&{\hat F_{_r}^{\rm T}} \end{array}} \right] \\ {\hat \Xi _{26}}{\rm{ = }}{L^{\rm T}} \otimes \left[ {\begin{array}{*{20}{c}} {{C^{\rm T}}\hat G_r^{\rm T}}&{{C^{\rm T}}\hat G_r^{\rm T}} \end{array}} \right] \end{array} $ |
$ \begin{array}{l} {\hat \Xi _{46}}{\rm{ = }}{I_N} \otimes \left[ {\begin{array}{*{20}{c}} {\hat G_r^{\rm T}}&{\hat G_r^{\rm T}} \end{array}} \right] \\ {\hat \Xi _{56}}{\rm{ = }}{I_N} \otimes \left[ {\begin{array}{*{20}{c}} {{B^{\rm T}}{U_{1r}}}&{{B^{\rm T}}{W_r}} \\ {{D^{\rm T}}\hat G_r^{\rm T}}&{{D^{\rm T}}\hat G_r^{\rm T}} \end{array}} \right] \end{array} $ |
${\hat \Xi _{66}}{\rm{ = }}{I_N} \otimes \left[ {\begin{array}{*{20}{c}} { - {U_{1r}}}&{ - {W_r}} \\ * &{ - {W_r}} \end{array}} \right]$ |
那么滤波误差系统是渐近稳定的, 并且
${F_r} = W_r^{ - 1}{\hat F_r},{G_r} = W_r^{ - 1}{\hat G_r}$ | (16) |
证明: 令
${U_r} = \left[ {\begin{array}{*{20}{c}} {{U_{1r}}}&{{U_{2r}}} \\ * &{{U_{3r}}} \end{array}} \right]$ |
其中,
${J_2} = diag\left\{ {\begin{array}{*{20}{c}} {{I_N} \otimes {J_1}}&{\begin{array}{*{20}{c}} {\begin{array}{*{20}{c}} I&I&I&I \end{array}}&{{I_N} \otimes {J_1}}&I \end{array}} \end{array}} \right\}$ |
根据
${\hat F_r} = {U_{2r}}{F_r}U_{3r}^{ - 1}U_{2r}^{\rm T}, {\hat G_r} = {U_{2r}}{G_r}$ |
通过简单的线性转换, 我们可以很容易得到式(14). 不难发现, 式(12)等价于式(14), 因此, 滤波误差系统(11)渐进稳定, 并且具有
在这一节中, 给出两个例子来说明所提出的方法的有效性. 第一种情况考虑文献[11]中网络拓扑结构是固定不变的. 考虑一个由4个智能体组成的多智能体系统, 其参数变量为:
$A = \left[ {\begin{array}{*{20}{c}} {0.1}&{0.4} \\ 0&{ - 0.5} \end{array}} \right], B = \left[ {\begin{array}{*{20}{c}} {0.2}, \\ {0.5} \end{array}} \right], C = \left[ {\begin{array}{*{20}{c}} {0.1}&0 \end{array}} \right]$ |
$D = 0.2, E = \left[ {\begin{array}{*{20}{c}} {0.1}&{0.1} \end{array}} \right]$ |
其中, 外部扰动
${L_1} = \left[ {\begin{array}{*{20}{c}} 1&{ - 1}&0&0 \\ 0&1&0&{ - 1} \\ 0&{ - 1}&1&0 \\ { - 1}&0&0&1 \end{array}} \right]$ |
令初始条件
${F_1} = \left[ {\begin{array}{*{20}{c}} {{\rm{ - 0}}{\rm{.0456}}}&{{\rm{0}}{\rm{.2285}}} \\ {{\rm{ - 0}}{\rm{.0539}}}&{{\rm{ - 0}}{\rm{.4570}}} \end{array}} \right], {G_1} = \left[ {\begin{array}{*{20}{c}} {{\rm{ - 0}}{\rm{.660}}} \\ {{\rm{ - 0}}{\rm{.2361}}} \end{array}} \right]$ |
图2描绘的是每个智能体的滤波误差信号
第二种情况考虑网络拓扑结构是切换的. 我们选取系统参数变量跟第一种情况一样. 图4代表两种有向的网络通信拓扑结构图. 不失一般性, 假设所有的权重都为1, 其对应的Laplacian矩阵如下:
${L_1} = \left[ {\begin{array}{*{20}{c}} 1&{ - 1}&0&0 \\ 0&1&0&{ - 1} \\ 0&{ - 1}&1&0 \\ { - 1}&0&0&1 \end{array}} \right], {L_2} = \left[ {\begin{array}{*{20}{c}} 1&0&0&{ - 1} \\ { - 1}&1&0&0 \\ 0&{ - 1}&1&0 \\ 0&{ - 1}&0&1 \end{array}} \right]$ |
图5表示马尔可夫链中两种模式的切换. 其中, 状态转移概率矩阵为:
$\Pi {\rm{ = }}\left[ {\begin{array}{*{20}{c}} {0.3}&{0.7} \\ {0.4}&{0.6} \end{array}} \right]$ |
假设最大时延上界
${F_1} = \left[ {\begin{array}{*{20}{c}} {{\rm{ - 0}}{\rm{.6257}}}&{{\rm{0}}{\rm{.2758}}} \\ {{\rm{ - 0}}{\rm{.1215}}}&{{\rm{ - 0}}{\rm{.0650}}} \end{array}} \right], {G_1} = \left[ {\begin{array}{*{20}{c}} {{\rm{ - 0}}{\rm{.1253}}} \\ {{\rm{ - 0}}{\rm{.0328}}} \end{array}} \right]$ |
${F_2} = \left[ {\begin{array}{*{20}{c}} {{\rm{0}}{\rm{.7681}}}&{{\rm{0}}{\rm{.5518}}} \\ {{\rm{0}}{\rm{.1453}}}&{{\rm{ - 0}}{\rm{.0218}}} \end{array}} \right], {G_2} = \left[ {\begin{array}{*{20}{c}} {{\rm{0}}{\rm{.1217}}} \\ {{\rm{0}}{\rm{.0328}}} \end{array}} \right]$ |
图6描绘的是每个智能体的滤波误差信号
注释3. 虽然事件触发机制能有效减少网络信号的传送, 节省网络带宽资源的占用率, 但是相对于周期触发方式, 事件触发方式需要复杂的数学计算与分析, 由于目前大部分结果仅关注事件触发网络控制系统的稳定性分析, 无法设计事件触发网络化系统的滤波器. 因此, 为弥补已有结果的不足, 我们引入相应的事件触发机制, 提出对应的事件触发滤波器联合设计的方法.
4 结论本文研究了具有切换拓扑的离散时间多智能体系统的事件触发
本文主要是对4个智能体组成的多智能体系统进行了研究与实验仿真, 由于智能体的数量相对较少, 测试场景也相对简单, 因此, 在接下来的工作中主要会对复杂情形的多智能体系统进行建模与分析.
[1] |
Oh KK, Ahn HS. Formation control of mobile agents based on inter-agent distance dynamics. Automatica, 2011, 47(10): 2306-2312. DOI:10.1016/j.automatica.2011.08.019 |
[2] |
Ge XH, Han QL. Distributed event-triggered H∞ filtering over sensor networks with communication delays
. Information Sciences, 2015, 291: 128-142. DOI:10.1016/j.ins.2014.08.047 |
[3] |
Chen SY. Kalman filter for robot vision: A survey. IEEE Transactions on Industrial Electronics, 2012, 59(11): 4409-4420. DOI:10.1109/TIE.2011.2162714 |
[4] |
Gao JY, Xu X, Ding N, et al. Flocking motion of multi-agent system by dynamic pinning control. IET Control Theory & Applications, 2017, 11(5): 714-722. |
[5] |
缪盛, 崔宝同. 带有通信时滞的二阶多智能体系统一致问题. 计算机系统应用, 2012, 21(8): 99-104. |
[6] |
Olfati-Saber R. Distributed Kalman filtering for sensor networks//2007 46th IEEE Conference on Decision and Control. New Orleans, LA, USA. 2008. 5492–5498.
|
[7] |
Shen B, Wang ZD, Hung YS. Distributed H∞-consensus filtering in sensor networks with multiple missing measurements: The finite-horizon case
. Automatica, 2010, 46(10): 1682-1688. DOI:10.1016/j.automatica.2010.06.025 |
[8] |
Shen B, Wang Z, Hung YS, et al. Distributed H∞ filtering for polynomial nonlinear stochastic systems in sensor networks
. IEEE Transactions on Industrial Electronics, 2011, 58(5): 1971-1979. DOI:10.1109/TIE.2010.2053339 |
[9] |
Hu SL, Yue D. Event-based H∞ filtering for networked system with communication delay
. Signal Processing, 2012, 92(9): 2029-2039. DOI:10.1016/j.sigpro.2012.01.012 |
[10] |
Liu JL, Tang J, Fei SM. Event-triggered H∞ filter design for delayed neural network with quantization
. Neural Networks, 2016, 82: 39-48. DOI:10.1016/j.neunet.2016.06.006 |
[11] |
Ding L, Guo G. Distributed event-triggered H∞ consensus filtering in sensor networks
. Signal Processing, 2015, 108: 365-375. DOI:10.1016/j.sigpro.2014.09.035 |
[12] |
Hu S, Zhang Y, Du Z. Network-based H∞ tracking control with event-triggering sampling scheme
. IET Control Theory & Applications, 2012, 6(4): 533-544. |